Tettares

News, fathomed

Insights

The day's industry stories, already run through our knowledge graph: who supplies whom, who's exposed, what actually changes. Noise in, insights out.

Friday, September 4

ReportedPhysical AI · Rokae

ROKAE Launches Turnkey Embodied Robotics Solution with Industry ...

CovariantQualcommRokae

ROKAE unveiled a turnkey embodied robotics package with Qualcomm at Auto China 2026, bundling hardware, software, and compute into one platform. How many other robotics makers and AI platforms already depend on Qualcomm's Snapdragon Rob compute stack, and what does that mean for ROKAE's competitive position?

ReportedAI Compute · Misorobotics

Miso Launches Next-Generation Flippy Fry Station - Miso Robotics

NVIDIA CorporationMiso Robotics

Miso Robotics rolled out its next-generation Flippy Fry Station on January 28, 2025, co-designed with NVIDIA and Ecolab. The launch rides on White Castle's earlier 100-location Flippy 2 rollout commitment. Miso's entire robotics stack depends on a single named supplier relationship in our graph.

High impactAI Compute · Korea Herald

Samsung narrows HBM gap with SK hynix

Micron TechnologyNVIDIA CorporationSK hynix Inc.SK Hynix

Samsung's HBM revenue share reportedly jumped from 21% to 33% in a single quarter, fueled by HBM4 shipments to major tech buyers. But Samsung, SK hynix, and Micron all draw from the same equipment vendors and chase the same customer base. Which upstream chokepoint could cap Samsung's ability to sustain that gain?

High impactAI Compute · Financial Times

Transcript: Memory chip mania isn't going away, with Dan Kim

Samsung Electronics Co., Ltd.Taiwan Semiconductor Manufacturing Company LimitedMicron Technology

Dan Kim says the memory chip mania is far from over, with DRAM shortages tightening across the industry. Samsung, TSMC, and Micron all sit on overlapping suppliers and customers in this squeeze. Micron's position as HBM supplier to NVIDIA, Tesla, Apple, and Samsung marks a specific bottleneck worth tracing.

High impactAI Compute · DIGITIMES

China's EUV gap may stretch 10–15 years, but DUV gains test Western export controls

BroadcomKioxiaNVIDIA CorporationMediaTekSK hynix Inc.SK Hynix

China's EUV shortfall could persist a decade or more, even as its DUV lithography narrows the gap enough to worry Western regulators. Everyone's watching Beijing's access to tools. But what happens when Taiwan Manufacturing and Applied Materials, the shared chokepoints behind NVIDIA, SK hynix, Broadcom, and MediaTek, hit capacity limits first?

High impactPhysical AI · Nikkei Asia

Japanese robots go beyond factory floor to take on tree cutting, lab work

HumanoidUnitree Robotics

Japanese robots are branching into tree cutting and lab work, pushing humanoid tech past the factory floor. That expansion puts Chinese leader Unitree squarely in the crosshairs after its volatile Shanghai debut. Track how deep Unitree's NVIDIA chip dependency runs before betting on its next production cycle.

NotablePhysical AI · PR Newswire

D-Robotics at IFA 2026: The Computing Platform Powering the Next Generation of Home Robots

TM Roboticsek roboticsD-Robotics

D-Robotics is showcasing its Sunrise chip platform at IFA 2026, with hey AiMe, SuperDog, and the xLean TR1 as the first consumer robots built on it. Eight robot OEMs, including UBTech's Walker S and ABB's YuMi, run on this single Chinese compute supplier's platform.

NotableAI Compute · Linuxgizmos

Rockchip RK3576 module powers edge AI dev kit with 20-TOPS RK1820 accelerator

DeepSeekRockchip

Seeed Studio's reComputer dev kit pairs Rockchip's RK3576 SoC with an optional 20-TOPS RK1820 accelerator, targeting local LLM and vision inference at the edge without cloud reliance. Rockchip already supplies compute to at least seven robotics OEMs. Does the RK1820's 2.5GB on-chip DRAM give it a real edge over Nvidia and Qualcomm at the silicon level?

NotablePhysical AI · Astutegroup

Humanoid robots are creating a new demand curve for precision actuators

HumanoidSchaefflerNVIDIA CorporationAgiBot

Humanoid robot makers are racing to lock down actuators as production ramps toward real volume. Schaeffler now sits at the center of this scramble, feeding roller screws to multiple humanoid OEMs while sourcing rare-earth magnets from its own suppliers. The full supplier map behind that chokepoint is inside.

Thursday, September 3

ReportedAI Compute · CNBC

South Korea's semiconductor exports just tripled year-over-year. Is it too much of a good thing?

SK hynix Inc.Samsung Electronics Co., Ltd.SK Hynix

South Korea's chip exports tripled in August as SK Hynix and Samsung both race to feed the AI boom. But behind the surge, the two rivals may be colliding over the same equipment vendors and buyer base. What happens when SK Hynix's Applied Materials tool allocation runs into Samsung's own fab timeline?

ReportedAI Compute · SCMP

Chinese investors scramble for 1-in-4,000 odds in Enflame's red-hot IPO

NVIDIA CorporationS TechEnflame

Enflame's Star Market IPO drew wild demand, with retail investors facing roughly 1-in-4,000 odds to land shares in the 6.12 billion yuan raise. Watch what happens to Taiwan Manufacturing's wafer allocation once a newly capitalized Enflame starts competing with NVIDIA Corporation for the same foundry chokepoint.

ReportedAI Compute · Yahoo Finance

NVIDIA's MediaTek Bet Shows How It Plans to Defend Its AI Moat

GroqNVIDIA CorporationMediaTek

NVIDIA is putting $3.5B into MediaTek, pairing new Rubin CPX and Groq 3 LPX products with a fresh $96.22B quarterly haul. MediaTek already ships networking components to NVIDIA and is chasing Google's TPU v9 orders. What does locking in MediaTek's 400G/448G SerDes IP mean for NVIDIA's Taiwan concentration risk?

ReportedAI Compute · Bloomberg

Kepco Seeks Billions from Samsung, SK Hynix to Finance Grid

SK hynix Inc.Samsung Electronics Co., Ltd.SK Hynix

Kepco is negotiating billions in grid financing with Samsung Electronics and SK hynix as both race to accelerate fab timelines in Yongin. But both manufacturers also lean on the same upstream equipment suppliers, including Applied Materials, Tokyo Electron, and Lasertec. Could that shared bottleneck undercut what Kepco's funding is meant to secure?

High impactAI Compute · Marketscreener

Ambarella, Inc Enters into A Long-Term Strategic Distribution Agreement with Macnica Americas, Inc. and Macnica Atd Europe S.A.S

Ambarella Inc.NXP Semiconductors

Ambarella has named Macnica Americas and Macnica ATD Europe as long-term distributors for its CV-series and N-series SoCs across the Americas and EMEA. The move aims to widen a thin customer base beyond direct buyers like Ottonomy and Hanwha. Watch whether Macnica's ISV and integrator recruitment can actually convert into volume before NXP's reported bid changes the picture.

High impactPhysical AI · Pymnts

Nscale Inks $3.5 Billion Deal With Robotics Firm Figure

TM Roboticsek roboticsFigure AI Inc.D-Robotics

Nscale just became Figure AI's preferred compute provider in a $3.5 billion deal, plugging a gap in the humanoid robot maker's training infrastructure. Figure's supply chain still runs through single-source suppliers for batteries and vision. The brief maps how those dependencies interact with Figure's new compute pipeline.

NotableAI Compute · Asiatimes

US, Japan race to secure chip-grade minerals as China curbs exports

CoherentLumentumNVIDIA Corporation

China's indium curbs are turning chip supply chains into a scramble, with Washington and Tokyo racing to lock down mineral sources before shortages bite. Lumentum and Coherent both flag indium phosphide as their real growth ceiling. Watch how NVIDIA's $2 billion bets on each supplier tie into that constraint.

NotableAI Compute · Tradingview

Will Credo's Neocloud Expansion Broaden Its AEC Growth Runway?

Credo TechnologyMarvell TechnologyBroadcom

Credo's push into Neocloud customers could widen its AEC and switch silicon business beyond its current hyperscaler base. But its top four customers already account for 84% of quarterly revenue. Does the $770 million DustPhotonics acquisition actually move the needle on that concentration problem?

NotablePhysical AI · Electropages

Serbia Opens Chinese Humanoid Robot Production Facility

HumanoidAgiBot

Serbia's new Šabac plant, a joint venture with AgiBot and Minth Group, aims to churn out over 5,000 humanoid robots a year, positioning Chinese robotics for EU market entry. But final assembly isn't the whole supply chain. Does relocating to Serbia actually remove AgiBot's reliance on Chinese vision and magnet suppliers like RoboSense?

NotablePhysical AI · Wsws

ICE plans to spend up to $2 million on robot dogs for immigration raids

Boston DynamicsAlphabet

ICE is set to spend up to $2 million on Boston Dynamics' Spot robots for immigration enforcement, its first large-scale law enforcement buy of the hardware. If federal agencies scale this deployment, which single supplier's lidar sensors become the bottleneck for every unit built?

NotableAI Compute · Seeking Alpha

Micron: The AI Memory Shortage Has Spread Beyond HBM (NASDAQ:MU)

Micron TechnologyDell

Reports suggest the AI memory crunch is no longer just an HBM story, with DRAM and NAND tightening as server demand surges. Micron sits at the center of this web, even sourcing its own HBM from rival SK hynix. Fathom It to see what that upstream dependency means for Micron's supply ceiling.

NotableAI Compute · Indiatimes

Broadcom raises AI chip forecast as Big Tech keeps writing bigger checks

BroadcomMeta Platforms

Broadcom just hiked its AI chip revenue forecast again, betting on Big Tech's continued spending spree through 2028. But the customer list behind those numbers has a curious gap. Trace where Meta's chip supply actually flows through Broadcom's network, and whether Celestica's repeated presence signals a manufacturing bottleneck in the making.

High impactPhysical AI · PR Newswire

Nscale and Figure Sign Strategic Partnership to Power the Next ...

Figure AI Inc.

Nscale and Figure just locked in a massive compute partnership, with up to 100,000 NVIDIA GPUs headed to Texas starting in 2027. Figure already leans on a tight supplier circle from LG batteries to Qualcomm chips. Does this new $6 billion GPU commitment solve Figure's compute bottleneck or just relocate the risk to a single Barstow facility?

High impactPhysical AI · Kyodo

Schaeffler Unveils Molded Wave Gearbox for Humanoid Robots

Schaeffler

Schaeffler says its new molding process turns wave gearbox production from minutes into seconds, slashing unit costs by over 25 percent and material use by over 75 percent. Roller screw customers like NEURA Robotics and Rainbow Robotics sit downstream of this exact bottleneck. The brief traces what tighter gearbox supply means for Schaeffler's grip on the humanoid actuator chain.

High impactPhysical AI · Yahoo Finance

Elon Musk Noted Natural Gas Will Still Fuel Data Centers, Gene Munster Says 'Makes Sense' to Put 'New Money to Work in Power'

Tesla, Inc.

Elon Musk says natural gas will still bootstrap solar for years, and Gene Munster likes Tesla putting fresh capital into power. But Tesla's supply graph shows zero sourced suppliers for turbines or solar gear. Watch for which turbine caster, not just GE Vernova, ends up filling that casting-bottleneck gap Musk flagged.

NotableAI Compute · Indiatimes

Broadcom launches VMware Private AI Cloud for enterprise AI workloads and data security

Broadcom

Broadcom rolled out VMware Private AI Cloud, pushing enterprise AI infrastructure into software-defined territory beyond raw chip supply. This lands as Broadcom seeks up to $80 billion in debt to fund AI chip production for Anthropic and others. Why would a custom ASIC supplier build a platform that welcomes rival accelerators into the stack?

NotablePhysical AI · PR Newswire

FANUC America Brings Robotics, Automation, Physical AI and CNC ...

FANUC Corporation

FANUC is showcasing Physical AI robotics at IMTS, teaming with Google Cloud, NVIDIA and AWS to push manufacturing automation forward. Watch how deeply NVIDIA silicon is now woven into FANUC's controller hardware, like the R-30 i B Mini Plus, before calling this just another cloud partnership announcement.

Wednesday, September 2

High impactAI Compute · DIGITIMES

Marvell sees silicon photonics scaling from late 2027, with Taiwan and TSMC at the center

Marvell TechnologyHewlett Packard Enterprise

Marvell's CTO now points to late 2027 for co-packaged optics to scale, with Taiwan and TSMC anchoring the buildout. HPE, mentioned alongside it, says its own bottleneck has moved from demand to memory supply already secured for FY2027. Marvell's CPO roadmap also touches optics-adjacent suppliers Celestial AI, XConn, and Lightwave Logic.

High impactAI Compute · Gadgethacks

Samsung Memory Long-Term Supply Deals 2031: What's Locked In

Samsung Electronics Co., Ltd.Meta PlatformsNVIDIA CorporationMicron TechnologyMicrosoft CorporationAlphabet

Samsung has reportedly locked long-term memory supply deals with five data-center giants, with five more in final talks, aiming to cover up to 70% of capacity through 2031. Trace what the penalty clauses on those five-year contracts actually mean for NVIDIA and Microsoft once spot-market flexibility disappears.

High impactAI Compute · Yahoo Finance

AMD Stock Alert: Advanced Micro Devices Teams Up With Cisco on AI Infrastructure

Advanced Micro DevicesCisco Systems, Inc.

AMD and Cisco have gone live with AI infrastructure in Saudi Arabia, pairing Instinct MI355X GPUs with Cisco's Silicon One networking silicon. It marks AMD's push beyond hyperscaler clients into sovereign AI builds. But what happens to AMD's Instinct rollout if Cisco's networking roadmap or pricing shifts?

High impactAI Compute · Yahoo

Microsoft Signed a 20-Year AI Power Deal With Chevron. Is Power Becoming the New Chip Shortage?

Microsoft Corporation

Microsoft just locked in 20 years of Chevron gas power for its AI buildout, betting energy capacity, not chips, is now the real bottleneck. Watch what this means for Microsoft's 300,000-unit Maia 300 chip order and whether the Kilby power deal actually eases or intensifies pressure on NVIDIA and Marvell.

High impactAI Compute · Venturebeat

39.4% of enterprises eye non-Nvidia chips | VentureBeat

NVIDIA CorporationAlphabetIntel Corporation

A new survey finds 39.4% of enterprises now evaluating non-Nvidia accelerators, up from 25.3%, even as switch-timing urgency cools rather than spikes. Nvidia still commands trillions in market cap and locked OpenAI capacity through 2028. So what design wins are Alphabet's TPU roadmap and Intel's Nova Lake rollout actually banking on?

NotableAI Compute · Business Standard

Data centre spending to reach $31.6 trillion by 2050 on AI boom

Amazon.com, Inc.

PwC projects global data centre capex will hit $31.6 trillion by 2050, outpacing every past infrastructure boom, as recurring chip upgrades rather than land or construction drive the spend. Amazon's chip sourcing for its Proteus AI ecosystem runs through a single named supplier across every mapped edge.

NotableAI Compute · Reuters

Taiwan says its companies plan to invest another $20 billion in US

Taiwan Semiconductor Manufacturing Company Limited

Taiwan's government says local firms plan another $20 billion in US investment beyond TSMC's Arizona buildout, riding the AI chip wave. Watch how this second wave threads through TSMC's own supplier web, from Tokyo Electron equipment to SK hynix memory, in the full brief.

NotableAI Compute · Chinanationalnews

TSMC plans Kaohsiung site expansion to boost semiconductor testing efficiency

Taiwan Semiconductor Manufacturing Company Limited

TSMC is building a 3-hectare testing and validation hub at Baipu Industrial Park in Kaohsiung, aiming to sharply speed up packaging validation. The move lands squarely on the industry's tightest chokepoint for AI chip supply. Does this expansion move the needle on TSMC's CoWoS capacity targets through 2027 and beyond?

NotableSpace · Yahoo

Rocket Lab (RKLB) Marks 94th Electron Mission With Perfect Customer Launch Record

Rocket Lab

Rocket Lab just completed its 94th Electron launch with a spotless customer success record, adding to a backlog of 16 more Synspective missions. Rocket Lab's supplier base includes single-source dependencies like Mynaric for optical systems, and its customer concentration still centers heavily on Globalstar.

NotableAI Compute · Yahoo

Can AMD's Saudi AI Buildout Challenge NVIDIA and Broadcom?

Advanced Micro Devices

AMD and partners are lighting up 250 MW of Saudi AI capacity via HUMAIN, scaling toward 1 GW by 2030 with MI400 GPUs and EPYC chips. Watch what this sovereign deal does to AMD's single-source Taiwan foundry exposure before you assume diversification means safety.

NotableAI Compute · Yahoo

Nvidia's $1 Billion Nokia Bet Was Worth About $2.2 Billion by June. Is 6G the Hidden AI Trade?

NVIDIA Corporation

Nvidia's $1B Nokia stake has already doubled to roughly $2.2 billion, fueling speculation that AI-enabled radio networks could become chipmaking's next frontier. Hedge funds are piling into both names, with Marshall Wace tripling its Nokia position. But does the AI-RAN partnership actually open a new market for Nvidia's compute business, or just add paper gains?

High impactPhysical AI · Barchart

Nidec Announces the Receipt of the Investigation Committee's Report

Nidec Corporation

Nidec says it has received the investigation committee's report into unauthorized changes to materials, processes and designs made without customer sign-off. Findings stay confidential until September 4, 2026. The report lands atop an already-strained Nidec, whose vertically integrated Dalian subsidiary sits outside direct customer visibility.

Tuesday, September 1

ReportedAI Compute · Bloomberg

MediaTek Shares Soar 10% After Blockbuster Nvidia Investment

NVIDIA CorporationMediaTek

MediaTek stock jumped 10% after Nvidia poured $3.5 billion into its convertible bonds, deepening an already tight partnership. The two chipmakers are now bound tighter than a simple investment suggests. Watch how the NVLink Fusion tie-in and MediaTek's TSMC dependency shape who really controls its Q4 2026 data-center ASIC ramp.

ReportedPhysical AI · Nvidia

NVIDIA Announces Jetson Orin Nano 2 Robotics Computer to Redefine Entry-Level Edge AI

Doosan GroupNVIDIA CorporationCognex

NVIDIA just unveiled the Jetson Orin Nano 2, its next entry-level robotics computer aimed at cost-sensitive edge AI deployments. Cognex has already signaled it is among the first adopters exploring the chip. What does Cognex's $642 million cash position mean for how fast Universal Robots and Yaskawa see this vision tech reach their production lines?

ReportedPhysical AI · Techstartups

Top Tech News Today, September 1, 2026: Amazon, Anthropic, Honda, OpenAI, Sony, Warner, Z.AI & More - Tech Startups

Top GroupNVIDIA CorporationApple

Apple's leadership shakeup lands with Ternus stepping in as CEO ahead of a September 9 iPhone and foldable reveal, while NVIDIA and OpenAI notch fresh compute deals. Watch how Micron's HBM allocation and Taiwan Manufacturing's foundry capacity get split between NVIDIA's Blackwell line and Apple's A20 Pro chips.

High impactAI Compute · Taipeitimes

Silicon photonics expected to dominate AI growth, TSMC vice president says

Taiwan Semiconductor Manufacturing Company LimitedPowertech Technology

TSMC's VP says silicon photonics will grab over half the optical transceiver market next year, fueled by AI infrastructure demand and 60 percent growth in high-end transceivers. The company is racing toward co-packaged optics production later this year. Trace how TSMC's CPO push reshapes its role as chokepoint for NVIDIA, AMD, and Apple's AI hardware.

High impactAI Compute · Techtimes

CXMT Ships HBM3E AI Memory to Alibaba and Cambricon: US Export Controls Failed to Stop It

ChangXin Memory TechnologiesCambricon

CXMT has reportedly shipped HBM3E samples to Alibaba's T-Head and Cambricon, a year ahead of the 2027 timeline analysts expected. US export controls aimed to keep China off this frontier. So what does CXMT's actual yield rate on these HBM3E stacks say about how close China really is to volume production?

High impactAI Compute · Indiatimes

China's CXMT starts small-scale HBM3E production to supply domestic AI chipmakers

ChangXin Memory TechnologiesT-HeadCambricon

CXMT has begun small-scale production of HBM3E memory to feed China's domestic AI chip supply chain, easing reliance on Korean and US suppliers. Which two Chinese AI chip designers are already receiving trial shipments of this homegrown HBM3E, and can CXMT's yields actually scale to meet their 2027 roadmaps?

NotableAI Compute · Asianetnews

China's CXMT starts HBM3E production amid US semiconductor curbs

ChangXin Memory TechnologiesT-Head

CXMT has begun shipping trial HBM3E chips to Chinese chip designers, a milestone for Beijing's push toward memory independence under US export curbs. Track how CXMT's yield gap against Samsung and SK hynix stacks up against its new dependence on customers like T-Head for advanced AI memory qualification.

NotableAI Compute · Heraldcorp

Nvidia to invest $3.5b in Taiwan's MediaTek, deepening long-term partnership

Taiwan SemiconductorNVIDIA CorporationMediaTek

Nvidia is putting $3.5 billion into MediaTek convertible bonds, deepening a partnership already built on networking component supply. This isn't just goodwill capital. What does the investment actually secure for Nvidia on MediaTek's 2nm SerDes IP timeline and its Taiwan Manufacturing foundry dependency?

NotableAI Compute · Motley Fool

Sundar Pichai Says Alphabet Can't Build AI Capacity Fast Enough, and Anthropic Has Secured 5 Gigawatts of It.

Broadcom

Pichai admits Alphabet can't build AI capacity fast enough, yet Anthropic just locked down 5 gigawatts of next-gen TPU capacity starting in 2027. That capacity reportedly routes through a Broadcom filing naming Anthropic as a customer whose access hinges on one specific financial condition.

NotableAI Compute · Koreajoongangdaily

China's CXMT begins HBM3E production as chip battle with Korean powerhouses widens

ChangXin Memory TechnologiesT-HeadCambriconXiaomi

CXMT has begun HBM3E production, pushing China's memory champion into direct competition with Samsung, SK hynix, and Micron in the high-margin AI accelerator segment. Watch the yield numbers behind CXMT's eight-layer HBM3 output before betting on how fast this scales into 12-layer configurations.

NotablePhysical AI · Financialexpress

Beyond dancing and boxing: Inside China's race to turn humanoid robots into industrial workforce

Beijing Humanoid Robot Innovation CenterLongcheer TechnologyUBTech RoboticsGalbotUnitree RoboticsXiaomi

China's humanoid robots are graduating from stage tricks to factory floors and even retail stores, with Unitree's IPO and Tiangong Ultra's record sprint grabbing headlines. But beneath the show lies a tight web of shared component suppliers. How many of these humanoid makers actually depend on the same harmonic reducer maker for their joints?

NotablePhysical AI · Interestingengineering

Four-legged robot learns to jump through tight gaps without human help

Unitree RoboticsIntel Electronics Malaysia Sdn Bh

A Unitree Aliengo robot taught itself to leap through gaps nearly its own size, hitting 2.5 meters per second with no human guiding its moves. The trick came from onboard vision and reinforcement learning, not factory presets. What does this new autonomy demand mean for Unitree's chip supplier concentration just days after its volatile IPO?

ReportedAI Compute · Yahoo Finance

MU And SNDK's Nascent Rebound Faces A New Test -- China's CXMT Reportedly Makes Major Breakthrough With New AI Memory Chip

ChangXin Memory TechnologiesCambriconSanDiskMicron TechnologyT-HeadAlibaba Group

China's CXMT reportedly ramped HBM3E production, with Cambricon and Alibaba's T-Head named as early qualifiers, pressuring Micron and SanDisk shares after a strong rebound. Cambricon's chip qualification with CXMT marks the first credible domestic HBM alternative Chinese AI designers have ever had, and the brief traces exactly how that supply shift redraws pricing power across shared equipment vendors.

High impactPhysical AI · Financial Times

China’s real robot revolution is not about humanoids

Unitree RoboticsBYD

BYD's Xi'an plant runs thousands of automation arms with near-full autonomy on key EV assembly steps, part of a wider Chinese shift from flashy humanoids toward dense industrial robotics. China now exports more industrial robots than it imports. So which robot supplier is BYD actually already buying from for its own assembly line automation?

NotableAI Compute · Trendforce

[News] Samsung to Break Ground on KRW 6T Onyang HBM Fab in Sept., P5 Eyes Triple-Fab Shift as Expansion Accelerates

Samsung Electronics Co., Ltd.SK hynix Inc.

Samsung will break ground in September on a KRW 6 trillion HBM fab at Onyang, while P5 could shift to a triple-fab structure by 2030. Watch what this build-out means for HBM4 yield and Samsung's push into NVIDIA's Blackwell supply chain, currently anchored by SK hynix.

NotablePhysical AI · Reuters

EXCLUSIVE: China's Geely to ship first Lotus EVs to Canada in July under Carney-Xi deal, ambassador says

BYD

Geely's Lotus brand will be first to ship EVs into Canada in July under the new Carney-Xi trade deal, but BYD and Chery are already lining up behind it, coordinating with Ottawa for their own entry as soon as next year. Watch what BYD's Melexis chip contract reveals about its capacity to scale into that 49,000-unit tariff lane.

ContextPhysical AI · CNBC

Hugging Face's new duck robot is selling fast. A Chinese chip powers it

RockchipPollen Robotics

Hugging Face's Pollen Robotics duck, Microduck, blew past 10,000 units and $5 million in sales within days, pushing delivery dates beyond Christmas. Its brain is a Chinese-made Rockchip RK3566 chip. How does this cost-optimized chip stack up against Nvidia's Jetson Orin Nano 2 in the widening edge AI robotics race?

ContextAI Compute · Reuters

EXCLUSIVE Meta to put AI chip into production in September as it looks to double computing capacity, memo shows

Meta Platforms

Meta will start manufacturing its in-house 'Iris' AI chip in September, part of a four-generation push to double compute capacity by 2027. A named design and foundry partner outside Meta's usual GPU suppliers is now central to how fast this chip roadmap can move.

ContextPhysical AI · Indiatimes

Einride to deploy 500 Tesla electric trucks for North American fleet

Tesla, Inc.

Einride will roll out 500 Tesla Semi trucks across US freight corridors starting September 2026, tripling its fleet and naming Amazon as a customer. Watch how this fresh 500-truck demand signal lands on Tesla's battery suppliers, CATL and LG Energy Solution, already stretched thin.

ContextAI Compute · Asianetnews

MU And SNDK's Nascent Rebound Faces A New Test -- China's CXMT Reportedly Makes Major Breakthrough With New AI Memory Chip

ChangXin Memory TechnologiesMicron TechnologySanDiskT-HeadCambricon

Micron and SanDisk shares are wobbling again after reports that China's CXMT cracked HBM3E production, threatening to reshape AI memory supply lines. Both stocks had just started clawing back losses. Which Chinese AI chip designers are already qualifying CXMT's silicon for production, not just testing?

Monday, August 31

ReportedAI Compute · Amd

AMD, Cisco and HUMAIN Expand Saudi Arabia’s AI Infrastructure as AMD Instinct Systems Go Live - AMD Newsroom

Screen HoldingsAdvanced Micro Devices

AMD's Instinct MI355X GPUs are now live in Saudi Arabia, running on Cisco Silicon One networking through partner HUMAIN. It's a rare deployment outside AMD's usual hyperscaler base. Does this Middle East pivot expose AMD's USD 5.0 billion capacity buildout to new export-control and geopolitical risk?

ReportedSpace · Europeanspaceflight

SES Awards OHB €1 billion for IRIS2 MEO Satellites

SESOHB

SES has awarded OHB a €1 billion contract to build eighteen satellite platforms for the EU's IRIS2 constellation, announced 31 August 2026. It's OHB's largest documented deal yet, but what happens to both companies if this single, newly formed supply link hits a delay?

ReportedSpace · Ohb

OHB erhält Auftrag im Wert von knapp 1 Milliarde Euro für die Entwicklung und den Bau von achtzehn Satellitenplattformen für das europäische Satellitenkommunikationssystem IRIS² –

SESOHBKnapp

OHB just landed a near-billion-euro deal to build eighteen satellite platforms for Europe's IRIS² network. The order comes from SES, a Luxembourg operator navigating major shifts in the satellite market. Watch how OHB's heavy-lift launch dependency factors into delivering these platforms on schedule.

High impactPhysical AI · Therobotreport

Skild AI unveils S1 flagship robot foundation model

HumanoidFoundationFetch RoboticsSkild AI

Skild AI just launched S1, a robot foundation model that reportedly learns new tasks from a single video, skipping the usual retraining slog. The startup already backs major names in industrial automation. Which of Skild's three named robotics customers stands to gain most, and where does the model's humanoid coverage still fall short?

High impactPhysical AI · Bloomberg

LG Energy Signs Lithium-Supply Deal With Smackover in Arkansas

LG Energy SolutionSmackover Lithium

LG Energy Solution has locked in a 10-year lithium carbonate supply deal with Smackover Lithium in Arkansas, starting in 2029. LG's existing lithium sourcing runs through SQM in Chile at 0.8 confidence. The Smackover agreement's 8,000-ton annual volume sits against LG's Lansing and North American LFP capacity plans.

High impactAI Compute · Taipeitimes

Semicon to put spotlight on CPO, silicon photonics

AlphabetPowertech TechnologyAmazon.com, Inc.GlobalWafersAppleMicrosoft Corporation

Semicon Taiwan is putting copackaged optics and silicon photonics center stage as AI networking hardware matures. Taiwanese suppliers are already positioning across wafers, OSAT, and switches for this shift. Trace how Powertech's Broadcom joint venture fits into the CPO assembly chain before the mass-production wave hits.

NotableAI Compute · Yahoo Finance

Nvidia investing $3.5 billion in MediaTek for AI chip partnership

NVIDIA CorporationMediaTek

Nvidia is putting $3.5 billion into MediaTek's convertible bonds, deepening a partnership that already spans networking components and consumer platforms. The two sit worlds apart in scale, one a $5.3 trillion design giant, the other a Taiwan fabless player racing rivals like Broadcom for custom silicon business. What exactly does NVLink Fusion lock MediaTek into?

NotableAI Compute · Wccftech

Samsung Locks 70% of Memory Capacity Into 2031 Deals – Long-Term HBM Deals Run 5x Cheaper Than Spot, as AI Demand Chokes the DRAM Market

Microsoft CorporationNVIDIA CorporationSK hynix Inc.SK Hynix

Samsung has locked 70% of its DRAM capacity into deals through 2031, with long-term HBM pricing running far below spot. Follow how Micron's role shifts once Samsung and SK hynix both lock in their contracted volumes.

NotableAI Compute · Barchart

AMD Stock Alert: Advanced Micro Devices Teams Up With Cisco on AI Infrastructure

Advanced Micro DevicesCisco Systems, Inc.

AMD and Cisco announced a joint AI infrastructure push, tying MI355X GPUs to Cisco networking gear for hyperscale deployments. Both firms already lean hard on the same customer base and foundry sourcing. Trace what a new sovereign buyer in Saudi Arabia does to their shared concentration risk in Amazon, Microsoft, and Meta.

NotableAI Compute · Heraldcorp

SK hynix to tap Intel foundry for next-gen HBM base die

Intel FoundrySK hynix Inc.Intel CorporationSK Hynix

SK hynix is reportedly weighing Intel Foundry as a base die source for next-gen HBM, breaking from its long TSMC reliance. This would touch a component where SK hynix currently has zero manufacturing diversification. The base die's cost multiple versus core dies in current HBM4 production is the number worth tracking.

NotableAI Compute · Techtimes

SEMICON Taiwan 2026 Kicks Off: AI Chips' Bottleneck Is Wires Connecting Them

Taiwan Semiconductor Manufacturing Company LimitedGlobalWafersHon Hai Precision Industry Co., Ltd.Powertech TechnologyBroadcomLam Research

SEMICON Taiwan 2026 opens with a stark theme: AI chips are fast, but the wires linking them are the real chokepoint. Equipment spending is set to blow past $150 billion for the first time. What pitch is TSMC's SoIC hybrid bonding actually targeting by 2029, and can it arrive in time?

NotableAI Compute · Yahoo Finance

AMD Expands Global AI Footprint With New Saudi Platform

Advanced Micro DevicesCisco Systems, Inc.

AMD is teaming with Cisco and Saudi operator HUMAIN to deploy Instinct MI355X GPUs and EPYC CPUs on Cisco Silicon One networking, now live in Saudi Arabia. Does this Silicon One pairing mark AMD's first hyperscale-class win outside Meta, Amazon, and Microsoft?

High impactAI Compute · Yahoo

Nvidia (NVDA) Invests $3.5 Billion in MediaTek. Is This the Next Leg of Its AI Growth?

NVIDIA CorporationMediaTek

Nvidia is putting $3.5 billion into MediaTek via convertible bonds, turning a supplier relationship into a strategic bet on edge, automotive, and PC chip design. Watch what this does to MediaTek's already heavy reliance on Taiwan Manufacturing and UMC for fabrication, and what that concentration means for Nvidia's own Blackwell supply chain.

High impactAI Compute · Investing.com

China memory leader CXMT begins small-batch production of HBM3E silicon - report By Investing.com

ChangXin Memory TechnologiesT-HeadCambricon

China's CXMT has reportedly started small-batch production of HBM3E, closing the gap with global memory leaders. State media frames it as a milestone for domestic AI chip supply chains. Which Chinese chip designers are already running qualification tests on these HBM3E samples ahead of 2027 commercial integration?

High impactAI Compute · 247Wallst

This Little-Known AI Company Is Quietly Building for the Data Center Boom

Applied OptoelectronicsLumilensPOET Technologies Inc.Lumentum

POET Technologies just landed its first real customer contract, a purchase order from AI compute supplier Lumilens that could grow into a framework worth $500 million over five years. The brief tracks how POET's optical engines actually reach hyperscalers, and where Lumentum and Applied Optoelectronics fit into the same supply chain. POET's per unit pricing implied by its 2026 shipment target sits well under $2,000.

NotableAI Compute · Bloomberg

SK Hynix Weighs Japan Memory Fab Partnership to Supply AI Boom

SK hynix Inc.SK On

SK Hynix is reportedly exploring a joint venture in Japan to boost memory output for the AI boom, adding to its existing US and Korea fab plans. Watch how a Japan facility timeline stacks up against Yongin and Cheongju to see which comes online first for NVIDIA's HBM needs.

NotablePhysical AI · Korea Herald

SK On to supply W1.5tr ESS LFP battery to US NeoVolta Power

SK On

SK On will supply NeoVolta Power with 1.09 trillion won worth of LFP batteries for US grid storage, a fresh push beyond its EV cell business. The deal includes a tolling arrangement alongside direct supply. What does the split between cell manufacturing and pack assembly reveal about SK On's Georgia and Tennessee capacity strategy?

High impactAI Compute · Forbes

AI Data Center Investment Could Reach $7 Trillion By 2030: Who Wins

Vertiv

McKinsey pegs global data center investment at $7 trillion through 2030, and infrastructure players are racing to capture it. Vertiv sits on both sides of the NVIDIA supply chain as buyer and supplier. Trace how rack power climbing from 227 kilowatts toward 400 kilowatts reshapes who actually profits from this buildout.

High impactAI Compute · Korea Times

Applied Materials strives to cooperate with chipmakers to shorten equipment lead time - The Korea Times

Applied Materials

Applied Materials is racing to work more closely with chipmakers to shrink the wait for critical equipment. The company just posted record Q3 revenue, but its own Korea chief admits near-term demand can't be met. Check how the EPIC Center program and its 2-3 year delivery lag could reshape SK hynix and Samsung's fab timelines.

High impactPhysical AI · TechCrunch

The U.S. is building barriers around drones and robots, but China has scale to get around them

AgiBot

New U.S. tariffs and Covered List expansions aim to wall off Chinese drones and robots, but Chinese makers already dominate 86% of global humanoid shipments. What happens to AgiBot's IPO-bound growth, and its dependence on suppliers like RoboSense, once North America closes off?

ContextAI Compute · Trendforce

[News] TSMC: SoIC, CoWoS to Drive 50× Compute by 2029; Silicon Photonics to Top 50% of Transceiver Market by 2027

Taiwan Semiconductor Manufacturing Company Limited

TSMC says SoIC and CoWoS packaging will drive a 50x jump in compute performance by 2029, while silicon photonics is set to dominate over half the optical transceiver market by 2027. A new optical PDK just cleared a key design hurdle for electronic-photonic chips. TSMC's VP flagged lasers, fibers, and connectors as the real bottleneck standing between here and that 2029 target.

Sunday, August 30

High impactSpace · Reuters

NASA launches powerful Roman Space Telescope, a spy satellite turned to the stars

SpaceXThomson

NASA's Nancy Grace Roman Space Telescope launched aboard a SpaceX Falcon Heavy, extending the observatory lineage that includes Hubble and Webb. Follow the Falcon Heavy stack back through SpaceX's chip and memory suppliers to see what a $4 billion flagship mission really depends on.

High impactAI Compute · Startupfortune

China Builds the World's Best Robot Bodies but Still Can't Give Them a Brain - Startup Fortune

DeepSeekUBTech RoboticsFourier Intelligence

Unitree's Shanghai IPO rocketed its valuation from $9 billion to near $50 billion in two weeks, cementing China's grip on humanoid robot hardware. But BYD and Foxconn deployments reportedly hit only 30-50% of human productivity on the factory floor. What is DeepSeek's $20.8 million bet on Unitree actually meant to fix?

High impactAI Compute · Yahoo Finance

Advanced Micro Devices (AMD) Is Exposed To Proposed US Chip Tariffs

Advanced Micro Devices

Proposed US chip tariffs could hit AMD right where it can't easily move: its offshore manufacturing chain for EPYC and Instinct GPUs. AMD doesn't own foundries, so costs from partners like TSMC would flow straight through to margins. Will AMD absorb the hit or pass it to customers like Meta and Microsoft?

High impactAI Compute · SCMP

CXMT vs Pentagon: how China's tech firms are challenging US blacklists in court

ChangXin Memory Technologies

China's DRAM giant CXMT is fighting the Pentagon in court over its blacklist status, following recent wins by Hesai and DJI against similar designations. CXMT just raised $8.6 billion in Asia's biggest IPO of the year. Watch what a ruling could mean for Dell's PowerEdge XE9680 server supply chain.

High impactAI Compute · Investorplace

The $50 Trillion Robot Boom Starts at $10 an Hour

NVIDIA CorporationBoston DynamicsHyundai Motor

JPMorgan says humanoid robots could soon undercut $30-an-hour factory labor, and BMW's Figure 03 pilot in Spartanburg is already testing that math on the line. But scaling to 30,000 units means someone has to build the supply chain first. Which single South Korean supplier could bottleneck Boston Dynamics' entire Atlas rollout?

High impactAI Compute · Korea Herald

AI demand drives broader Samsung Electro-Mechanics MLCC price hikes

Samsung Electro-Mechanics

Samsung Electro-Mechanics is pushing MLCC price hikes beyond distributors, now hitting OEM and ODM manufacturers directly in Q4 2026. Mainstream capacitor inventories have dropped below 30 days as capacity shifts toward AI-grade parts. Lead times for high-capacitance MLCCs now stretch to roughly 21 weeks. The brief breaks down what this means for Samsung Electronics, Qualcomm, Tesla, Groq, and BYD's input costs.

NotablePhysical AI · The Verge

China’s robots race ahead

HumanoidDexterity Inc.Boston DynamicsNVIDIA CorporationAgiBot

Humanoid robots raced, jumped, and sometimes caught fire at Beijing's World Humanoid Robot Games, with Tiangong Ultra clocking a sub-Bolt 100m sprint. Behind the medal count sits a fragile web of chip, battery, and magnet suppliers keeping these machines upright. Trace how NVIDIA's Vera chip and Ningbo Yunsheng's rare earth magnets tie into the failures nobody's highlighting.

NotablePhysical AI · Cubaheadlines

ICE Plans to Acquire Advanced Robotic Dogs: When Will They Be Deployed?

Boston Dynamics

ICE has moved to acquire Boston Dynamics' SPOT quadrupeds in a $1-2 million deal, its first sustained operational rollout rather than a one-off trial. The robots rely on a tight supplier web spanning lidar, compute, and battery components. Does this procurement channel signal a bigger shift in who else is lining up to buy SPOT next?

NotablePhysical AI · Yahoo Finance

Electrovaya Targets Data Centers, Defense as Jamestown Battery Plant Nears 2027 Launch

Electrovaya

Electrovaya's Jamestown, New York plant is set to hit commercial production by early 2027, pushing the battery maker beyond forklifts into data centers and defense. Watch how the onshore facility reshapes a supply chain currently concentrated among five Asia-based suppliers feeding its ceramic-separator cell production.

NotablePhysical AI · Townhall

ICE Is Eyeing Another Huge Tech Purchase

Boston Dynamics

ICE is reportedly lining up a fresh multimillion-dollar Boston Dynamics purchase, expanding Spot's footprint into federal law enforcement. See how this field-deployment pivot bears down on the same chokepoint suppliers, NVIDIA, Velodyne, and LG, that already constrain Atlas production for Alphabet and Hyundai.

NotableAI Compute · Yahoo

Celestica Stock Leads 3 Cash Flow Picks In AI Infrastructure

Celestica

Celestica tops a list of cash-flow picks riding the AI infrastructure boom, with strong Q2 numbers and expanding capacity. But the stock's relationship with Broadcom runs both directions, as a supplier and a customer. What does that dual coupling mean for Celestica's risk profile if the AI buildout slows?

Saturday, August 29

High impactAI Compute · Techtimes

China Chip Equipment Hits 35%; Vacuum Valves and RF Power Remain 80% Foreign

SMICTaiwan Semiconductor Manufacturing Company LimitedLam ResearchApplied Materials

China says its chip-tool supply chain is 'significantly safer,' with domestic equipment now at 35% of installed fab tools, up from 25% a year ago. But vacuum valves and RF power supplies remain 80-90% foreign-sourced. If NAURA and AMEC dominate the visible process layer, why does the sub-component chokepoint still hold?

High impactAI Compute · DIGITIMES

Delta Electronics acquires Alps Alpine power inductor business assets

Delta Electronics

Delta Electronics is buying Alps Alpine's power inductor and powder materials business through its Japan and Korea units, folding a key passive component into its own supply chain. Trace what this vertical integration means for Delta's power rails feeding NVIDIA's next-gen AI systems.

High impactAI Compute · Reuters

China's CXMT to supply memory chip for Xiaomi's upcoming folding phone

ChangXin Memory Technologies

CXMT will supply LPDDR6 memory for Xiaomi's flagship 18 Fold foldable phone launching in September 2026. The deal follows CXMT's record-breaking $8.6 billion Shanghai IPO and a lawsuit against the Pentagon over a national security blacklist. Does this design win lock Xiaomi into an all-domestic supply chain that reshapes CXMT's exposure to Western customers like HP and Dell?

High impactAI Compute · Business Standard

Chipmaker CXMT sues Pentagon over 'Chinese military company' designation

ChangXin Memory Technologies

China's CXMT, fresh off a Shanghai IPO, is suing the Pentagon to shake off its 'Chinese military company' label. The designation threatens more than contracts. CXMT chips are already qualified inside Dell's PowerEdge XE9680 servers and HP's Pavilion and Spectre lines, putting real OEM exposure on the line.

High impactPhysical AI · WSJ

Nvidia Wants to Run the World's Robots. China Is an Eager Customer.

AgiBot

Nvidia is pitching itself as the brains behind humanoid robots, and China's manufacturers are lining up as buyers. One of them, AgiBot, is racing toward a Hong Kong IPO on the back of mass-produced units already in the field. Watch how AgiBot's entirely domestic supplier list, from magnets to vision systems, shapes what Nvidia can actually claim as its China footprint.

High impactAI Compute · Bloomberg

DWS Shifts AI Fund Bets Into Hyperscalers From Chipmakers

Alphabet

DWS is rotating AI fund capital out of chipmakers and into hyperscalers, betting the real margins sit with platforms that consume silicon rather than sell it. Fathom It maps Alphabet's own supply chain to test that thesis. With Frozen v2 slated for 2028, is Alphabet's internal chip roadmap already outgrowing its Broadcom and Samsung dependencies?

NotableAI Compute · Indiatimes

Chinese company that Apple made a 'special request' to the US government for has sued Pentagon

ChangXin Memory Technologies

CXMT, China's largest DRAM maker, is suing the Pentagon over its military-affiliated designation, a label that has hammered its commercial standing since January 2025. Apple reportedly lobbied US officials on the company's behalf while testing its memory chips for China-market iPhones and MacBooks. Watch what Apple's supply plans for HP, Dell, and SK hynix reveal about the stakes.

NotablePhysical AI · Gizmodo

ICE Wants a Boston Dynamics K9 Unit

Boston Dynamics

ICE is planning to buy 13 to 27 Boston Dynamics SPOT units for roughly $1 to 2 million, marking a new law enforcement customer segment for the robotics maker. See what this government order means for Boston Dynamics' existing NVIDIA, Velodyne, and LG supply relationships behind Atlas and SPOT production.

NotableAI Compute · Yahoo

IBM Introduces New Mainframe Processor Combining Its Own Tech With Arm Architecture

Arm Holdings

IBM has unveiled a new mainframe processor blending its own architecture with Arm technology, built on a 2-nanometer process with AI inference accelerators for fraud detection. Arm's prior chip relationships ran through NVIDIA, Amazon, Samsung, Alphabet and Microsoft in mobile and hyperscale markets. What does this IBM partnership signal about Arm's move into mission-critical enterprise compute?

NotablePhysical AI · Newsmax

ICE Plans to Buy Up to $2 Million in Boston Dynamics Robots

Boston Dynamics

DHS posted an acquisition forecast to buy up to $2 million in Spot robots and accessories for ICE, with solicitation set for Sept. 4 and award expected in Q4 FY2026. The deployment site is Fort Benning, Georgia, where ICE trains roughly 13,000 field officers.

NotableAI Compute · Investing.com

How networking became the critical layer of AI By Investing.com

Marvell Technology

Citi analysts at Hot Chips say networking, not raw compute, now decides who wins the AI race, naming Marvell among the key beneficiaries. Marvell just struck a fresh custom silicon deal with Google spanning inference, storage, and networking components. Watch how Marvell's Taiwan Manufacturing dependency shapes the risk behind that expanding hyperscaler footprint.

NotablePhysical AI · Thehill

ICE plans to spend $2 million on dog-like robots for immigration enforcement

Boston Dynamics

ICE is set to spend up to $2 million on Boston Dynamics' Spot robots for immigration enforcement, marking the first major US federal law enforcement order for the dog-like machines. But with LG Innotek and LG Energy Solution supplying critical Korean-sourced components, how exposed is this deal to supply chain concentration risk?

High impactAI Compute · Yahoo

How Amkor's AI-Focused Packaging Push At Amkor Technology (AMKR) Has Changed Its Investment Story

Amkor Technology

Bank of America now calls Amkor an underappreciated AI packaging beneficiary, pointing to capacity expansions and long-term HPC deals. Track the $100 million customer advance payment tied to a standby letter of credit and see whether it de-risks the Songdo facility ramp or just masks concentration risk in Amkor's AI-heavy customer base.

High impactAI Compute · Straitstimes

Chinese memory chipmaker CXMT sues Pentagon over 'Chinese military company' designation

ChangXin Memory Technologies

China's DRAM giant CXMT is fighting its Pentagon blacklist designation in federal court, just weeks after a blockbuster IPO valued it near half a trillion dollars. Trace what a court loss would mean for CXMT's HBM supply into HP and Dell's server and PC lines.

High impactAI Compute · Indiatimes

Andreessen Horowitz raises $1.1 billion for AI infrastructure fund

Cerebras Systems

Andreessen Horowitz just closed a $1.1 billion fund aimed squarely at AI infrastructure scarcity, from chips to networking to storage. The move validates the exact bottleneck companies like Cerebras exist to solve. So who actually feels the squeeze first when this capital hits the wafer queue at Taiwan Manufacturing?

NotablePhysical AI · Yahoo

Sanctuary AI Built A Robot Body . Now It Also Selling A Robot Brain

Sanctuary AIMeta Platforms

Sanctuary AI is repackaging its Physical AI system as a hardware-agnostic brain that can slot into existing industrial robots rather than relying solely on its own Phoenix bodies. It reports strong results on a wire-plugging task with Tier 1 automotive suppliers. What does this pivot mean for Sanctuary's Apptronik and Microsoft dependencies?

Earlier

Week of August 24 67 stories ▾

Friday, August 28

High impactPhysical AI · TechCrunch

Chinese automakers are following Tesla’s bet that robots are the next big profit machine

Mobileye GlobalXPengBoston Dynamics

Mobileye's $900 million grab of Mentee Robotics signals Intel's supplier arm wants a piece of the humanoid robot race, echoing Tesla's Optimus ambitions. Chinese rivals like XPeng are already pouring billions into their own robot supply chains. Which of Mobileye's chip and actuator partners now face new exposure as it pivots from perception software to full robot hardware?

High impactAI Compute · Indiatimes

SK Hynix to start AI chip output in Indiana in 2029, sees memory shortage through 2030

Alphabet

SK Hynix will begin volume production of HBM4E chips at its new Indiana plant in 2029, with its CEO warning the memory shortage could persist through 2030. Alphabet is named among SK Hynix's major customers, yet its supply chain concentration risk with the 58%-share HBM leader remains unquantified.

High impactAI Compute · Straitstimes

Sk Hynix $5 billion memory hub to boost US chip supply

SK hynix Inc.

SK Hynix is building a $5 billion HBM packaging hub in West Lafayette, Indiana, its first outside Asia, targeting mass production in late 2029. Federal CHIPS Act funding and thousands of jobs are riding on it. Watch which of the 100+ suppliers SK Hynix is vetting for the plant, including equipment names like Applied Materials, land the orders.

High impactAI Compute · DIGITIMES

ASIC makers see CSP orders stay strong through 2028

Broadcom

ASIC makers report cloud giants have locked in orders through 2028, with 2026 capex jumping 80% from last year. Broadcom sits deep in this chain, sourcing from Applied Materials and Samsung while guiding for over $100 billion in FY2027 AI revenue. Substrate lead times now stretch past 12 months, reshaping what counts as guaranteed demand.

High impactAI Compute · TechCrunch

Open-weight AI companies are the Valley’s hottest acquisition targets

DeepSeekNVIDIA Corporation

Open-weight AI startups have become the Valley's must-buy assets, with Nvidia and Stripe both writing multibillion-dollar checks. Nvidia sits at the center of this wave while its own Nemotron models struggle for adoption. Trace how Nvidia's Blackwell supply chain, dependent on Micron, SK hynix, and TSMC, connects to its acquisition push into the open-weight developer base.

High impactPhysical AI · Reuters

South Korea's export boom seen sustained in August on strong chip demand - Reuters poll

ThomsonSK hynix Inc.

South Korea's exports are on track to keep climbing 62.6% year-over-year in August, extending a 15-month streak fueled by AI chip demand. Samsung and SK hynix are the names behind the surge. SK hynix's new Yongin Y2 fab timeline hinges on whether this demand phase is cyclical or structural.

NotableSpace · SCMP

China’s chip supply chain ‘safer’ amid self-sufficiency drive: top planner

SpaceXHumanoidChangXin Memory TechnologiesNVIDIA Corporation

China's top planner says its chip supply chain is now 'significantly improved,' shifting from scattered breakthroughs to full-chain synergy. Foundry utilization reportedly tops 90 percent. See which household-name device makers have already qualified a homegrown memory maker's chips for their flagship products.

NotablePhysical AI · Nbcnews

ICE plans to buy robot ‘dogs’ for some immigration enforcement operations

Boston Dynamics

ICE is set to spend up to $2 million on Boston Dynamics robot dogs for enforcement work, opening a new government sales channel outside the company's commercial roadmap. If federal orders scale up, what happens to Boston Dynamics' supplier bottlenecks in lidar, batteries, and camera sensors already stretched thin by the Hyundai Metaplant deployment?

NotableAI Compute · Yahoo Finance

Kyndryl, Broadcom deepen partnership to build AI-ready private clouds

Broadcom

Kyndryl and Broadcom are joining forces to push VMware Cloud Foundation into regulated industries building AI-ready private clouds. Broadcom already guides FY2027 AI revenue above $100 billion on 10 gigawatts of hyperscaler shipments. Watch how this consulting push into VCF and Tanzu reshapes Broadcom's customer base beyond Alphabet, Apple, and Arista.

NotableAI Compute · Nasdaq

Intel Raises $23B to Fund AI Chip Capacity as 14A Roadmap Gains Momentum

Intel Corporation

Intel just secured $23 billion to fuel AI chip capacity as its 14A process hits defect-density milestones ahead of schedule. The raise reshapes capex plans across Arizona, Ireland, and Oregon fabs. Watch what this means for Intel's advanced packaging bet with Ibiden and AT&S on EMIB-T interposer-free tech.

NotablePhysical AI · Yahoo

Belden (BDC) Unveils New IT OT Networking Products

Belden

Belden just rolled out new IT/OT networking gear spanning industrial connectivity, wireless orchestration, rugged switching and data center infrastructure, building on its RUCKUS Networks integration. But does this launch actually loosen Belden's single-source dependency on Shin-Etsu Chemical for the rare earth magnets in its PatchSee products?

Thursday, August 27

High impactAI Compute · CNBC

SK Hynix CEO says Indiana will be key memory production base by 2030, first U.S. facility now underway

SolidigmNVIDIA CorporationSK hynix Inc.

SK Hynix broke ground on a $4 billion Indiana facility, with its CEO calling it a key HBM production base by 2030 and total U.S. investment set to top $45 billion. The Indiana plant handles one specific step in the memory supply chain, not the whole thing.

High impactAI Compute · EE Times

First Benchmarks Revealed for OpenAI's Jalapeño AI Chip

DeepSeekOracle CorporationDiodesDialogMicrosoft CorporationCelestica

OpenAI's Jalapeño chip just posted its first public benchmarks against rival AI silicon. The numbers look strong, but the real story is who built the pipeline behind them. Trace the Broadcom-Celestica manufacturing loop feeding this chip to see why the capacity buildout matters more than the specs.

High impactPhysical AI · Reuters

China’s humanoid robots aren’t smart enough to take your job – yet

HumanoidDexterity Inc.AgiBot

Reuters finds China's humanoid robots still lag on real intelligence, but investors aren't waiting. AgiBot is already prepping a Hong Kong IPO targeting a multi-billion valuation. Watch its reliance on Ningbo Yunsheng magnets and RoboSense vision systems, the two chokepoints that could decide whether that valuation holds.

High impactAI Compute · Ndtvprofit

Amazon To Buy Two Million Nvidia Chips For Data Center Build-Out

NVIDIA Corporation

Amazon is locking in two million Nvidia chips for 2027-2028 data center expansion, stacking onto a prior million-unit order. Nvidia's HBM comes from just two suppliers feeding this surge. With Amazon also building Trainium chips via Annapurna Labs, how much strain does this order actually put on Taiwan Manufacturing's wafer capacity?

High impactPhysical AI · Techtimes

Tesla AI5 Claims 2 - 3x Nvidia Efficiency ; Jalapeño Offers Closest Comparison

Tesla, Inc.Broadcom

Musk says Tesla's AI5 chip beats Nvidia 2-3x on efficiency, with Jalapeño cited as the nearest real-world comparison. But Jalapeño runs in data centers, not battery-powered robots. See how AI5's 250-watt Optimus power envelope stacks up against Jalapeño's benchmarked edge over Nvidia's GB300 before taking Musk's number at face value.

High impactAI Compute · Yahoo

Nvidia's latest earnings report shows its SpaceX relationship deepening

NVIDIA CorporationSpaceX

Nvidia's earnings show SpaceX moving from occasional buyer to core account, with chip and server ties expanding fast. Foxconn and SK hynix already thread through the supply chain behind that growth. So what does Vera CPU deployment and a jump to roughly 5% of Nvidia's revenue actually signal about SpaceX's hyperscaler status?

NotableAI Compute · Nextplatform

Peeling Apart That Supposed $120 Billion Chip Deal Google Inked With Marvell

Marvell Technology

Google and Marvell announced a commercial pact covering custom AI chips, storage controllers, and network silicon, pegged at roughly $120 billion over the deal's lifespan. The agreement leans on Marvell's existing foundry chain, and Taiwan Manufacturing's capacity allocation under this warrant structure is exactly what the brief maps out.

NotableAI Compute · Smirkingchimp

The New Cold War of Rocks vs. Chips

Intel CorporationMP MaterialsShenghe ResourcesBYDBaidu

Trump's Beijing meeting reportedly greenlit less-advanced Nvidia chips for China while leaving rare-earth access unresolved. The rocks-versus-chips framing hides an asymmetry in who actually gets squeezed. Trace how MP Materials' 500-tonne monthly magnet target and its Apple prepayment intersect with Intel's absence from BYD's chip supplier list.

NotablePhysical AI · Investing.com

SoftBank in talks for majority stake in humanoid robotics startup 1X- The Information By Investing.com

1X Technologies

SoftBank is reportedly in talks to take a majority stake in humanoid robotics startup 1X at a hefty valuation. The deal would inject serious capital and clout into the young Physical AI player. What does 1X's reliance on a single chipmaker for NEO's brain mean now that SoftBank owns the upside?

NotableAI Compute · Co

AWS Nvidia 2 Million GPU Deal Reshapes Agentic AI

NVIDIA Corporation

AWS just locked in 2 million Nvidia GPUs over multiple years, a bet on agentic AI reshaping cloud infrastructure. Every one of those chips still funnels through the same Blackwell supply chain, with HBM memory from just SK hynix and Micron capping how many usable units reach any customer.

NotableAI Compute · WSJ

Nvidia’s $279 Billion Supply-Chain Gamble

NVIDIA Corporation

Nvidia just doubled its supplier commitments to $279 billion in a single quarter, a massive bet to keep AI chip production running at full tilt. Trace where that capital actually flows through Nvidia's HBM and foundry bottlenecks to see which single-source partners are absorbing the biggest slice of the bet.

Wednesday, August 26

High impactAI Compute · TechCrunch

Anthropic continues compute-gobbling streak in $45B deal with Nscale

SpaceXAdvanced Micro DevicesBroadcom

Anthropic just added a $5 billion AMD compute deal on top of its $45 billion Nscale pact, deepening its infrastructure sprawl. Both moves route back to the same silicon chokepoint. Which single supplier's Vera Rubin chip rollout could stall Anthropic's late 2027 compute buildout before it even starts?

High impactPhysical AI · CNBC

Temasek-backed investor behind Unitree pitches Singapore as Chinese robots' path to U.S.

HumanoidMeituanPhysical IntelligenceUnitree RoboticsHorizon Robotics

A Temasek-linked investor is floating Singapore as a workaround for Chinese robot makers to reach American buyers, with Unitree as the flagship case. Unitree's Shanghai IPO pop has already faded fast amid a fresh U.S. humanoid ban. The brief maps Unitree's NVIDIA chip dependency against its 18% U.S. revenue exposure.

High impactAI Compute · Independent

AI in everyday devices: tokenised intelligence at the edge

GlobalFoundries

AI is moving off the cloud and into everyday devices, processing tokens locally instead of shipping them to a data center. That shift favors chipmakers built for low-power, deterministic silicon over general-purpose GPUs. Trace GlobalFoundries' foundry ties to Groq and Apple to see who actually supplies this edge-native shift.

High impactAI Compute · Moneycontrol

Amazon to buy 2 million Nvidia chips for data center build-out- Moneycontrol.com

Amazon.com, Inc.

Amazon is adding 2 million Nvidia GPUs for 2027 and 2028, stacking onto 1 million already announced for this year. That's 3 million units even as Amazon builds its own Annapurna Labs silicon. If Trainium is production-ready enough for OpenAI's gigawatt-scale commitment, why is Amazon still buying Nvidia at this volume?

High impactAI Compute · Asianewstoday

Lam Research Breaks Ground on New Oregon Lab to Accelerate AI Era Semiconductor Research & Development Locally, Globally

Lam ResearchMicron TechnologyIntel Corporation

Lam Research broke ground on a $3 billion Tualatin, Oregon lab set to expand cleanroom space over 50% by 2028, with Intel and Samsung executives on site. The move signals faster chipmaking R&D cycles as AI demand surges. Trace how Micron's HBM4E ramp timeline hinges on this facility's output.

NotableSpace · Spacenews

SpaceX to develop Starship launch site in Louisiana

SpaceXUnited Launch Alliance

SpaceX is planting a $100 billion Starship launch complex on 125,000 acres of Louisiana Gulf Coast, with ten pads and its own propellant and power systems. But what happens to SpaceX's semiconductor and compute buildout, including its Terafab investment, as this new site pulls capital away from Texas? Fathom It to see the supply chain read.

NotablePhysical AI · Techtimes

China's Humanoid Robots Win Factory Contracts: VLA Models Drive Orders

BYDHon Hai Precision Industry Co., Ltd.Galbot

Chinese humanoid robots are landing real factory orders as VLA foundation models replace task-specific code, splitting winners from demo-stage players. BYD and Galbot both sit at this intersection, one as buyer and supplier, one as a $3B+ pure-play with a thin but telling parts list. Galbot's supply chain includes Leaderdrive harmonic reducers.

NotableAI Compute · Techtimes

Diamond Rapids Disclosed: Intel Xeon 7 Packs 256 Cores and Gigabyte Cache, Drops SMT

Taiwan Semiconductor Manufacturing Company LimitedIntel CorporationMeta PlatformsQualcommSamsung Electronics Co., Ltd.Microsoft Corporation

Intel's Diamond Rapids Xeon 7 packs 256 cores and 1.28 GB of cache, dropping SMT in a major architectural overhaul unveiled at Hot Chips 2026. The chip also ditches EMIB for the open UCIe-S standard and builds entirely on Intel's own nodes. Trace what this foundry shift means for Intel's dependence on TSMC.

NotableAI Compute · Yahoo Finance

Applied Optoelectronics Has a Rare Problem Thanks to a Possible U.S. Ban on China

Applied Optoelectronics

A potential U.S. ban on Chinese optical transceivers could hand Applied Optoelectronics a flood of new hyperscaler orders, but its factories are already booked solid through mid-2027. Can AAOI actually scale fast enough to capture the demand, or does its own supplier Nexustest quietly cap how much it can build?

NotableAI Compute · Indiatimes

HP issues clarification for partnering with Chinese company that has been labeled as national security threat by America

ChangXin Memory Technologies

HP clarified its Huawei Wi-Fi patent license as routine cross-licensing, not a new partnership. But HP's memory sourcing tells a deeper story already underway. Trace HP's CXMT dependency for Pavilion and Spectre HBM chips, and watch what ASML's export exposure could mean for that supply line.

NotablePhysical AI · Nationalpost

Robot dogs are here and they're cheap, Canada, but their software still needs a human and their security has flaws

HumanoidUnitree Robotics

Chinese-made robot dogs from Unitree are now roaming Canadian campuses and fire departments for about $6,000, even as security researchers flag takeover flaws and the US bars new foreign robot models without defense clearance. Trace how Unitree's NVIDIA chip and LG battery sourcing ties into its post-IPO valuation swings.

High impactAI Compute · Bloomberg

Amazon to Buy 2 Million Nvidia Chips for Data Center Build-Out

Amazon.com, Inc.NVIDIA CorporationAmazon Annapurna Labs

Amazon has locked in 2 million Nvidia GPUs for its data centers over the next two years, deepening a supply tie that already spans its Proteus infrastructure. But Amazon also builds its own AI chips through Annapurna Labs. So why is it still betting this heavily on Nvidia instead of its in-house silicon?

High impactAI Compute · Ndtv

Nvidia's Costlier AI Chips May Force Silicon Valley To Learn From China

NVIDIA CorporationMeta PlatformsMicrosoft Corporation

Nvidia is hiking server prices over 15 per cent as memory costs climb, even as hyperscaler GPU utilization reportedly sits under 15 per cent. Meta and Microsoft keep buying anyway, locked into massive 2026 spend plans. So what happens to Nvidia's supply chain when a technique like DeepSeek's multi-head latent attention undercuts the hardware math entirely?

High impactPhysical AI · Indiatimes

Kerala's Maveli takes a robo avatar this year

Unitree Robotics

This Onam, Kerala malls are welcoming a robotic Maveli, powered by Unitree's G1 Edu platform and rented through local firm Rentobot, greeting shoppers and pulling in crowds. The deployment lands just after Unitree's turbulent Shanghai IPO. Watch how its LG Energy Solution batteries hold up against Kerala's monsoon humidity.

NotablePhysical AI · Statesman

Tesla is now operating a first-of-its-kind facility at Gigafactory Texas

Tesla, Inc.

Tesla is now running what it calls the first large-scale lithium battery cathode plant in the Americas, right at Gigafactory Texas, feeding its 4680 cells for Cybercab, Semi, and Model Y. This backward integration recalibrates Tesla's leverage over Contemporary Amperex and LG Energy Solution on battery supply and pricing.

NotableAI Compute · Indiatvnews

OpenAI takes on Nvidia with its first AI chip Jalapeo: Promises faster, more efficient AI - India TV News

Broadcom

OpenAI unveiled Jalapeo, its first custom AI inference chip co-designed with Broadcom, claiming 1.9x more work per watt and rollout by end of 2026. Watch how Broadcom's 10 gigawatt 2027 shipment ceiling, already shared across Google, Meta, and Anthropic, absorbs OpenAI's new silicon demand without straining capacity.

High impactAI Compute · Yahoo Finance

AMD's Next AI Accelerator Carries 432 Gigabytes of Memory -- 50% More Than the Last One. Here's the Bill That Comes With It.

Advanced Micro Devices

AMD's MI455X chip packs 432GB of HBM4 memory, a 50% jump that fuels its Helios rack-scale systems now entering production. That memory boost doesn't come free. Watch which single supplier now holds outsized leverage over AMD's margins as memory content per accelerator climbs to unprecedented levels.

High impactAI Compute · Cnbctv18

OpenAI claims its new chips can outperform Nvidia processors in tests - CNBC TV18

Broadcom

OpenAI says its new Jalapeno inference chip beats Nvidia's GB300 on throughput and latency while sipping just 700 watts. Broadcom co-developed the silicon, adding a fifth name to its custom ASIC roster alongside Google, Meta, and Anthropic. With Broadcom guiding over $100 billion in FY2027 AI revenue, how concentrated is that growth now on four buyers?

High impactAI Compute · The Register

Meta's new MTIA 400 chip has a split personality: Training AI and serving ads

Meta Platforms

Meta's MTIA 400 chip splits its workload between AI training and ad-serving, with the 450 and 500 already lining up for more bandwidth and compute. But whose IP framework actually stitches its multi-die design together, and what does that mean for Meta's chip independence?

NotableAI Compute · X

Samsung, SK Hynix to Increase Share of 8-High HBM4 Shipments, Signaling Changes in Nvidia’s Supply Chain (Zdnet Korea) Samsung Electronics and SK hynix are expected to increase the

Samsung Electronics Co., Ltd.LG전자(주)SK hynix Inc.

Samsung and SK hynix are set to ship more 8-high HBM4 to Nvidia in H2 2026, pulling back from the 12-high stacks tied to Vera Rubin accelerators. Watch what thermal limits on HBM4's 2,048-channel I/O reveal about why Nvidia wants both configurations in its supplier mix.

ContextPhysical AI · Apnews

A robot sprinter shatters a 100-meter record again as sparks fly at China’s robot competition

Humanoid

Tiangong Ultra blazed to a new 100-meter record at Beijing's World Humanoid Robot Games, adding another spectacle to China's robotics showcase. But the win lands weeks before a Q4 Beta rollout for a freshly minted unicorn. Could this podium moment reshape supply commitments with Bosch, Schaeffler, and Siemens?

Tuesday, August 25

High impactPhysical AI · Businessinsider

The US wants to lead the robotics race. Its hardware still leans on China.

HumanoidStandard BotsPhysical Intelligence

Washington wants a homegrown robotics industry, but the parts inside US-branded robots often come from Chinese factories. Startups like Standard Bots build for Amazon while sourcing hardware from multiple Chinese suppliers. Standard Bots' supply chain for its Amazon-supplied robot platform runs through named Chinese component makers.

High impactAI Compute · Therobotreport

Jetson Orin Nano 2 doubles inference performance for robotics on the edge, says NVIDIA

NVIDIA CorporationAlphabet

NVIDIA says its new Jetson Orin Nano 2 doubles edge inference performance while cutting power draw 40%, aiming squarely at robotics builders who already run on its chips. Trace what this efficiency jump means for Tesla's Optimus, Figure AI, and Boston Dynamics' Atlas deployments before you assume it's just a spec bump.

High impactAI Compute · Asianetnews

Infineon to acquire Bengaluru's C2i Semiconductors for AI data centres

Infineon Technologies AG

Infineon is buying Bengaluru-based C2i Semiconductors to bolt software-defined power control onto its chip lineup for AI data centres, with the deal expected to close in Q3 2026. What does C2i's multiphase controller tech actually add to Infineon's existing supply ties with NVIDIA's Blackwell platform?

High impactAI Compute · Hani

Why SK Hynix is turning to Japan for its next memory hub

Taiwan Semiconductor Manufacturing Company LimitedMicron Technology

SK Hynix is eyeing a Miyagi hub to lock in Japanese suppliers and dodge export-control risk. Micron and TSMC already have deep Japan footprints, from Hiroshima's DRAM line to Kumamoto's fabs. Watch what Tokyo Electron's 104 billion yen Miyagi plant signals about who gets privileged access next.

High impactSpace · Payloadspace

White House Wants 1,000 Annual Launches and Reentries by 2030

SpaceXRocket Lab

A new White House space policy targets 1,000 launches and reentries a year by 2030, a tenfold jump from today, and scraps NASA's old heavy-lift mandate in favor of commercial providers. Watch how SpaceX's NVIDIA and SK hynix dependencies factor into whether that cadence is even supply-chain feasible.

High impactPhysical AI · Barchart

Aptiv Accelerating Production-Ready Physical AI with Support for NVIDIA Jetson Orin Nano 2

Aptiv

Aptiv is deepening its NVIDIA ties, adding support for the Jetson Orin Nano 2 to push physical AI from prototype to production. The move bundles Aptiv's own sensing hardware with NVIDIA's compute stack for robotics and autonomy customers. What does Aptiv's PULSE camera and radar package actually plug into on the Orin Nano 2 platform?

NotablePhysical AI · Indiatimes

Oak Ridge and Idaho National Lab Print New Steel Pressure Vessel to Combat U.S. Forging Shortage

Lincoln Electric

Oak Ridge and Idaho National Lab printed a steel pressure vessel using large-scale wire arc additive manufacturing, targeting components weighing 10,000 pounds or more to ease America's forging capacity crunch. Which under-the-radar manufacturer built the path-planning software and monitoring systems behind this 100 pounds per hour deposition rate?

NotableAI Compute · Yahoo

TSM Rides on AI, Advanced Nodes & Capacity Expansion: Worth a Buy?

Taiwan Semiconductor Manufacturing Company Limited

TSM keeps riding AI demand, advanced node wins, and capacity expansion headlines as investors ask if the rally has room left. But the stock's fate hinges on names most headlines skip. Watch how tightly NVIDIA and AMD concentration ties into TSM's revenue volatility before you call this a buy.

NotableAI Compute · Barchart

Applied Optoelectronics Has a Rare Problem Thanks to a Possible U.S. Ban on China

Applied Optoelectronics

A possible US ban on Chinese optical transceivers could flood Applied Optoelectronics with orders it can barely fill. The company already runs at full capacity, booked solid into Q2 2027, while hyperscale giants Alphabet, Amazon, and Meta drive demand. AAOI's 2027 target of over 930,000 monthly units sits at the center of whether it wins or misses this window.

NotableAI Compute · Wccftech

NVIDIA CoWoS AI Chip Packaging Demand To Be Foreshadowed In 2027 By AMD Due To Agentic AI CPU Ramp , Says Morgan Stanley

NVIDIA Corporation

Morgan Stanley sees NVIDIA's TSMC CoWoS allocation slide from 53.4% in 2026 to 45.6% in 2027 even as its wafer demand climbs 57% to 1.2 million, with AMD's agentic AI ramp eating the difference. Watch how NVIDIA's monolithic Vera CPU stacks up against AMD's wafer-hungry MI455X design under this shift.

NotablePhysical AI · Businessinsider

Amazon's (AMZN) New "Tetromino" Plan Aims to Automate Delivery Stations

Agility Robotics

Amazon's Tetromino plan bets big on automating delivery stations, with Boxbot tapped for sortation while Digit humanoids get tested nearby for package handling. Watch how Agility Robotics, the maker of Digit, juggles NVIDIA chip allocation and precision motion suppliers as it races to serve both Amazon and its existing Schaeffler contract.

NotableAI Compute · Hothardware

Intel Crescent Island GPU Flexes 32 Xe3P Cores , 480GB LPDDR5X For Agentic AI

Intel Corporation

Intel's Crescent Island GPU pairs 32 Xe3P cores with up to 480GB of LPDDR5X, targeting the memory-hungry prefill stage of LLM inference. Onboard cards ship with 160GB. Which foundry and packaging partners actually get Intel's die-area gamble to market, and does its new sales chief have the customer base to move it?

Monday, August 24

High impactAI Compute · Design Reuse

TSMC Completes Development of 1 . 6nm A16 Process ; Mass Production Set for Q4

Taiwan Semiconductor Manufacturing Company LimitedIntel Corporation

TSMC says its 1.6nm A16 process is ready for Q4 mass production, promising faster, denser chips for AI leaders like NVIDIA, AMD, and Tesla. But process leadership isn't the only supply story here. Watch where TSMC's CoWoS packaging overflow is actually landing, and what that means for Intel's Malaysia facility.

High impactAI Compute · Semiengineering

Multi-Die Assemblies Dominate at 2nm and Below

Lam ResearchXilinx

Chipmakers below 2nm are ditching single-die designs for stacked, heterogeneous chiplet architectures to keep AI performance climbing. Lam Research sits at the center of this shift, shipping new panel-scale tools to customers racing to assemble multi-die packages. What do Lam's 510x515mm panel system shipments reveal about which chipmakers are already committing capex to this transition?

High impactAI Compute · SCMP

Cut off from Europe, Nexperia China eyes self-reliance through domestic wafers

Nexperia

Nexperia's China unit is ditching European 8-inch and 6-inch wafers for a domestic 12-inch supplier, severing a supply line that once fed roughly 70 per cent of its global output. This follows months of governance conflict between Nexperia's Dutch management and Chinese parent Wingtech. Vishay Intertechnology's fab relationship with Nexperia now sits on uncertain footing.

High impactAI Compute · Techtimes

Goldman Sees China 34% Short of Chip Self-Sufficiency in 2035; SMIC Yield Is Fragile Key

SMICASMLTaiwan Semiconductor Manufacturing Company LimitedHuawei Technologies Co., Ltd.Alibaba GroupBaidu

Goldman projects China's chip self-sufficiency gap narrowing to 34% by 2035, resting almost entirely on SMIC's fabrication capacity as the sole domestic alternative for advanced-node AI chips. Follow the wafer-start math behind that forecast, and see what it actually assumes about SMIC's DUV yield without EUV access.

High impactSpace · Reuters

Taiwan issues indictments over alleged illegal export of AI servers to China

SpaceXThomsonNVIDIA Corporation

Taiwan has charged nine people, including employees at NVIDIA and Super Micro Computer, over a scheme to smuggle AI servers into China through Indonesia, Japan, and Hong Kong using falsified end-user certificates. Only 74 of 130 Super Micro B300 servers reportedly reached China. What happened to the other 56, and who caught them?

High impactPhysical AI · Newatlas

Humanoid 'Olympics' reveal why the US can't compete with China's robots

HumanoidXiaomiXPeng

China's robots dominated the World Humanoid Robot Games in Beijing, with one machine even beating Usain Bolt's 100-meter time. The win says less about robotics engineering than about who already owns the supply chain. Our graph maps how Xiaomi and XPeng route batteries, LIDAR, and actuators through the same suppliers powering their EV lines.

NotablePhysical AI · Fntimes

Hyundai Steps Up U.S. Robot Lobbying as China Race Heats Up

Boston Dynamics

Hyundai Motor Group hiked its U.S. lobbying spend 73% in H1 2026, adding robotics as a standalone legislative priority as competition from Chinese firms like Unitree intensifies. Trace how a Velodyne Lidar sole-source dependency on the Atlas platform could be reshaped by the national robotics framework Hyundai is pushing Washington to define.

NotableAI Compute · Cryptobriefing

Largan Precision and TSMC partner to chase AI opportunities in co-packaged optics

Taiwan Semiconductor Manufacturing Company Limited

TSMC and Largan Precision are teaming up to build co-packaged optics for AI data centers, aiming to fix bandwidth bottlenecks in chip interconnects. What does a sub-0.3-micron collimator requirement mean for supplier concentration risk once CPO output actually starts feeding AMD's chip programs?

NotableAI Compute · Businessinsider

Amazon plots a new 'Tetromino' warehouse where robots tackle work that's notoriously hard to automate

Amazon.com, Inc.

Amazon's new Tetromino delivery stations aim to process packages at 2.5 times the speed of existing designs by automating the industry's toughest final-mile bottleneck. A startup called Boxbot is emerging as a potential vehicle-loading technology partner for the rollout, and it doesn't appear anywhere in Amazon's existing supplier network.

ContextAI Compute · Chipsandcheese

Hot Chips 2026: CUDA Targets RISC-V - by Chester Lam

NVIDIA Corporation

Nvidia is bringing CUDA support to RISC-V CPUs, moving beyond its longtime x86-64 and aarch64 lock-in. The move sets strict new hardware bars for any chip hoping to join the CUDA stack. Check the RVA23 and PCIe cache coherency requirements Nvidia is quietly imposing on RISC-V server CPU vendors.

Week of August 17 68 stories ▾

Sunday, August 23

High impactSpace · TechCrunch

TechCrunch Mobility: The custom chip driving Waymo’s robotaxi ambitions

SpaceXServe RoboticsSocionextNVIDIA CorporationEinrideSanDisk

Waymo just unveiled a custom 5nm ASIC pushing over 1,000 TOPS to power its Ojai robotaxis in LA, Phoenix, and San Francisco. The chip is framed as a profitability lever and a shift away from merchant silicon. What does this vertical integration mean for Nvidia's DRIVE AGX Thor and its wider AV footprint?

High impactAI Compute · Wccftech

SK Hynix Unpacks Advanced Packaging Technologies, Including Intel EMIB, For Its Next-Gen HBM Memory As It Eyes 3D Structures For The Future

SK HynixSK hynix Inc.

SK Hynix is bringing in Intel's EMIB packaging tech to push HBM beyond current stacking limits as it eyes 3D structures for future memory. Watch how Hybrid Bonding and sub-18-micron TSV pitch targets aim to unlock scaling past 16-Hi stacks without thermal collapse.

High impactAI Compute · Koreajoongangdaily

Samsung, SK hynix pour fresh investment into aging China factories despite U.S. curbs

Samsung Electronics Co., Ltd.

Samsung and SK hynix are ramping up spending at their older Chinese memory plants, reversing years of caution despite Washington's export curbs. Samsung's Xi'an investment jumped 68 percent and SK hynix poured over 1 trillion won into Wuxi and Dalian. Trace how the Xi'an plant's 40 percent NAND share collides with Samsung's 2028 capacity wall.

High impactPhysical AI · Baomoi

China Controls 85% of Robots, Preparing for New Tech Shock

AgiBot

China now ships roughly 85% of the world's humanoid and quadruped robots, and the industry is bracing for its next tech leap. But as Shanghai's AgiBot pushes toward a multi-billion dollar Hong Kong IPO, what happens to its supply chain if a single foreign chipmaker still powers its robots' brains?

High impactAI Compute · WSJ

Nvidia Is Spending $6 Billion to Build a Powerful U.S. Alternative to Chinese AI

DeepSeekNVIDIA Corporation

Nvidia is putting $6 billion behind its own open-weight AI model to counter Chinese rivals like DeepSeek. The deal doesn't touch chip supply, but it does say something about how Nvidia views its hardware moat. Poolside's licensing terms sit inside the brief.

High impactAI Compute · Wccftech

Micron Says AI’s Memory Wall is Worsening as Compute outruns HBM Bandwidth by 3x every Two Years, Advanced Packaging & Process To Tackle Rising Thermal Constraints

Micron Technology

Micron warns that AI compute is outpacing HBM bandwidth growth, with stack heights already hitting thermal ceilings at 16-high. The company's response leans on HBM4E ramping in 2027. Watch how Micron's SK hynix dependency and the Hiroshima 2028 timeline could reshape supply commitments to NVIDIA, Tesla, and Samsung.

NotablePhysical AI · Cnx Software

Quectel Smart Modules SH602FA and SE505FE Enable Android 16 Support in 4G LTE and 5G IoT Devices

Quectel

Quectel just launched two Android 16 smart modules, the 4G SH602FA and 5G SE505FE, built on MediaTek chips instead of Qualcomm. Both target robotics and IoT devices with drop-in form factors for easy 4G to 5G swaps. Could this chipset shift funnel into next-gen platforms at D-Robotics or Autel?

NotableAI Compute · CNBC

What's bogging down the data center trade has nothing to do with demand

Micron TechnologyBroadcom

Micron is pouring $100 billion into four new fabs in Boise and Clay, chasing DRAM capacity rather than chasing demand. But CXMT's IPO just rattled its stock twice in one month. Can Micron's HBM4E ramp outpace Chinese competition before 2027 arrives?

NotablePhysical AI · 36Kr

Zhiyuan Robotics Deep Dive: The Evolution Theory of Humanoid Robot Unicorns

AgiBot

AgiBot is chasing a Hong Kong IPO near HK$40-50 billion as it scales humanoid production toward volume manufacturing. Track how deep the RoboSense vision-system relationship really runs across its Expedition A2 line before you assume that supplier tie is easily replaceable.

Saturday, August 22

High impactPhysical AI · Autonocion

A 3-by-5-foot American steel pressure vessel just got printed in Tennessee by three robotic arms laying molten wire at 100 pounds an hour — Oak Ridge and Idaho National Lab want 10

Lincoln Electric

Three robotic arms in Tennessee just printed a steel pressure vessel by laying molten wire at 100 pounds an hour, and the labs behind it want ten more, including 10,000-pound nuclear parts made without a forge. Who actually built the software steering those three live welding arcs without collision?

High impactPhysical AI · Seeking Alpha

Fast track to the future: China's Humanoid Robot Games goes mainstream (HUMN:BATS)

Fourier Intelligence

Beijing's Humanoid Robot Games this weekend put 2,056 robots from 666 teams through real assembly and rescue tasks, pushing China's humanoid sector from lab demo to commercial showcase. Fourier Intelligence's GR-1 and GR-2 platforms are on the floor. Fourier's chip supply chain for those platforms runs through a single dominant foreign vendor.

High impactPhysical AI · Yahoo

8 Experts Weigh In On The FCC Foreign Robot Ban: Good Or Bad?

Apptronik

The FCC's new Covered List ban on foreign robots has experts split on whether it protects US industry or just raises costs. See what Apptronik's supplier map, including its single-source NVIDIA chip dependency, reveals about who actually wins this fight.

High impactSpace · Jatan

Skyroot Vikram-1, the least worst case scenario for India’s launch crisis

SpaceXSkyroot AerospaceRocket Lab

Skyroot's Vikram-1 reached orbit on its first try, a rare feat for any rocket program. But India's launch bottleneck isn't about proving a rocket works, it's about capacity math. Where does a ~300kg small-lift vehicle actually fit against Rocket Lab's Electron and India's own SSLV in an already thin market?

ContextAI Compute · Digitaltrends

AMD’s next-gen Medusa APUs are getting ready for launch, and Linux just spilled the beans

Advanced Micro Devices

Linux firmware leaks just confirmed AMD's Medusa Point APUs are packing RDNA 4m graphics and Zen 6 cores ahead of their official reveal. Watch how this pilot-stage ramp lands on Taiwan Manufacturing's capacity queue, already stretched by AMD's $5 billion capex push through 2028.

Friday, August 21

High impactSpace · TechCrunch

Starcloud raises $250 million for orbital data centers as launch options dry up

SpaceXNVIDIA CorporationRocket Lab

Starcloud just landed $250 million to build data centers in orbit, betting big as rocket availability tightens. The raise leans on a next-gen NVIDIA chip built specifically for space. With Falcon 9 set to retire in 2028 and no proven successor yet flying, what happens to Starcloud's launch pipeline in the meantime?

High impactPhysical AI · Therobotreport

Schaeffler plans to mass produce gearboxes for humanoid robots in 2027

HumanoidSchaeffler

Schaeffler says it will mass produce strain wave gearboxes for humanoid robots by 2027 using a new forming process instead of machining. The company already ships components to makers like Rainbow Robotics and NEURA Robotics. What does a component comprising roughly half of humanoid manufacturing costs look like once forming replaces machining at scale?

High impactPhysical AI · Businessinsider

Chinese humanoids steal the spotlight at San Francisco's robot party

HumanoidPhysical IntelligenceNVIDIA CorporationGenesis AI

San Francisco's Actuate conference drew 1,500 attendees and Chinese humanoid makers dominated the buzz, even as record robotics funding pours in. Yet actual hardware stayed scarce on the floor. Trace where NVIDIA's chip supply actually reaches across the humanoid stack, from Tesla to Figure AI to 1X.

High impactAI Compute · Chosun

South Korea's Humanoid Robots Rely on Chinese Bodies, American Brains

NVIDIA Corporation

South Korean robotics labs are pairing Chinese Unitree humanoid bodies with American NVIDIA compute, sidelining domestic makers like Robotis and Rainbow Robotics. The two nations dominate hardware and brains alike. But how does Jetson Orin's grip on Unitree's SDK ripple through NVIDIA's own supply chain, from Taiwan Manufacturing to HBM partners like SK hynix?

High impactSpace · Foxnews

Trump to transform spaceflight from spectacle to status quo targeting 3 launches per day

SpaceXRogers

Trump's new space policy pushes for over 1,000 launches a year by 2030, roughly three a day, cementing SpaceX as the dominant U.S. launch provider. That scale-up forces a reckoning with SpaceX's chip suppliers. The Terafab semiconductor complex in Grimes County, Texas, backed by a $16.8 billion investment, sits at the center of how SpaceX plans to feed that demand.

High impactAI Compute · Chosun

Samsung, LG Develop Robot Vision; Startups Focus on AI, Joints

Samsung Electro-MechanicsLG Innotek

Samsung and LG are racing to turn smartphone camera know-how into robot vision, while Korean startups chase AI chips and joint actuators. Samsung's new RX Task Force and LG Innotek's fresh TDK tie-up hint at bigger moves. Watch how LG Innotek's Boston Dynamics Atlas supply role ties into Hyundai Mobis's actuator plans.

NotableSpace · Arstechnica

Trump’s space transportation policy calls for new spaceport on federal land

SpaceXAstraRocket Lab

The White House wants a new federal spaceport, adding to a launch infrastructure push that already favors incumbents. SpaceX, Rocket Lab, and Astra sit at wildly different scales in this race. SpaceX's own compute capacity target of 2 gigawatts by year-end 2026 shows how far ahead it's positioning.

NotableAI Compute · Thenextweb

Brazil is building two AI supercomputers, one Chinese and one American

NVIDIA CorporationHuawei Technologies Co., Ltd.

Brazil is splitting a $444 million AI compute buildout between Huawei and Nvidia, one supercomputer in Rio and one still pending tender, framed by Lula's government as a deliberate hedge against depending on a single company or country. Watch the roughly $193 million Nvidia tender for what it signals about NVIDIA's own Taiwan-concentrated supply chain.

NotablePhysical AI · Ibtimes

NSK, Atom sign Japan humanoid robot MOU

NSK Ltd.

Bearing giant NSK has signed an MOU with Tokyo humanoid startup Atom, pairing precision components with a new physical AI robot maker. NSK already ships roller screws to at least one other humanoid developer, RT Corporation. Trace how NSK's actuator supply commitment to Atom fits its wider robotics push.

ContextPhysical AI · TechCrunch

Tesla, Uber, and Waymo all get the OK to operate thousands of robotaxis in Nevada

Tesla, Inc.

Nevada regulators gave Tesla, Uber, and Waymo the green light to run thousands of robotaxis across Clark County starting now. Tesla can deploy up to 5,000 vehicles, but its own Cybercab chief engineer testified under oath about a far lower real target. What number is Tesla actually planning to hit, and why?

Thursday, August 20

High impactAI Compute · CNBC

Japan exports beat estimates in July as robust chip shipments power fifth straight month of gains

SK HynixSK hynix Inc.

Japan's chip exports rose 23.2% in July, beating forecasts for a fifth straight month, with semiconductor equipment shipments jumping 49.1% on AI-driven demand. Which memory chipmaker's multi-year fab buildout, including a 2027-2029 HBM expansion, does this equipment surge actually de-risk? Fathom It to see the supply chain it feeds.

High impactPhysical AI · Reuters

Chinese companies to watch at the World Robot Conference

HumanoidFourier IntelligenceDeep RoboticsThomsonRoboteraLimX Dynamics

China's humanoid and quadruped makers are heading to the World Robot Conference to court investors and customers ahead of showcase season. Trace which single chipmaker sits underneath nearly every one of these five exhibitors' product lines before you assume their supply chains are diversified.

High impactAI Compute · DIGITIMES

Analysis : Google keeps Sunfish at Broadcom , Zebrafish at MediaTek Marvell US$12 . 2bn warrant buys a category nobody was defending

BroadcomMarvell TechnologyMediaTek

Google is splitting its TPU designs, keeping Sunfish at Broadcom while handing Zebrafish to Marvell via a $12.2bn warrant for 58,970,907 shares. Broadcom and Marvell serve almost entirely different hyperscale customers, yet both quietly depend on the same foundry chokepoint. Samsung Electronics sits underneath this whole arrangement.

High impactAI Compute · Wccftech

Intel’s Razor Lake Will Reportedly Tap TSMC’s 2nm N2X Process, Squeezing AMD Out of Already-Constrained Fab Capacity

X-FabIntel Corporation

Intel's next-gen Razor Lake chips will reportedly split production between its own 18A node and TSMC's cutting-edge N2X process, tightening an already squeezed fab queue. That allocation decision has direct consequences for how much TSMC capacity AMD can secure for its server CPUs. Intel's own $15 billion equipment bet suggests 18A still can't carry the full load alone.

High impactAI Compute · Nasdaq

Cerebras Unveils CS-4, OpenAI and AMD Partnerships to Accelerate AI Inference

Advanced Micro Devices

Cerebras launched its CS-4 platform with new OpenAI and AMD partnerships aimed at speeding up AI inference. AMD supplies the GPU muscle for one half of this split-stack setup. Trace what AMD's role in the prefill-decode split, its Taalas acquisition, and that fresh $4-5 billion debt raise actually signal about its inference ambitions.

High impactPhysical AI · SCMP

Growth driver: Hesai bets on lidar boom amid surge of smart cars in China

Hesai Group

Hesai is riding China's lidar boom as smart-car adoption accelerates across mid-market vehicles. The company supplies sensors to Xiaomi, XPeng, BYD, and even Amazon's warehouse robots. Trace what happens if Xilinx, Hesai's sole AI-compute source for the AT128, hits a capacity or export-control wall.

NotableAI Compute · Forbes

Why Nvidia Is Building Data Centers, Not Just The Chips Inside Them

Meta PlatformsNVIDIA CorporationSpaceXAlphabet

Nvidia's chip dominance was never the whole story, and now the company is pouring money into physical data center infrastructure rather than just silicon. What does Nvidia's $1.5 billion SB Energy investment and its role in the PORTS-Pike Technology Campus reveal about who actually controls AI compute going forward?

NotableAI Compute · Indiatimes

Marvell Technology partners with Google for custom AI chips, Google gets $12.2 billion stake option

Marvell Technology

Marvell just landed a warrant deal with Google worth up to $12.2 billion, sending shares nearly 8% higher. Broadcom, the incumbent Google custom-chip partner, dropped over 5% on the news. Trace how Marvell's foundry dependence on Taiwan Manufacturing and its existing ASIC ties to Amazon, Meta, and Microsoft shape this new concentration risk.

NotableAI Compute · Yahoo

Alchip Technologies Ltd (TPE:3661) (Q2 2026) Earnings Call Highlights: N3 AI Accelerator Drives ...

Alchip Technologies

Alchip's N3 AI accelerator ramp pushed Q2 revenue to USD 241.7 million, up 82.6% sequentially, with advanced nodes now 47% of output. North America revenue jumped from 23% to 51%. But networking, the component Alchip supplies to d-Matrix and Amazon's Proteus platform, is now just 16% of revenue. What does that shift mean for those two sourcing relationships?

NotableAI Compute · Cyprus Mail

Samsung unveils next-gen 3D-memory vision at FMS 2026, charting AI infrastructure future

Samsung Electronics Co., Ltd.

Samsung used FMS 2026 to lay out its next-gen 3D-memory roadmap, including a stacked HBM concept aimed at AI accelerators. The zHBM design sits directly above the chip rather than beside it, and Samsung's own claimed performance multiples raise questions about its current Micron HBM sourcing arrangement. The Fathom It brief maps what changes.

Wednesday, August 19

High impactPhysical AI · Reuters

Unitree soars in Shanghai debut, a milestone for China's humanoid robotics sector

Boston DynamicsAgiBotHumanoidDeep RoboticsMech-MindDeepSeek

Unitree's Shanghai IPO exploded nearly six-fold, valuing the humanoid robotics maker near $50 billion despite few robots actually deployed outside labs and universities. Trace the strategic investor behind Unitree's IPO placement whose own AI models now feed the humanoid's brain, and see how that tie-up collides with a US import ban on foreign humanoid robots.

High impactSpace · The Register

Google pits Marvell against Broadcom as it chases AI crown

SpaceXAlphabetS TechMicrosoft CorporationBroadcomCelestial AI

Google just handed Marvell a massive warrant tied to its TPU roadmap, a direct challenge to Broadcom's long-standing custom silicon grip. Broadcom shares dropped 4% on the news. What does Marvell's Celestial AI photonic tech actually let Google do to its TPU clusters that Broadcom couldn't?

High impactAI Compute · Chosun

SK Hynix Leads NVIDIA Sales, Samsung Diversifies

SK HynixNVIDIA CorporationSamsung Electronics Co., Ltd.SK hynix Inc.

SK Hynix rides NVIDIA's Blackwell platform as its top customer, while Samsung leans on a broader chip roster spanning Alphabet and Tesla. Track how NVIDIA's share of SK Hynix's H1 2026 revenue moved even as its total sales jumped 61.7 percent year-on-year.

High impactPhysical AI · Reuters

Germany's VDMA lobby calls on Europe to step up on humanoid robotics as Unitree surges

HumanoidUnitree RoboticsD-Robotics

German lobby group VDMA is urging Europe to build its own humanoid robotics supply chain as China's Unitree rides a record-breaking IPO wave. Track how deep Unitree's dependence on NVIDIA processors and Chinese vendors like Sharpa and Hesai actually runs, and see who Europe's answer might be, before betting on either side of this race.

High impactAI Compute · Techtimes

Marvell Lands Google AI Chip Deal, Closing the Custom Silicon Hyperscaler Sweep

BroadcomMicrosoft CorporationAlphabetMeta Platforms

Marvell now designs custom silicon for Microsoft, Meta, and Google, completing a hyperscaler sweep that Broadcom has long dominated. The Google deal came with a stock-vesting warrant tied directly to future purchase volume. What total revenue would Marvell need from Google for that warrant to fully vest by 2033?

High impactAI Compute · Techtimes

Samsung Commits $170M to Gwangju CDU Factory as AI Cooling Demand Exceeds Supply

VertivSamsung Electronics Co., Ltd.NVIDIA Corporation

Samsung is pouring $170M into its Gwangju plant to build coolant distribution units for AI servers, betting that liquid cooling demand will outpace supply. It's the site's biggest manufacturing bet in 37 years, converting an old appliance factory into cooling hardware infrastructure. Vertiv's capacity timeline against this new entrant is worth examining.

NotableAI Compute · Benzinga

Nvidia H200 Chips Head Back to China: Report - NVIDIA (NASDAQ:NVDA)

NVIDIA CorporationLenovo

Nvidia's H200 chips are trickling back into China after Beijing quietly cleared a limited batch for select buyers. But the shipment size tells a different story than the headline suggests. Track the exact unit counts ByteDance and Tencent actually received against the 75,000-chip ceiling Washington approved back in May.

NotablePhysical AI · Reuters

China robot makers flock to Beijing show, seek path to mass adoption

RoboteraD-Robotics

Chinese humanoid makers packed the Beijing show as Unitree's IPO soared nearly six-fold, fueling talk of new listings from Lumos Robotics and Chery's AiMOGA division. Yet analysts say 50-70% of this year's humanoids may end up idle in 'data factories.' Where does Robotera's STAR1 chip sourcing from NVIDIA and Intel leave it exposed?

NotableAI Compute · Korea Times

KT puts Korean AI chips, language model into one enterprise server - The Korea Times

Rebellions

KT launched an enterprise NPU LLM Station combining its own Mi:dm K 2.5 Pro language model with a Korean-made AI chip in one appliance. Rebellions' ATOM-MAX NPU sits at the hardware core of this system, tying KT's rollout to its silicon and memory supply lines. The brief traces Rebellions' dependence on Samsung and SK hynix behind that chip.

NotableAI Compute · Business Standard

Cerebras launches new server chip, system designed to speed AI chatbots

Cerebras Systems

Cerebras unveiled its CS-4 server system, packing three WSE-3 Turbo chips into one rack to accelerate AI chatbot inference. The design uses 50% fewer components than its predecessor, a shift meant to ease assembly. Trace how this changes Cerebras' reliance on TSMC's 5nm wafers and contract manufacturer Flex.

NotableAI Compute · Bloomberg

Cerebras Says Its New Computer Boosts AI Speed Advantage Over Nvidia

Cerebras Systems

Cerebras claims its newest system widens the speed gap over Nvidia's AI chips. That bravado rests on a fragile chain: contract manufacturers scaling 10x, a foundry partner exposed to cross-strait risk, and a GPU supplier that's also its rival. Can Flex and Taiwan Manufacturing actually keep pace with Cerebras's ambitions?

Tuesday, August 18

High impactAI Compute · Reuters

Einride forecasts faster growth, adds 500 Tesla Semi trucks to fleet

Tesla, Inc.EinrideCG Semi

Einride is tripling its fleet to roughly 750 trucks, adding 500 Tesla Semis alongside its existing DAF lineup. The Swedish freight tech firm says third-party funding will cover the expansion, not its own balance sheet. Our graph tracks how this reshapes Einride's supplier base beyond VDL and BYD, and what it means for its Amazon customer concentration.

High impactPhysical AI · TechCrunch

Einride strikes deal to add 500 Tesla Semis to its fleet

EinrideTesla, Inc.

Einride, the Swedish autonomous trucking firm, has agreed to add 500 Tesla Semis to its fleet over the next two years. The purchase would triple Einride's current lineup and lean heavily on Tesla's newly ramped Nevada production line. Watch how Tesla's 4680 battery cell output shapes whether this delivery timeline actually holds.

High impactAI Compute · Yahoo

Advanced Micro Devices (AMD) Unveils Instinct Coder For Private Enterprise AI

Advanced Micro Devices

AMD launched Instinct Coder, a turnkey enterprise AI inference stack built with MI series accelerators, marking its push beyond chips into full-stack platforms. The move ties AMD closer to partners handling integration and deployment. What does bundling this offering with Supermicro really expose AMD to downstream?

High impactAI Compute · 247Wallst

Nvidia Is Cutting Rubin Ultra’s Memory. Is Micron’s HBM Boom in Danger?

Keysight TechnologiesMicron TechnologyKeysightNVIDIA CorporationSK hynix Inc.

Nvidia is reportedly testing Rubin Ultra configs as low as 192GB of HBM, a steep drop from the originally planned 1TB, citing supply constraints and qualification delays. Watch Micron's HBM4E timeline and its 2027 ramp-up target to see whether this despec is temporary or a lasting demand hit.

High impactAI Compute · EE Times

Marvell Targets AI Bottlenecks with Memory-Disaggregation Portfolio

Marvell Technology

Marvell just unveiled a new AI memory portfolio, including the Bravera SC6 SSD controller and Structera CXL memory expanders, aimed at hyperscaler bottlenecks. The move deepens ties across its foundry and customer web, from TSMC to Meta. Track how Structera's compression multiplier reshapes Meta's DDR4 recycling strategy at scale.

High impactPhysical AI · Reuters

China's Lumos eyes fresh funding, potential listing as factory robot sales grow

Mitsubishi ElectricD-Robotics

Chinese factory robot maker Lumos Robotics is raising fresh capital and weighing an IPO as industrial deployments accelerate. Mitsubishi Electric quietly backs the startup as its largest outside shareholder. With Mitsubishi ramping its own humanoid production at Kyoto to 1,000 units monthly by 2027, what does its Lumos stake signal about which robot form factor actually wins in factories?

NotableAI Compute · Korea Times

Samsung Display to expand capacity to meet growing foldable phone demand - The Korea Times

Apple

Samsung Display is expanding foldable OLED capacity in Asan to keep pace with surging demand. But its new A7 plant won't finish until 2028, well after the first foldable iPhone is expected to launch. With Apple reportedly locked into an exclusive multi-year supply deal, what happens to iPhone foldable volume in the gap?

NotableAI Compute · Indiatimes

India semiconductor startup ecosystem: Cadence AI cuts chip timelines

Cadence Design Systems

India's chip startups are riding Cadence's AI push, which is squeezing design timelines dramatically. Cadence's own tools now turn five week engineering tasks into single-week sprints, reshaping how fast customers can ship silicon. Watch what ChipStack's roughly 25 production-stage customer engagements mean for the capital cycle across Cadence's entire chipmaker roster.

NotablePhysical AI · Seeking Alpha

Fabrinet anticipates $1.375B-$1.425B Q1 FY2027 revenue while outlining $12.5B-$14B capacity plan (NYSE:FN) | Seeking Alpha

Fabrinet

Fabrinet guided Q1 FY2027 revenue to $1.375B-$1.425B alongside a $12.5B-$14B capacity buildout, implying a huge jump from FY2025's $3.42B. But management also flagged uncertainty over potential bans on Chinese transceivers. Could that threat to optical component supply unravel Fabrinet's aggressive NVIDIA Blackwell production trajectory?

NotablePhysical AI · Chinadaily

Unitree Robotics to debut on STAR market, focusing on humanoid and quadruped robots

Unitree Robotics

Unitree Robotics debuted on Shanghai's STAR Market, becoming the first humanoid robotics firm listed on a mainland Chinese exchange after delivering over 5,500 humanoid units in 2025. Shares have since slumped 45% from their debut peak. Which H1 subsystem supplier shows up twice in Unitree's sourcing list, hinting at a concentration risk investors haven't priced in?

NotableAI Compute · Motley Fool

Why This Memory Chip Boom May Have More Staying Power Than History Suggests | The Motley Fool

Micron TechnologySK hynix Inc.ChangXin Memory Technologies

Analysts argue this DRAM upcycle could outlast history's boom-bust pattern, with AI demand straining supply. Micron and SK hynix are already posting triple-digit revenue growth and record margins. Track how HBM's 30% capacity draw for just 13% of memory bits is quietly squeezing conventional server DRAM output through 2027.

Monday, August 17

High impactPhysical AI · Reuters

China's Unitree unveils 'Superman' robot as fervour builds ahead of Shanghai debut

HumanoidDeepSeekThomson

Unitree's 'Superman' robot debuts in Shanghai as hype builds toward its high-profile reveal. Behind the scenes, an AI compute firm has already staked millions on Unitree's IPO and agreed to co-develop its robot brains. What does DeepSeek's stake in Unitree actually mean for who controls the humanoid robotics supply chain?

High impactPhysical AI · Reuters

Serve Robotics partners with Grubhub in robot delivery expansion push

ThomsonServe Robotics

Serve Robotics is rolling out sidewalk robots for Grubhub across Chicago, Los Angeles, and Alexandria while adding DoorDash coverage in San Jose and Washington, D.C. But this expansion follows a bigger shakeup on the platform side. Watch what Uber Eats' exit does to Serve's revenue concentration and how Ouster and Magna factor into its scaling plan.

High impactSpace · Arstechnica

Satellite operators are in panic mode due to a worsening launch crisis

SpaceXUnited Launch AllianceAmazon Leo

Satellite operators are scrambling as launch slots dry up, with commercial customers reporting SpaceX won't book missions past 2028. Falcon 9 flights already peaked in 2025 and are set to decline in 2026. Watch what happens to Amazon Leo's stalled satellite production once booked launches never materialize.

High impactPhysical AI · EE Times

Nvidia Bets on the Classical Side of Quantum Computing

FoundationNVIDIA Corporation

Nvidia says it won't build its own quantum chips, choosing instead to supply the classical backbone that makes quantum machines run. A hybrid deployment is already live at Barcelona's MareNostrum 5. What does Nvidia's GPU integration into that facility's three quantum computers reveal about its next demand pipeline?

High impactSpace · Breakingdefense

Space Force brings 5 companies on-board Space Data Network

SpaceXFoundationYork Space SystemsRocket LabAmazon LeoLockheed Martin Corp.

Space Force just added five companies to its Space Data Network, the backbone for military satellite communications. Rocket Lab, Lockheed Martin, and SpaceX all show up as both buyers and suppliers across this build. What does SpaceX's $2.9 billion MILNET backbone contract actually depend on further down its supply chain?

High impactSpace · GlobeNewswire

Viasat Selects Rocket Lab to Build GEO Satellite for U.S. Space Force’s Protected Tactical SATCOM-Global Program

ViasatRocket Lab

Viasat tapped Rocket Lab to build a GEO satellite bus for the Space Force's Protected Tactical SATCOM-Global Swarm 1 program, pairing Viasat's X/Ka-band payload with Rocket Lab's Lightning platform. Trace how this bus contract fits against Rocket Lab's pending Iridium acquisition and its shift from launch vendor to full-stack spacecraft supplier.

NotableAI Compute · Techtimes

China Bought It: NBS Confirms 81.8% AI Equipment Surge as Broader Investment Falls

Alibaba GroupBaiduHuawei Technologies Co., Ltd.NVIDIA Corporation

China's statistics bureau confirmed internet-sector equipment purchases jumped 81.8% year-on-year through July 2026, turning vague AI capex talk into hard procurement numbers. Tencent alone posted its first negative free cash flow since 2005. So why did Wall Street slash Baidu's earnings estimates by 18% right as its Q2 capex surged?

NotableAI Compute · Yahoo Finance

The Global Market for Computing and AI for Data Centers Report 2026-2040: Evaluating GPUs, Custom ASICs, Arm vs x86, and Strategies of Nvidia, AMD, Google, and AWS

NVIDIA CorporationMicrosoft CorporationAlphabetMeta PlatformsIntel CorporationSamsung Electronics Co., Ltd.

A new report maps the data-center AI chip race through 2040, from Nvidia's GPU dominance to the rise of custom ASICs at Google, Microsoft, and Meta. Microsoft's Maia 300 is targeting capacity beyond 300,000 units for 2027 delivery, a number that says a lot about where hyperscaler chip spending is headed.

NotablePhysical AI · Restofworld

Ban on Chinese robots leaves U.S. startups stranded

Unitree RoboticsSZ DJI Technology Co., Ltd.

A new FCC ban on Chinese robots is squeezing U.S. startups that built their prototyping pipelines on Chinese suppliers. Firms like Persona AI now scramble for non-Chinese parts amid claims components are too costly or unavailable domestically. Trace which specific component, from Unitree's Hesai lidar to Sanhua's hydraulics, has no real American substitute.

NotableSpace · Sciencepost

US Space Force aims to develop 'flat satellites' for low-Earth orbit deployment

Rocket Lab

The Space Force wants stackable 'flat satellites' for LEO deployment, and Rocket Lab just landed the job to build them. Track how the $397M SB-AMTI contract layers onto Rocket Lab's existing $266M Alaska launch deal and pending Iridium acquisition to reshape its supplier and customer concentration.

ContextPhysical AI · Technode

Robot companies are becoming AI companies as AgiBot reveals the new logic of embodied AI competition

AgiBot

AgiBot is racing toward a Hong Kong IPO valued at HK$40-50 billion, rolling out its A3 robot and Genie G2 humanoid while chasing tens-of-millions-of-hours of training data. But as the company pivots from hardware to AI models, what happens to suppliers like Ningbo Yunsheng and Arm once chips and magnets stop being the differentiator?

Week of August 10 81 stories ▾

Sunday, August 16

High impactPhysical AI · Taiwannews

Humanoid robots to headline Taipei automation exhibition | Taiwan News | Aug. 16, 2026 16:09

Unitree Robotics

Humanoid robots take center stage at Taipei's automation show this week, with Taiwan makers like HIWIN and Delta showcasing rivals to China's fast-scaling players. Trace Unitree's component supply chain, from NVIDIA chips to LG batteries, to see exactly where Taiwan's leverage over its newly IPO-funded competitor actually sits.

High impactAI Compute · Khan

China's offensive intensifies in the NAND market as well... YMTC surpasses Kioxia to take third place by shipments

Yangtze Memory TechnologiesMicron TechnologyKioxiaSolidigmSamsung Electronics Co., Ltd.SK hynix Inc.

YMTC just leapfrogged Kioxia into third place in global NAND shipments, its first time ever ranking above the Japanese memory maker. Kioxia's stock has been swinging wildly on AI financing jitters. What does YMTC's new placement inside Lenovo's ThinkBook 14 G9 lineup signal about its next OEM target?

High impactAI Compute · Thestreet

Apple's cheap-chip plan just hit a wall in Washington

Apple

Apple's push to use cheap Chinese DRAM to dodge a historic price spike is running into Washington opposition. DRAM contract prices reportedly jumped roughly 90 to 95 percent quarter over quarter in early 2026, and Apple already raised prices $100 to $300 on several products. CXMT sits on a federal list that blocks Apple from customizing the chips it buys.

High impactAI Compute · Google

Big Tech’s data centre boom poised to drive up carbon emissions

Meta Platforms

Meta's $1 billion Beaver Dam, Wisconsin data centre is one piece of a Big Tech buildout that could push emissions higher as gas and coal plants stay online to feed the grid. Track how Meta's 14-gigawatt 2027 capacity target reshapes its Taiwan Manufacturing foundry dependence once power, not chips, becomes the bottleneck.

High impactPhysical AI · Supplychaintoday

Unitree As2-W: "Super Athlete" Robot Shows How Fast the Future of Automation Is Arriving. -

Unitree Robotics

Unitree's new AS2-W robot can hit 6 m/s, climb 45-degree slopes, and clear 80 cm obstacles, showing how fast quadruped automation is scaling toward real-world logistics work. Fresh off a record-breaking Shanghai IPO, Unitree is ramping production fast. Which single supplier's LiDAR capacity could bottleneck the entire rollout?

High impactAI Compute · Tom's Hardware

Google reportedly taps AMD to design next-generation TPU — hybrid AI ASIC could integrate on-package CPU cores for reinforcement learning

AlphabetAdvanced Micro Devices

Google is reportedly bringing in AMD to design its next-gen TPU, a hybrid AI ASIC that could pack on-package CPU cores for reinforcement learning workloads. Track what this does to Alphabet's existing Samsung and Broadcom design ties as AMD's EPYC-proven supply chain enters the TPU picture.

ContextPhysical AI · Marketbeat

Ouster Says Rev8 Lidar Gains Early Multimillion-Dollar Orders, Targets Broader Platform Growth

Ouster, Inc.

Ouster's new Rev8 lidar platform is already pulling in multimillion-dollar orders as the company eyes broader platform growth. Watch what happens to Rev7 production, still embedded in customer programs, as Ouster runs dual manufacturing lines through suppliers like Benchmark and Fabrinet during this transition.

Saturday, August 15

High impactAI Compute · Hindustantimes

Three plants shipping, nine more on the way: Unpacking PM Modi's chips-&-minerals pledge in Independence Day speech | Hindustan Times

CG SemiSTARS MicroelectronicsMicron TechnologyQualcommTata ElectronicsKaynes Semicon

Modi touted three chip plants shipping and nine more in the pipeline during his Independence Day address. But the assembly plants shipping today aren't the same as the fab India actually needs. What's happening to Tata's Dholera fabrication timeline, and does it still hold December 2026?

High impactAI Compute · Technews

TSMC, Silicon Motion, Foxconn Supply Chain Support: NVIDIA Announces Spectrum-X Ethernet Silicon Photonics Switch Mass Production

NVIDIA CorporationTaiwan Semiconductor Manufacturing Company LimitedLumentumHon Hai Precision Industry Co., Ltd.

NVIDIA has moved its Spectrum-X Ethernet switch into mass production, embedding silicon photonics directly onto the switching chip with TSMC, Foxconn, and Silicon Motion in the supply chain. What does shifting Lumentum from bulk transceiver sales into a lower-volume laser-source role do to its margins?

High impactAI Compute · Techtimes

Co-Packaged Optics Standard Gets Q4 Deadline: OCP Blueprint Targets Wasted AI Compute

NVIDIA CorporationMeta PlatformsTaiwan Semiconductor Manufacturing Company Limited

OCP just published a 300-page blueprint pushing co-packaged optics toward a formal Q4 2026 spec, aiming to fix wasted AI compute from copper interconnect limits. Watch what happens to Nvidia's Spectrum-6 single-vendor grip once TSMC-fabbed rivals get a shared standard to build against.

High impactAI Compute · Memeburn

Meta and Nvidia plant a flag in China-led open-weight AI race - Memeburn

Meta PlatformsNVIDIA Corporation

Meta and Nvidia are pushing open-weight AI models to counter China's Moonshot AI, Z.ai, Alibaba, and DeepSeek in the developer race. Meta's new 30-billion-parameter model runs on a single consumer GPU. But does this edge push actually shrink Meta's need for its own 14 gigawatts of 2027 compute capacity?

High impactAI Compute · 247Wallst

Nvidia Is Looking to Own Another Layer of the AI Ecosystem

Advanced Micro DevicesOracle CorporationLumentumNVIDIA Corporation

Nvidia's push into networking hardware is pulling its supply chain even tighter, with co-packaged optics now soldered directly onto Spectrum-X switches. CoreWeave, Lambda, and Oracle are already running the platform at full production volume. Lumentum's role in this buildout carries a revenue figure worth checking.

High impactPhysical AI · Supplychaintoday

Robot Fight Clubs May Be Building the Humanoids That Transform Supply Chain. -

EngineAIAgility RoboticsFigure AI Inc.Apptronik

Combat-style robotics contests are quietly becoming proving grounds for humanoid machines headed straight for warehouse floors. Firms like Figure AI and EngineAI are already courting factory and logistics deals. Trace who actually supplies these fighters' brains and batteries, and watch what a Chinese humanoid FCC ban does to EngineAI's NVIDIA-dependent chip pipeline.

NotableAI Compute · Investing.com

US presses Apple to avoid Chinese memory chips amid shortage - WSJ reports By Investing.com

Apple

Washington is reportedly pushing Apple to steer clear of Chinese memory suppliers as an AI-driven chip shortage bites into iPhone and Mac production. Apple's own supply map shows just one mapped memory source covering this critical component. Can that single supplier actually absorb the demand Apple now needs?

NotableAI Compute · Khan

'Memory shortage' prompts Apple to consider Chinese-made memory···Trump administration "We oppose it, find another solution

Apple

Apple is weighing Chinese-made memory to plug a worsening shortage, but Washington says no and Micron is pushing back hard. Watch the loophole around off-the-shelf commodity memory purchases from CXMT or YMTC, since it sidesteps the approval gate that blocks customized chip specs entirely.

NotableAI Compute · Yahoo

COHR Q2 Deep Dive: Data Center Demand and Capacity Expansion Define Outlook

Coherent

Coherent's Q2 revenue jumped 33.7% year-on-year, beating estimates as data-center and communications demand surged alongside a major capacity expansion. Track how the 6-inch Indium Phosphide line buildout ties directly into Coherent's optical module commitments to NVIDIA's Blackwell architecture and what that means for supply continuity risk.

NotableAI Compute · Wccftech

AMD & ARM Have Eaten Intel’s Server Lead, Yet x86 Is Poised To Own The AI CPU Wave Through 2030 As TAM Crosses $200b

Advanced Micro DevicesIntel Corporation

Server CPU spending is set to top $200 billion by 2030, with AI chips driving most of that growth as x86 fends off ARM's advance. AMD and Intel are both racing to fund next-gen capacity, but who actually controls the foundry and packaging chokepoints behind EPYC's scale-up?

NotableAI Compute · Siliconangle

Chipmaking equipment supplier Lam Research to spend $3B+ on new labs

SK hynix Inc.

Lam Research plans to pour over $3B into new engineering labs over five years, aiming to boost its experiment capacity by more than 50%. The equipment maker supplies the world's three largest memory chipmakers. Does that lab buildout line up with SK hynix's Yongin Y2 fab timeline for HBM production?

NotableAI Compute · Dongascience

NVIDIA's SONIC trains humanoid G1 to use hands and feet for trash disposal

NVIDIA CorporationUnitree Robotics

NVIDIA's SONIC framework taught Unitree's G1 humanoid to grab trash by hand and pop a pedal-operated bin lid with its foot in one balanced motion. Follow the chip pipeline behind the demo: NVIDIA already supplies Unitree, Atlas, and 1X's NEO, while G1's other parts trace back to suppliers like RoboSense and Sanhua.

Friday, August 14

High impactSpace · Reuters

Chinese humanoid robot maker Unitree powers up for stellar Shanghai debut

SpaceXHumanoidDeepSeekThomsonBoston DynamicsDobot

Unitree's Shanghai debut is drawing headlines as the humanoid robot maker gears up for a blockbuster IPO. But behind the hype sits a tighter deal: DeepSeek didn't just invest, it committed to co-develop Unitree's AI models. Does that partnership give Unitree an edge over rivals like Boston Dynamics?

High impactSpace · Arstechnica

Rocket Report: Rocket Lab shows off its flexibility; Blue Origin’s two-pad plan

Rocket Factory AugsburgArianeGroupRocket Lab

Rocket Lab keeps flexing its launch options while Blue Origin advances a two-pad strategy for its own cadence. The gated brief traces how a portable spaceport concept and a pending Iridium acquisition are reshaping Rocket Lab's role in the satellite supply chain. Neutron's Stage 1 tank timeline is already narrowing toward late 2026.

High impactAI Compute · EE Times

Intel at a Memory Crossroads, Again - EE Times

SK hynix Inc.

Intel's CEO is talking up embedded DRAM stacking and memory innovation beyond commodity chips. He just hired away SK Hynix's former CEO too. Watch what that talent move signals for SK Hynix's exclusive 12H HBM3E grip on NVIDIA's GB300 and its Yongin Y2 expansion timeline.

High impactPhysical AI · Datamintelligence

Trump Drone Tariffs 2026: 100% Duty on Chinese UAV Imports Reshapes Global Drone Market

SZ DJI Technology Co., Ltd.Autel RoboticsSkydio

New 100% tariffs on Chinese drones over 25kg hit just as DJI faces FCC bans and a projected $1.5B loss. But even DJI's flagship Matrice line leans entirely on non-Chinese chipmakers for IMUs and controllers. Does that Western sourcing shield DJI from the tariff hit, or does it seal the trap?

High impactAI Compute · Wccftech

Samsung May Push HBM Base Dies to Cutting-Edge 2nm, Extending Its Own 4nm HBM4 Roadmap

Samsung Electronics Co., Ltd.NVIDIA CorporationSK hynix Inc.Micron Technology

Samsung is reportedly eyeing 2nm production for HBM base dies at a repurposed Giheung line, layered on top of its existing 4nm HBM4 plans. The move would pull base-die work in-house instead of relying on TSMC. The targeted facility, NRD-K Line 2, has an opening timeline worth checking against SK hynix and Micron's own capacity schedules.

High impactPhysical AI · Torquenews

Tesla and SpaceX Terafab Has Officially Started Construction , And Here Is The First Image

Tesla, Inc.SpaceX

Tesla and SpaceX's jointly funded Terafab has broken ground in Texas, with the first construction photo now public. Both companies currently lean on outside chipmakers for critical compute. Watch what the $16.8 billion facility signals about Tesla's Marvell and Qualcomm dependencies once captive silicon comes online.

NotableAI Compute · Wccftech

NVIDIA’s Optics Partners Crush CPO Delay Rumors, With Coherent’s Scale-Out Revenue Starting Late 2026

NVIDIA CorporationLumentumCoherentAdvanced Micro Devices

Coherent and Lumentum are pushing back on reports that NVIDIA's co-packaged optics rollout is slipping, pointing to scale-out revenue landing later this year. Both suppliers just backed up the timeline in their latest earnings calls. So what does NVIDIA's $2 billion bet on each partner actually buy in return, and when does it pay off?

NotableAI Compute · Google

LG joins forces with Nvidia to build AI-powered robots and hardware for self-driving cars - GSMArena.com news

LG ElectronicsNVIDIA Corporation

LG and Nvidia signed an MOU spanning humanoid robots, factory robotics, and a future data center buildout in South Korea. Track how deep the Jetson Thor and Isaac GR00T commitments run through LG's manufacturing and AI plans, especially the 80 MW Cheonan facility timeline.

NotableAI Compute · Webpronews

Microsoft Scales Back Cloud and AI Operations in China Amid Tensions

Microsoft Corporation

Microsoft is pulling back its cloud and AI operations in China, shifting resources to Singapore and India as staff face reassignment or severance. The move comes even as its global backlog just hit $678 billion. Watch what happens to Meituan's AI compute pipeline once its Microsoft link starts to fray.

NotableSpace · Sciencedaily

Caltech breakthrough brings fiber-optic performance to silicon chips

EMCORECoherent

Caltech engineers have printed fiber-optic-grade light paths directly onto silicon wafers, mimicking fiber performance at chip scale. The advance lands squarely on top of a supply chain already straining to keep pace with AI data center demand. Could this silicon photonics leap dissolve the indium phosphide bottleneck now capping Coherent's laser output growth?

NotableAI Compute · Reuters

Chinese chipmaker SMIC increases prices on strong AI demand

SMIC

SMIC is hiking wafer prices after strong AI-driven demand, adding 8,000 wafers of monthly 12-inch capacity in Q2. With utilisation already at 93.7% and revenue 90% concentrated in China, what does tighter supply mean for customers like Huawei and Cambricon locked into this chokepoint?

ContextPhysical AI · Automation Mag

Automation International | News

MolexHumanoid

Molex has unveiled a hybrid connector packing 15 A power and 1 Gbps Ethernet into a 5.65 mm profile built for humanoid robot joints and AMRs. It looks like a routine product launch. Follow where Schaeffler's thousands-of-units robot deployment fits into Molex's connector roadmap before bills of material lock in.

Thursday, August 13

High impactAI Compute · Financial Times

TSMC and Sony team up, China's AI stocks swing

Taiwan Semiconductor Manufacturing Company Limited

TSMC and Sony are forming a joint venture to mass-produce next-gen CMOS image sensors in Kumamoto by 2029, while China's AI stocks whipsawed on the news. TSMC has never taken an equity stake in a sensor fab before. What does a 40/60 split with Sony actually lock in for Apple's camera roadmap and TSMC's captive customer base?

High impactSpace · SCMP

Chinese chip foundries surge as AI boom drives triple-digit profit growth

SpaceXDeepSeekLenovoCambriconNVIDIA CorporationIluvatar CoreX

SMIC and Hua Hong Grace posted triple-digit profit jumps as China's AI chip demand runs domestic foundries at full capacity. Cambricon's revenue surged 108% in H1, but the brief maps a supply bottleneck tied to its sole reliance on SMIC for fabrication.

High impactAI Compute · CNBC

An inside look at SK Hynix $720 billion AI-fueled buildout that's taking over South Korea

AlphabetMicron TechnologySolidigmNVIDIA CorporationSamsung Electronics Co., Ltd.SK hynix Inc.

SK Hynix is pouring $720 billion into new fabs across South Korea to feed the world's AI chip demand, funded partly by a record Nasdaq IPO. But the memory maker already controls 58% of the HBM market that Nvidia's Blackwell GPUs depend on. Track when Yongin Y2 and Cheongju M17 actually come online before calling this crisis solved.

High impactPhysical AI · Neura Robotics

NEURA Robotics acquires ACTIVE Shuttle from Bosch Rexroth

FoundationNEURA RoboticsBosch Rexroth

NEURA Robotics is acquiring the ACTIVE Shuttle AGV platform from Bosch Rexroth, folding its fleet software and navigation stack into NEURA's Physical AI ecosystem. What happens to ACTIVE Shuttle's existing industrial customer base once NEURA Mobile Robots GmbH takes over service and integration on October 1, 2026?

High impactAI Compute · X

SK Hynix, Samsung, and Micron have reportedly already sold out all of their 2027 DRAM and HBM memory capacity. $MU $NVDA $AMD

SK hynix Inc.Samsung Electronics Co., Ltd.Micron Technology

SK Hynix, Samsung, and Micron have reportedly sold out their entire 2027 DRAM and HBM output, tightening supply for AI chipmakers. Micron sits on both sides of this squeeze, buying HBM from SK Hynix while also supplying it to NVIDIA and Tesla. The brief maps how that dual role could reshuffle who gets allocated first.

High impactPhysical AI · X

Tesla and SpaceX just broke ground on Terafab. This land will soon hold 100M+ sq ft of the most advanced AI chip factory ever built. It will supply enough AI compute every year to

Tesla, Inc.SpaceX

Tesla and SpaceX broke ground on Terafab, a 100M+ sq ft chip plant promising enough AI compute to outpace global demand. Both firms already lean hard on outside partners for silicon and fabrication. Does Terafab finally end Tesla's single-source dependency on Taiwan Manufacturing for Optimus foundry work?

NotablePhysical AI · Techtimes

DEEP Robotics DR02 Conquers Outdoor Stairs as China's Humanoid IPO Race Accelerates

Deep RoboticsNVIDIA CorporationUnitree Robotics

DEEP Robotics showed its DR02 humanoid climbing outdoor stairs unassisted as it pursues a CNY 2.5 billion STAR Market IPO. Watch what powers that 500-1,000 Hz balance loop: a single Jetson AGX Orin chip from NVIDIA, with zero backup compute path in DEEP Robotics' supply chain.

NotablePhysical AI · Economictimes

Govt's Rs 7,280 crore rare earth magnet scheme draws 20 bids, including L&T, Coal India, ReNew

Neo Performance MagnetsProterial

India's rare earth magnet incentive scheme pulled in 20 bidders, from L&T and Coal India to ReNew, all chasing a slice of the Rs 7,280 crore pot. One bidder already has sourced supply edges to three named downstream customers in our graph, and a capacity expansion timeline that predates the scheme itself.

NotablePhysical AI · PR Newswire

LG to Unveil Its Next-Gen Humanoid Robot, Built on NVIDIA Isaac GR00T

HumanoidFoundationLG Energy SolutionNVIDIA CorporationLG InnotekLG Electronics

LG is set to reveal its next humanoid robot powered by NVIDIA's Isaac GR00T platform, positioning the conglomerate deeper into the physical AI race. But LG's own units already sit inside rival robots' supply chains. Trace how LG Energy Solution's battery ties to Optimus, Figure 03, and Atlas complicate this launch.

NotableSpace · Lightreading

Lockheed Martin preps drone detection service using Verizon 5G

Lockheed Martin Corp.NVIDIA CorporationKeysight Technologies

Lockheed Martin is building a NetSense drone-detection service on Verizon's 5G network, aiming for pilot runs in late 2026. The move leans on outside chipmaking and RF testing partners to make it work across any network. Which supplier's RF simulation tools quietly become a chokepoint for this dual-use system?

NotablePhysical AI · Spglobal

CATL’s 350 Wh/kg prismatic battery achieves aviation-grade safety standard

CATL

CATL's 350 Wh/kg prismatic battery just cleared aviation-grade safety testing in China, clearing the runway for mass production. The cells are slated for Autoflight's manned eVTOL aircraft, a segment CATL hasn't served before. With Tesla, Xiaomi, and Stellantis already on its books, how does an eVTOL debut fit CATL's existing customer concentration?

NotableAI Compute · Tom's Hardware

Memory maker CXMT overtakes Tencent to become China's most valuable company 17 days after its IPO — now worth $524 billion

CXMT

CXMT's Shanghai debut sent it past Tencent to become China's most valuable company at $524 billion, just 17 days after listing. See which rival memory maker, SK hynix, shows up in our graph as a buyer of CXMT's HBM despite being a direct competitor.

High impactAI Compute · X

HBF TIME! SanDisk $SNDK says its HBF architecture can deliver HBM-like read bandwidth with 8–16x more capacity, targeting increasingly memory-heavy AI inference workloads. The comp

SanDisk

SanDisk unveiled HBF, a memory architecture promising HBM-like read speeds at 8-16x the capacity for AI inference. The launch comes days after a competitor's IPO sent SNDK shares tumbling despite a huge 2026 rally. Could write endurance limits undercut HBF's bid to defend SanDisk's margins with NVIDIA and Microsoft?

ContextPhysical AI · Interestingengineering

BYD's new humanoid robot with 31 Degree of Freedom enters showrooms

BYDHumanoid

BYD's Xiao Di humanoid, 1.61 meters tall with 31 degrees of freedom, is now greeting customers at Di Space centers in Zhengzhou, Shenzhen, and Shanghai. The robot's bill-of-materials draws on suppliers already inside BYD's stack, including Ningbo Tuopu for actuators and Hesai for vision.

Wednesday, August 12

High impactAI Compute · Mt

HBM Gaps Filled by CXL... Memory Capacity Wars Intensify - MoneyToday

Micron TechnologySamsung Electronics Co., Ltd.SK hynix Inc.

Samsung, SK hynix, and Micron are all racing toward CXL memory to ease HBM bottlenecks squeezing AI compute. The pivot looks coordinated, not competitive, across all three suppliers. Samsung's own research claims CXL-based DRAM can cut token generation latency by up to 16x versus GPU VRAM-only systems.

High impactAI Compute · Livetradingnews

Nvidia's Nemotron 4: The Trillion-Parameter Open Model That Tightens the AI Power Map

NVIDIA CorporationMeta Platforms

Nvidia's Nemotron 4 pushes a trillion-parameter open model into an AI stack already reshaping enterprise compute. But Nvidia's own supply chain runs through a narrow set of chokepoints, from HBM sourcing to a single foundry jurisdiction. What does the Nemotron Coalition's compute-sharing structure mean for buyers like Meta who are quietly building their own silicon?

High impactAI Compute · Google

Oracle Stock Climbs After Quantum Partnership, Layoffs Report

AlphabetOracle CorporationMicrosoft CorporationCoherentNVIDIA CorporationSanDisk

Oracle shares jumped over 5% to $153.28 after news of a quantum computing partnership and a layoff report both hit the same day. Trace how deep Oracle's new dependency on Quantinuum's Helios hardware runs before betting on the quantum-AI narrative holding up.

High impactPhysical AI · Aibusiness

How US Ban on Chinese Humanoid Robots Could Affect the Industry

AgiBot

The FCC's new ban on Chinese-made humanoid robots lands just as one of China's top manufacturers pushes toward a major market debut. Watch how AgiBot's Hong Kong IPO timeline collides with a blocked U.S. pathway and a supply chain running through names like Ningbo Yunsheng and RoboSense.

High impactAI Compute · Chosun

TSMC Invests $29.4 Billion to Expand AI Semiconductor Production

Taiwan Semiconductor Manufacturing Company LimitedSony Group

TSMC is pouring $29.4 billion into AI chip capacity, including a new Sony joint venture in Kumamoto set to start mass production in 2029. But the real capacity crunch may not be wafer fabs at all. What happens when TSMC's CoWoS packaging roadmap becomes the actual bottleneck for NVIDIA and AMD?

High impactAI Compute · Chosun

ChangXin Memory Targets Mobile DRAM, Challenging Samsung and SK Hynix

ChangXin Memory TechnologiesCXMT

Apple is reportedly testing CXMT DRAM for iPhones and MacBooks, opening early supply talks with the Chinese memory maker. This would push CXMT beyond its PC-commodity base into mobile DRAM long dominated by Samsung and SK Hynix. What does CXMT's new LPDDR6 development, rated up to 12.8 Gbps, signal about its readiness for that leap?

NotableAI Compute · Wccftech

TSMC Exec Makes Big Announcement & Confirms 98 % Yield For CoWoS Packaged AI Chips

Taiwan Semiconductor Manufacturing Company Limited

TSMC says it's hitting 98-99% yield on AI chips packaged through its CoWoS 5.5x process. That's a big claim for such a complex assembly step. Watch what this means for TSMC's roadmap to 14x packaging by 2029 and the capex bet riding on it.

ContextAI Compute · Yahoo

Did ACMR's Raised 2026 Revenue Outlook and Strong Orders Just Shift Its Investment Narrative?

ACM Research

ACM Research raised its 2026 revenue outlook to $1.13-$1.18 billion after orders jumped 105% in the first half of the year. But three of its six major customers are China-based chipmakers facing export scrutiny. Does that order surge hold up if Washington tightens tool restrictions further?

ContextAI Compute · EE Times

Meta Recycles DDR4 Memory: Can Others Follow?

Meta PlatformsMarvell Technology

Meta is stretching aging DDR4 memory using CXL expansion tech, cutting server counts 25% across millions of machines instead of buying new DRAM. This reuse trend threatens volume for DRAM suppliers and server vendors alike. Marvell's Structera X memory expansion line sits directly in the path of this shifting demand.

ContextAI Compute · Trendforce

[News] Micron Flags HBM4E as Start of Custom HBM Era; HBM Market Could Tilt Toward Dual-or-Single Suppliers

Micron TechnologySK hynix Inc.

Micron says HBM4E marks the start of custom HBM, warning the market could shift from today's three-supplier split toward dual- or single-source deals. It's also moving from 1-beta to 1-gamma DRAM for the new standard. Watch the DDR-to-HBM bit tradeoff ratio Micron just disclosed for the real capacity squeeze.

ContextPhysical AI · Google

Quantinuum and Oracle are partnering to bring quantum computing to the cloud

Oracle Corporation

Oracle will host Quantinuum's Helios quantum computer inside its U.S. AI data centers, folding quantum access into OCI's cloud stack within months. It's a fresh capability layer for Oracle's cloud, not just a reseller deal. What does a 98-qubit trapped-ion system do to Oracle's existing physical AI supply footprint?

ContextPhysical AI · N Tv

From Scrap Parts to Empire: The Success Secret Behind China's New Robot Giant Unitree

Unitree RoboticsHumanoid

Unitree's G1 humanoid stands 1.32 meters, weighs 35 kilograms, and sells for roughly 13,500 dollars, built on the founder's scrap-to-empire cost philosophy. Track the single-source dependency on NVIDIA chips powering its AI inference, a chokepoint that could matter far more than the robot's price tag.

Tuesday, August 11

High impactPhysical AI · EE Times

AMD’s Ryzen AI X100 Takes On GPU-Centric AI

HumanoidSiemens EDANVIDIA CorporationSiemens

AMD is challenging Nvidia's GPU dominance with the Ryzen AI X100, pitching deterministic latency for robotics brains over raw parallel compute. But Nvidia's Jetson Thor already threads through Figure AI, LimX Dynamics, and a fresh LG humanoid partnership. Trace who else, like Siemens and Humanoid, is quietly wired into that Jetson-fed supply loop before AMD's pitch lands.

High impactAI Compute · Fortune

AI's biggest buildout is here. These stocks offer a way to invest in the data center boom | Fortune

Marvell Technology

Fortune flags data center stocks riding the AI buildout, with Marvell Technology named as a networking infrastructure play. But Marvell's revenue leans heavily on custom ASIC deals with a handful of hyperscalers. How concentrated is Marvell's exposure to its single largest customer relationship?

High impactSpace · TechCrunch

Uber surprised robotics company Serve by selling its entire stake

SpaceXHumanoidMind RoboticsServe Robotics

Uber quietly offloaded its entire stake in Serve Robotics, catching the delivery-bot maker off guard after a five-year partnership. The sale ends a 17-quarter growth streak through Uber's app and puts fresh pressure on Serve's supplier ties to Magna and Ouster as it scrambles to prove its 2,000-robot pipeline can survive without its biggest customer.

High impactAI Compute · Semiengineering

HBM Becomes Testbed For 3D Assembly Yield

SynopsysSiemens EDAAdvantestSiemensSK hynix Inc.

SK hynix, Samsung, and Micron are turning HBM into the proving ground for 3D chip assembly, where defect detection at each layer decides whether AI hardware ships on time. Follow the chain from SK hynix's HBM buyers to the test vendors racing to catch flaws before HBM5's 24-layer stacks hit production.

High impactAI Compute · Nikkei Asia

TSMC takes rare step of teaming with Sony as Samsung, China rivals loom

Sony GroupTaiwan Semiconductor Manufacturing Company LimitedSony Semiconductor Solutions

Sony and TSMC are pouring $6.3bn into a new Kumamoto image sensor plant to fend off Samsung and Chinese rivals. It's a big shift from customer to co-owner. What does Sony's 60% stake in this joint venture actually mean for TSMC's foundry capacity allocation?

High impactPhysical AI · Google

China humanoid robot makers hold 97% of shipments in H1, report says

Tesla, Inc.Figure AI Inc.Unitree RoboticsDeep RoboticsLeju Robotics

Chinese manufacturers reportedly controlled 97% of humanoid robot shipments in H1, widening the gap on US rivals Tesla and Figure AI. Track how deep this split runs beneath the surface: watch Unitree's supplier list, from RoboSense to Hesai to Leaderdrive, for signs of which dependencies actually flow only one direction.

NotableAI Compute · Taipeitimes

SPIL breaks ground in Yunlin - Taipei Times

SPIL

SPIL has broken ground on a NT$100 billion CoWoS packaging plant in Yunlin, aiming for first output by 2028. The site scales advanced chip-on-wafer-on-substrate capacity central to AI accelerator supply. Trace whether this buildout is tied to AMD's EPYC roadmap or a broader bet on rival GPU makers.

NotablePhysical AI · Nvidia

NVIDIA and Local AI Community Fuel Open Source Models and Intelligent Agents

FoundationDeepSeekAppleNVIDIA Corporation

NVIDIA is pushing free, optimized open models like DeepSeek-V4-Flash and Cosmos 3 Edge onto DGX Spark, Jetson, and RTX hardware, deepening developer lock-in beyond chip sales alone. But with a claimed 26x speedup over Apple's M3 Ultra circulating, does Apple's custom silicon strategy leave it any foothold in local AI at all?

NotableAI Compute · Google

Exclusive: SK hynix to Increase NAND Production in China by 50% SK hynix is resuming construction of its second NAND flash plant in Dalian, China, expanding its local production ca

SK hynix Inc.

SK hynix is restarting construction on its second Dalian NAND plant, a project frozen for four years by the memory downturn and US export curbs. Combined with Fab 1, local China output climbs from 100,000 to 150,000 wafers monthly. Cheongju M17's role in this bifurcated capacity strategy sits inside the brief.

NotablePhysical AI · Financial Times

Chinese robot maker’s IPO 5,500 times oversubscribed by retail investors

Unitree RoboticsHumanoid

Unitree's IPO drew 5,500 times retail demand, valuing the humanoid robot maker at $9bn after raising Rmb6.1bn. Trace the NVIDIA chip dependency sitting inside Unitree's G1 flagship before betting on China's newest robotics champion.

Monday, August 10

High impactSpace · Fortune

Musk’s planned $16.8 billion factory is five times bigger than the world’s current largest building

SpaceXTesla, Inc.

Elon Musk's Terafab project would dwarf every building on Earth, backed by $16.8 billion and 100 million square feet of factory space. Tesla and SpaceX already lean on outside chipmakers for everything from Optimus memory to AI servers. Could this facility finally cut their dependence on suppliers like Micron?

High impactAI Compute · Supplychaindive

FedEx, Amazon pursue expanded use of robotic arms

Amazon.com, Inc.Dexterity Inc.Boston Dynamics

FedEx and Amazon are both scaling robotic arms, with FedEx leaning on Dexterity's Mech loaders at Hagerstown Hub and Amazon doubling its fleet through 2026. Follow the roller screw supply chain behind Dexterity's single-source deal with FedEx before betting on a smooth rollout across 20+ hubs.

High impactAI Compute · Business Standard

Apple tests China's CXMT chips for iPhones, MacBooks amid supply crunch

CXMT

Apple is reportedly testing memory chips from China's CXMT for iPhones and MacBooks as global supply tightens, with early talk of China-market devices first. HP and Acer are already shipping CXMT chips overseas. See which other Apple product lines and OEM customers already sit on CXMT's supplier list.

High impactSpace · Manufacturingdive

SpaceX to invest $16.8B on first phase of Terafab project

SpaceXIntel Corporation

SpaceX plans $16.8 billion for the first phase of its Terafab semiconductor campus in Texas, with Intel lending manufacturing expertise. The move signals a shift from chip buyer to chip maker. But what happens to SpaceX's existing reliance on Foxconn, NVIDIA, and Super Micro once Terafab scales up?

High impactAI Compute · Wccftech

NVIDIA’s Vera Rubin Memory Bill Doubles Grace Blackwell’s, As DRAM Cost Swells 2.5x In A Single Generation, With One Rack Holding 4,500 Smartphones’ Worth Of Memory

NVIDIA CorporationAdvanced Micro Devices

Nvidia's Vera Rubin racks now pack the memory equivalent of 4,500 smartphones each, as DRAM costs jump 2.5x in one generation. With dual-source HBM3E from SK hynix and Micron underpinning Blackwell, how much of Vera Rubin's total superchip cost does memory now claim?

High impactPhysical AI · Inelectronics

Aeva moves lidar photonics into optical interconnects

AevaSICK

Aeva is pushing its lidar-grade Semiconductor Optical Amplifier tech into data-centre optical interconnects, launching a new Optical Connectivity unit tied to a hyperscaler-bound joint development deal. See how this pivot stacks against Aeva's existing sensing customer base, including AMD, Amazon, NVIDIA and Bosch, before betting on the ramp.

NotableAI Compute · Sammobile

Samsung crosses major milestone in flagship AI memory chip production

Samsung Electronics Co., Ltd.SK hynix Inc.Advanced Micro DevicesNVIDIA CorporationMicron Technology

Samsung says it's hit a major yield milestone on its next-gen AI memory chips, narrowing the gap with HBM leader SK Hynix. The company is now chasing a bold 2026 market-share target. Could this finally win Samsung a foothold in Tesla's HBM supply chain?

NotableAI Compute · Seattletimes

Microsoft plans production boost for AI chips, report says

Microsoft CorporationAmazon.com, Inc.

Microsoft is reportedly negotiating to lock down manufacturing for over 300,000 Maia 300 chips for 2027 delivery, a sharp scale-up in its custom silicon ambitions. That capacity hinges on a single named supplier whose exposure our graph hasn't yet mapped: TSMC.

NotableAI Compute · Trendforce

[News] Samsung’s HBM4 Yield Reportedly Hits 80% as HBM Race Heats Up, SK hynix Labor Talks Add a Twist

SK hynix Inc.Samsung Electronics Co., Ltd.

Samsung claims its HBM4 yield jumped to 80%, well ahead of schedule, just as SK hynix scrambles to lock in equipment orders for second-half capacity. Both chase the same NVIDIA Blackwell demand with overlapping supplier bases. Watch Samsung's targeted 38% HBM market share by year-end 2026 against SK hynix's Yongin Y2 fab timeline.

NotablePhysical AI · Reuters

What is Unitree and why are China’s humanoid robot makers racing to list?

Unitree RoboticsAgiBot

Unitree just priced its Shanghai IPO at 150.8 yuan a share, raising $904 million as China's first publicly listed humanoid robot maker. The retail frenzy was intense, but the real story is upstream. Which suppliers, from NVIDIA to LG Energy Solution to RoboSense, now face mounting volume pressure as Unitree scales globally?

NotablePhysical AI · Cleantechnica

$10 Billion Tesla Solar Factory in Texas? And $20–25 Billion TERAFAB Project in Midst of AI Bubble?

Tesla, Inc.

Tesla's Project Crystal Sun, a $10.1 billion Texas solar factory, targets 9,712 permanent jobs and Q1 2029 operations. But a second, larger disclosure looms in the same filings. Watch what TERAFAB's joint-venture structure with SpaceX implies for Tesla's reliance on Marvell, Micron, and Qualcomm.

NotableQuantum · Thestreet

Pentagon may see one of quantum computing’s first practical winners

IonQ

IonQ just landed a $28 million DARPA deal to build 25 Evergreen-05 optical atomic clocks, with an option for 100 more worth up to $58 million total. Trace the acquisition behind this hardware pivot back to where the Evergreen-05 technology actually came from.

ContextAI Compute · Gizmodo

Apple Reportedly Testing Controversial Chinese Chips as Memory Supply Crunch Intensifies

Apple

Apple is reportedly testing memory chips from China's CXMT for China-market iPhones and MacBooks as the global DRAM crunch worsens, pending US government sign-off. Apple's own earnings call flagged rising memory costs with 'limited impact' from current fixes. Could a China-only CXMT deal quietly become a global procurement shift for Micron's replacement?

ContextAI Compute · Macrumors

Apple Testing Chinese Memory Chips in iPhones and MacBooks

AppleHPCXMT

Apple is reportedly testing CXMT memory chips in iPhones and MacBooks, following HP's move to use the Chinese supplier in devices sold outside the U.S. Apple's memory lineup has long leaned on Micron, SK Hynix, and Samsung. With a Senate deadline looming, can CXMT's capped production even support Apple's needs?

Week of August 3 66 stories ▾

Sunday, August 9

High impactAI Compute · Techtimes

TSMC Revives Longtan for 1.4nm Fabs and CoWoS Hub as Both Sell Out Globally

Taiwan Semiconductor Manufacturing Company Limited

TSMC is bringing Longtan Phase 3 back from the dead, adding two to three 1.4nm fabs plus CoWoS packaging lines as both sell out worldwide. Watch how this single Taoyuan cluster reshuffles allocation certainty for the four chipmakers most starved for advanced capacity.

High impactAI Compute · Yahoo Finance

Western Digital Says AI Customers Are Already Negotiating Storage Deals Through 2031: 'Visibility Remains Very Strong'

Micron TechnologyWestern Digital CorporationSK hynix Inc.

Western Digital says AI customers are locking in storage deals stretching to 2031, calling demand visibility 'very strong.' That's a longer horizon than even Micron's multi-year HBM commitments or SK hynix's fab expansion timelines. What does a seven-year hyperscaler commitment signal about how durable this capex cycle really is?

High impactAI Compute · Mashable

Apple Reportedly Tests CXMT Memory Chips For iPhones And MacBooks Amid AI-Driven Shortage

ChangXin Memory Technologies

Apple is reportedly testing memory chips from China's CXMT for iPhones and MacBooks amid an AI-driven shortage. CXMT already ships to HP, Huawei, Lenovo, and Samsung, but Apple approval hinges on Washington's export-control review of the Trump administration's stance toward the state-backed chipmaker.

High impactSpace · Newatlas

Space Force invests millions in advanced threat detection 'flatellites'

SpaceXRocket Lab

Space Force is pouring millions into flat, low-cost 'flatellites' for advanced threat detection. Rocket Lab and SpaceX both sit deep in the military space supply chain already. Trace how the $397 million Flatellite award stacks against SpaceX's separate multi-billion-dollar SB-AMTI stake in the same program.

High impactAI Compute · Reuters

Apple tests China's CXMT memory chips for iPhones and MacBooks, WSJ reports

CXMTHPApple

Apple is reportedly testing CXMT memory chips across iPhones and MacBooks as HP already ships CXMT-equipped devices outside the U.S. to dodge shortages. CXMT just IPO'd on July 28 with shares up 466% on debut, and its scope of talks with Apple centers on a specific geography.

High impactAI Compute · SCMP

China's 'little Nvidia' plans Hong Kong listing after 147% revenue jump

Moore Threads

Moore Threads is heading to Hong Kong after first-half 2026 revenue hit 1.7 billion yuan and net losses shrank to 11.6 million yuan. Watch what the listing plan actually reveals about its Taiwan Manufacturing foundry dependency and whether offshore capital can buy it sourcing optionality.

NotablePhysical AI · Xataka

2026 marks the beginning of humanoid robots' meteoric rise: this chart illustrates it perfectly

Unitree RoboticsXiaomi

A viral chart says 2026 is humanoid robotics' liftoff year, with Unitree leading China's charge and freshly flush from a record-shattering Shanghai IPO. But the growth story rides on parts most headlines skip. Trace Unitree's G1 supply chain down to its harmonic reducer supplier and see where the real bottleneck sits.

NotablePhysical AI · Khaleejtimes

Khor Fakkan Port to receive giant ship with 6,068 new energy vehicles from China

BYD

A giant ro-ro vessel is delivering 6,068 new-energy vehicles from China to Khor Fakkan Port, opening a fresh Shenzhen-to-UAE shipping lane. What does routing this much volume through a single Gulftainer-run port mean for BYD's exposure to chokepoint risk? Fathom It to see the discharge-rate math.

ContextPhysical AI · Korea Herald

BYD's humanoid push challenges Hyundai's early lead

BYDHumanoid

BYD's new Xiao Di humanoid stands 1.61 meters tall with 31 degrees of freedom and millimeter-precise hands, adding a new front to its rivalry with Hyundai in robotics. The bot's LIDAR precision traces back to a specific BYD supplier already embedded in its EV vertical integration.

Saturday, August 8

High impactAI Compute · Techtimes

SK Hynix Board Locks Korean AI Memory Through 2033; US Has Zero HBM Wafer Fabs

AppleMicron TechnologyTaiwan Semiconductor Manufacturing Company LimitedSamsung Electronics Co., Ltd.SK hynix Inc.

SK Hynix's board just locked in two new Korean fabs, Yongin Y2 and Cheongju M17, running through 2033. Even Micron's US HBM depends on wafers sourced from SK Hynix and fabricated in Japan. So when does America's only HBM alternative actually stop relying on Korea for the wafers themselves?

High impactPhysical AI · Zerohedge

Digging Deeper Inside Physical AI's Hardware Stack Powering Humanoids

Harmonic Drive SE

Physical AI's hardware stack is getting a closer look, from actuators to sensors powering the humanoid boom. One name keeps surfacing at the gearbox level: Harmonic Drive. With Boston Dynamics, Figure AI, and Universal Robots all relying on its reducers, what happens to humanoid scaling if this single chokepoint can't keep pace?

High impactAI Compute · Polandsun

Samsung, SK Hynix evaluate Chinese chipmaking equipment

Samsung Electronics Co., Ltd.SK hynix Inc.Applied MaterialsLam ResearchAMECKLA Corporation

Samsung and SK Hynix are reportedly testing Chinese etching tools for their China plants, even after denying it to Reuters. Both firms lost validated end user status this year and now face annual licensing for Western equipment. AMEC's 14nm etching verification at SMIC marks a quiet shift in who backs up Lam Research and Applied Materials.

High impactAI Compute · Myanmarnews

Korean chipmakers test Chinese tools as export risks grow

Advanced Micro DevicesSK hynix Inc.Samsung Electronics Co., Ltd.Applied MaterialsLam ResearchAMEC

Samsung and SK Hynix are reportedly trialing Chinese etching tools from AMEC as Washington tightens export licensing on their China fabs. Both firms deny formal testing, but sourcing shifts rarely start with press releases. Trace what happens to their 'validated end user' status now that annual licensing replaced it in 2026.

High impactAI Compute · Motley Fool

Forget the Artificial Intelligence (AI) Capex Bubble: SpaceX CEO Elon Musk Says the Memory Shortage Will Drive Costs Even Higher

Micron Technology

Elon Musk says memory supply is growing 20% a year while AI demand grows 200%, calling the chip shortage a bigger cost driver than any AI capex bubble. Watch Micron's HBM4E ramp and its supplier tie to SK hynix, the pressure point that could decide who actually captures the pricing power.

High impactAI Compute · Techspot

SK Hynix is spending $38 billion to build two new chip fabs

SK hynix Inc.Samsung Electronics Co., Ltd.Micron TechnologyNVIDIA CorporationAdvanced Micro DevicesSpaceX

SK Hynix is pouring $38 billion into two new fabs, Yongin Y2 for DRAM and HBM, and Cheongju M17 for NAND. But construction timelines stretch into 2029 while a new unified labor union just formed across its plants. Will Y2's cleanroom hit its target date before NVIDIA's Blackwell demand outpaces supply?

NotablePhysical AI · Reuters

Apple says Mac users in China can connect to Alibaba's Qwen AI service

LenovoThomsonAppleSK hynix Inc.

Apple is letting Mac users in mainland China plug Alibaba's Qwen into Siri and Writing Tools, giving it a local AI assist without building its own model. Mac's China market share sits at just 9% against Lenovo and Huawei. Watch the fine print on what Qwen is contractually barred from doing with user data.

Friday, August 7

High impactPhysical AI · Yicai Global

Unitree Robotics to Raise USD843 Million in Shanghai IPO

Unitree Robotics

Unitree Robotics is set to raise USD843 million on Shanghai's STAR Market, with retail bids topping 53 million shares against just 6.5 million allocated. Trace the 13 supplier ties behind the G1 and H1 platforms, from Chinese-made reducers and bearings to one lingering NVIDIA chip dependency.

High impactAI Compute · Economictimes

Trump rolls out trade measures to counter China on solar and chips, slaps 15% tariff on polysilicon

CorningShin-Etsu ChemicalHanwhaSamsung Electronics Co., Ltd.Hemlock Semiconductor

Trump's new 15% polysilicon tariff and import price floors hit a supply chain with just two U.S. producers left standing. The move lands weeks after one of those producers, Hemlock's parent Corning, saw shares fall more than 50% from their June 2026 high. The brief maps Corning's Hemlock stake against its wider chip and hyperscaler exposure.

High impactAI Compute · Techtimes

AMD Buys Taalas to Hardwire AI Models Into Silicon, Bypassing GPU Memory Wall

GroqAdvanced Micro DevicesMeta PlatformsCerebras Systems

AMD has agreed to acquire Toronto startup Taalas, whose chips hardwire AI model weights directly into silicon rather than reading them from memory each time. The deal targets the notorious memory wall that slows every GPU-based inference system. Trace what Taalas's HC1 chip actually does to Llama 3.1's weights before you assume this just tweaks the Helios roadmap.

High impactPhysical AI · Techrepublic

Boston Dynamics’ Atlas Robot Can Change Its Own Battery in Under 3 Minutes

Boston DynamicsTesla, Inc.

Boston Dynamics' new Atlas can swap its own battery in under three minutes, ending the 90-minute charging pause that idled earlier units. That speed isn't just convenience, it's a spec upgrade running through one battery supplier. What does hot-swappable design demand from LG Energy Solution as Hyundai scales toward 30,000 robots a year?

High impactAI Compute · CNBC

SK Hynix to invest $38 billion building new memory chip plants as demand soars

SK hynix Inc.

SK Hynix will pour $38.1 billion into two new fabs, Y2 in Yongin and M17 in Cheongju, chasing surging memory demand. Watch the HBM supply line feeding NVIDIA's Blackwell roadmap and see why analysts say these plants won't ease prices before 2028.

High impactAI Compute · Webpronews

China Claims Global Lead in Semiconductor Output by Redefining 'Advanced' Chips

Taiwan Semiconductor Manufacturing Company LimitedSamsung Electronics Co., Ltd.Intel CorporationSMICHuawei HiSiliconASML

China is reclassifying which chips count as 'advanced' to boost its output rankings, a definitional shift centered on 28-nanometer production. But the real story sits in equipment dependency. Trace SMIC's ASML sourcing and the EUV chokepoint that still separates rhetoric from actual fab capability.

NotableSpace · Foxbusiness

SpaceX and Tesla choose Texas for AI chip manufacturing plant that will be world's largest building

SpaceX

SpaceX and Tesla are teaming up to build Terafab, a giant Texas plant for AI chips and edge computing gear, pitched as the world's largest building. SpaceX says it will invest $16.8 billion upfront. If SpaceX already sources chips from NVIDIA, Intel, and Foxconn, why is it justifying Terafab as a fix for its own supply chain?

NotablePhysical AI · Interestingengineering

China expands world’s largest rare earth mine by 50% to 15 million tonnes annually

China Northern Rare Earth High-Tech Co., Ltd.

Baogang Group is pouring over 500 million yuan into Bayan Obo, lifting mining capacity 50% to 15 million tonnes a year at the world's largest rare earth deposit. More ore doesn't mean more rare earth oxide reaches the market. CNREH's 95,000-tonne 2025 output sits downstream of a quota system that mining volume alone can't move.

NotableAI Compute · Korea Times

Sk hynix to invest $38.3 bil. in Yongin's Y2 fab, Cheongju's M17 fab - The Korea Times

SK hynix Inc.

SK hynix is pouring $38.3 billion into new Y2 and M17 fabs at Yongin and Cheongju, chasing a government-backed timeline pulled forward by over a decade. Watch what Y2's 2029 HBM output timing means for NVIDIA's Blackwell supply chain once the cleanroom opens.

NotableAI Compute · Indiatimes

AMD AI inference: AMD deepens AI inference bet with Taalas deal as chip race heats up

Advanced Micro DevicesGroq

AMD is buying Taalas, a Toronto startup that hard-wires AI model weights into chips, to sharpen its inference lineup as Nvidia and Groq push rival tech. It's AMD's fourth AI acquisition in under a year. But does adding Taalas' silicon tighten AMD's exposure to Taiwan Manufacturing rather than reduce it?

NotableAI Compute · Financialpost

AMD’s acquisition of Toronto chip startup Taalas shows Canada needs to do more to keep companies at home

Advanced Micro DevicesTenstorrent

AMD is buying Toronto's Taalas to sharpen its AI inference edge, marking its second Canadian chip acquisition in just over a year. The deal targets weight-printing tech that bypasses a specific bottleneck AMD's current GPU lineup can't avoid. Fathom traces how this move fits AMD's existing Taiwan Manufacturing dependency and Nvidia rivalry.

NotableAI Compute · Nikkei

Hitachi and Intel to develop silicon quantum technology; target 1,000 qubits by fiscal 2030

Intel CorporationHitachi, Ltd.

Hitachi and Intel are teaming up on silicon quantum chips, aiming for 1,000 qubits by fiscal 2030. The two already trade GPUs and physical AI tech under a 2026 pact. Trace how Intel's Taiwan-heavy foundry ties to TSMC and UMC could shape this quantum ambition before the qubit race even starts.

Thursday, August 6

High impactAI Compute · Indiatimes

Tata's Dholera fab eyes Entegris boost for chemical supply

Entegris

Tata Electronics is reportedly in early talks with Entegris to supply specialty chemicals for its Dholera fab, a greenfield semiconductor project in India. Entegris already counts SK hynix, Samsung, and NVIDIA among its customers. Does this India expansion signal something deeper about how Entegris is positioned amid its own leadership transition?

High impactAI Compute · Korea Herald

China's homegrown DUV jolts chip industry, but yield test looms

ASML

China says it has built a homegrown immersion DUV lithography system, rattling ASML investors overnight. Early reports point to a single-stage design still far from the dual-stage architecture used in ASML's machines. The brief tracks the yield threshold this system must clear before it threatens ASML's chokepoint on advanced chipmaking.

High impactAI Compute · Eweek

Samsung, SK Hynix Reportedly Test Chinese Chip Tools Amid U.S. Export Risks

Samsung Electronics Co., Ltd.SK hynix Inc.Applied MaterialsLam ResearchKLA CorporationAMEC

Samsung and SK Hynix are reportedly piloting Chinese-made chip equipment as U.S. export controls tighten pressure on their toolchains. Both memory giants still lean heavily on Lam Research, ASML, and Entegris for core production. Watch AMEC's 14nm etching tool verification at SMIC for a read on how far this substitution could actually go.

High impactSpace · Satnews

Viasat Advances Equatys D2D Joint Venture Toward Initial Constellation Procurement

ViasatRocket Lab

Viasat's Equatys venture with Space42 is moving from planning to active procurement for a constellation of up to 2,800 satellites. The joint venture needs a bus manufacturer at massive scale, and one company keeps surfacing as the likely pick. Can Rocket Lab's production capacity actually absorb an order that size without becoming Equatys' single point of failure?

High impactAI Compute · GCN

Transistors smaller than a strand of DNA just packed nearly 100 billion switches onto a fingernail, and the stacking trick IBM used to cross a barrier physicists thought would stop

Intel CorporationSamsung Electronics Co., Ltd.NVIDIA Corporation

IBM says its new 0.7nm 'nanostack' architecture crams almost 100 billion transistors onto a fingernail-sized chip by stacking transistors vertically instead of shrinking them further. Track how Intel's 18A ramp at Fab 52 and Samsung's SF1.4 roadmap now measure up against a rival that isn't even in the foundry race.

High impactAI Compute · Techspot

Apple A20 Pro production is going smoothly , except DRAM supply is bottlenecking iPhone assembly

AppleMicron TechnologySK hynix Inc.Samsung Electronics Co., Ltd.BYD

A20 Pro chips are rolling off TSMC's N2 line smoothly, but the real holdup is memory needed to finish the job. What's actually piling up at TSMC while Apple waits on packaged DRAM for its WMCM integration step, and what does that mean for BYD and Foxconn's ramp before iPhone 18 launch?

NotablePhysical AI · Investors

Next AI Battleground: U.S., China Scramble To Lead In Humanoid Robots

Tesla, Inc.NVIDIA CorporationSpaceXUnitree RoboticsUBTech RoboticsAgiBot

Washington and Beijing are racing to dominate humanoid robotics, with Tesla, NVIDIA, and Chinese giants like Unitree and UBTech building rival supply chains. Behind the headlines, the component sourcing tells its own story of dependency and leverage. Which chip, sensor, or rare-earth supplier sits at the choke point of both nations' humanoid ambitions?

NotableSpace · Statesman

SpaceX makes Terafab official, confirms initial $16.8 billion investment

SpaceXTesla, Inc.Intel Corporation

SpaceX has officially confirmed Terafab, a joint venture backed by an initial $16.8 billion investment. The move formalizes a chipmaking partnership with major implications for existing supplier networks. Our graph tracks how Tesla and Intel already feed SpaceX's compute and robotics stack through overlapping vendor dependencies, including Intel's own foundry outsourcing arrangements.

NotableAI Compute · Stocktwits

NVDA Reportedly Weighs Lower-Memory Rubin Ultra GPU Designs To Ease HBM Bottleneck — Why Retail Is Watching Micron

NVIDIA CorporationMicron Technology

NVIDIA is reportedly testing Rubin Ultra GPU variants with less HBM than the original 1TB spec, a sign that memory supply, not design ambition, is now capping the roadmap. Trace how Micron's HBM4E timeline for 2027 lines up against a shifting, fragmented demand picture across these lower-memory variants.

NotableSpace · AL.com

3 companies getting $615 million for satellites to track aircraft and missile threats

Rocket LabSpaceX

The Space Force just handed $615 million to three companies, including Rocket Lab, to build satellites that spot aircraft and missile threats from orbit. Rocket Lab's $397 million slice covers designing and operating a fleet of Flatellite spacecraft, not just launching them. What does that shift mean for its pending Iridium deal?

NotablePhysical AI · Reuters

SpaceX, Tesla to initially spend $16.8 billion on Terafab chip plant in Texas

Tesla, Inc.SpaceXTeraFab

Tesla and SpaceX will initially sink $16.8 billion into Terafab, a Texas chip complex meant to feed Optimus robots, Cybercabs, and space-based data centers. Marvell, Micron, and Qualcomm currently anchor Tesla's chip sourcing, and Terafab's build-out plan reveals what that dependency is really worth.

NotablePhysical AI · Reuters

US telecoms regulator chief says curbs on Chinese tech imports aim to spur production, counter security risks

Tesla, Inc.

FCC chair Brendan Carr says import curbs on Chinese tech aim to boost US production and cut security risks, as Tesla and SpaceX pour $16.8 billion into the Terafab chip plant in Texas. Watch how CATL's battery cells fit into Tesla's newly exposed supply chain.

High impactSpace · WSJ

SpaceX, Tesla to Build Chip Plant in Texas to Meet Compute Demand

SpaceXTesla, Inc.TeraFab

SpaceX and Tesla are teaming up on a $16.8 billion Texas chip fab aiming for 1 terawatt of annual compute. Both firms already lean hard on outside suppliers for critical silicon and memory. Which merchant vendor, tied to Tesla's Optimus HBM needs, does Terafab most threaten to sideline?

High impactAI Compute · Reuters

Merck KGaA raises profit guidance on momentum in Life Science, Electronics units

Merck KGaA

Merck KGaA lifted its 2026 EBITDA guidance to €5.9 billion to €6.3 billion, citing stronger demand for drug manufacturing supplies and semiconductor materials. The company's AI chip supply ties to TSMC and Texas Instruments sit behind that guidance bump. Fathom It breaks down what those relationships and Merck's €4 billion M&A spree actually signal for chip material continuity.

High impactAI Compute · EE Times

GlobalFoundries’ Growth Makes the Case for a U.S. Photonics Buildout

GlobalFoundries

GlobalFoundries is scaling up capacity in Vermont, Malta, and Singapore as demand for photonics and specialty chips surges. Q2 revenue hit $1.786 billion with communications infrastructure up 62% YoY, prompting a full-year guidance hike. Watch how silicon germanium capacity, already oversubscribed through 2027, reshapes who gets supplied first.

NotableAI Compute · Reuters

DeepSeek invests $20.8 million in Unitree's Shanghai IPO

Unitree RoboticsDeepSeekHumanoid

DeepSeek is putting $20.8 million into Unitree's blockbuster Shanghai IPO, pairing up to build AI models for humanoid robots. Unitree's G1 already leans on suppliers like RoboSense and Leaderdrive for its hardware. So what technical gap in DeepSeek's own AI stack is this robotics tie-up really meant to patch?

ContextPhysical AI · Bloomberg

China’s Unitree Seeks $904 Million in First Mainland Robotic IPO

Unitree Robotics

Unitree Robotics is raising about 6.1 billion yuan ($904 million) in Shanghai's first mainland humanoid robotics IPO, selling 40.4 million shares at 150.80 yuan each. But which single foreign chipmaker underpins the G1's processing, and what happens to Unitree's supply chain if that link gets cut?

Wednesday, August 5

High impactAI Compute · Reuters

China AI hardware stocks slump after news US plans to ban imports of Chinese components

Eoptolink

Chinese AI hardware stocks dropped after reports the US is drafting a ban on Chinese component imports. Optical module maker Eoptolink saw shares slide roughly 8% amid the news. Trace what a transceiver ban means for Eoptolink's Microsoft, Amazon, NVIDIA and Meta supply lines, and its pending $5B Hong Kong listing.

High impactAI Compute · Indiatimes

Global Market: China optical stocks sink on reports of potential US import ban

Eoptolink

Chinese optical stocks slid after reports the US may ban imports of new Chinese data-center components. Shares of major suppliers dropped sharply as investors weighed the fallout. With hyperscaler contracts and Broadcom-based ASIC supply chains at stake, does Eoptolink's Thailand expansion and Hong Kong listing plan offer enough cover if the ban actually lands?

High impactAI Compute · Trendforce

[News] U.S. Export Curbs Backfire? Samsung, SK hynix Reportedly Mull AMEC's Tools as China Fab Backup

SK hynix Inc.Samsung Electronics Co., Ltd.AMEC

Samsung and SK hynix are reportedly eyeing China's AMEC as a fallback for etching tools used at their Xi'an, Dalian, and Wuxi fabs, hedging against tightening U.S. export rules. AMEC currently has no confirmed supply relationship with either memory giant, only tools under verification at SMIC.

High impactSpace · Breakingdefense

Space Force awards 3 firms $615M to track airborne targets

Rocket LabSpaceX

Space Force handed out $615M for airborne target-tracking satellites, spreading work across three contractors. Rocket Lab lands the biggest slice of this tranche. Follow how a $397M task order tied to Rocket Lab's Flatellite constellation design stacks up against its $602M annual revenue base.

High impactAI Compute · Engadget

Samsung Announces Next-Gen 3D Memory When All We Want Is Reasonably Priced DDR5

Samsung Electronics Co., Ltd.SK hynix Inc.

Samsung just unveiled zHBM and V10 BV-NAND, flashy specs that won't fix your DDR5 shortage anytime soon. The real story is where all this new memory capacity is actually headed. With Samsung and SK Hynix controlling the bulk of global production, which buyers are quietly locking up the supply that's left?

High impactAI Compute · Nikkei Asia

China hits back at US FCC bans with sanctions and drone curbs

ChangXin Memory TechnologiesHPCXMT

China answered US FCC bans with sanctions and drone export curbs, escalating a fight that's already reshaping hardware supply chains. HP, Asus and Acer have quietly turned to China's CXMT for memory chips as shortages bite. Our graph maps how deep that CXMT dependency now runs across HP's AI PC lineup.

NotablePhysical AI · Financial Times

China hits back at US with sanctions and tighter drone export rules

SZ DJI Technology Co., Ltd.EoptolinkZhongji InnoLightHuawei Technologies Co., Ltd.

China's new export controls target drones and components heading to the US, hitting DJI just as it fights FCC restrictions in court. Trace how DJI's Matrice line depends on Western chipmakers like STMicroelectronics and Texas Instruments to see where this bilateral squeeze actually bites.

ContextAI Compute · Wccftech

CXMT Says No Thank You to Apple's Demand For A Price Cut, as Huawei And Xiaomi Hand It Rare Leverage - Report

CXMTXiaomi

CXMT has turned down Apple's request for a discount on DRAM chips, after Huawei and Xiaomi locked in pricey advance contracts. The Chinese memory maker now quotes prices matching Samsung and SK hynix instead of undercutting them. Does this mean Apple just lost its favorite bargaining chip against Korea's memory giants?

Tuesday, August 4

High impactAI Compute · Korea Herald

SK hynix, Sandisk set first HBF standard for AI memory

AlphabetTenstorrentSK hynix Inc.SanDisk

SK hynix and SanDisk just published the first HBF standard, a new memory tier sitting between HBM and SSDs for AI workloads. Both suppliers already feed giants like NVIDIA, Microsoft, and Tesla amid fab expansions and stock swings. Trace how Tenstorrent's acquisition rumors and Yongin Y2 timelines factor into this new spec's rollout.

High impactSpace · Reuters

Focus: SpaceX's satellite ambitions squeeze out rivals reliant on its rockets

IridiumRocket Lab

Falcon 9's manifest is filling with SpaceX's own Starlink satellites, squeezing out rival operators who depend on those launches. Iridium and its supplier Rocket Lab feel it most, with over 10,000 satellites in orbit and no backup ride to space. Watch what Rocket Lab's Neutron timeline and the pending Iridium acquisition actually mean for that bottleneck.

High impactAI Compute · Wccftech

SK hynix, In Collaboration With SanDisk, Unveils The New High Bandwidth Flash (HBF) Standard, Helping To Resolve AI Inference Bottlenecks, Targeting Up To 3TB/s Bandwidth

SK hynix Inc.SanDisk

SK hynix and SanDisk just introduced High Bandwidth Flash, a new memory standard aimed at easing AI inference bottlenecks with speeds up to 3TB/s. It slots between HBM and SSDs as an open ecosystem play. What does that mean for Samsung's rival CMX effort tied to NVIDIA's V10/V11 NAND stacks?

High impactAI Compute · Wccftech

The FCC Mulls Banning China-Sourced Optical Transceivers, Threatening Innolight’s 27% Global Market Share As Coherent And Lumentum Prepare To Pounce

LumentumCoherent

The FCC is weighing curbs on China-sourced optical transceivers, putting Innolight's 27% global share in the crosshairs while Coherent and Lumentum wait in the wings. A full ban looks unlikely given compatibility testing hurdles, but a narrower 1.6T upgrade restriction is on the table. Coherent's Indium Phosphide capacity constraint shapes how much of that upside it can actually capture.

High impactAI Compute · Indiatimes

Trump administration drafting ban on Chinese data center devices, sources say

Coherent

The White House is drafting a ban on Chinese data center components, with optical transceivers reportedly the primary target. That would shake up who supplies the gear behind AI buildouts at hyperscalers like AWS. Watch how Coherent's Pluggable Optics ramp and its NVIDIA Blackwell ties factor into who fills the gap.

High impactAI Compute · Nikkei Asia

HP, Asus and Acer begin using CXMT chips amid memory shortage

ChangXin Memory TechnologiesHP

Facing a global memory crunch, HP, Asus and Acer have quietly started pulling small volumes of chips from China's CXMT. It sounds like a minor stopgap, but design wins on HP's AI PC and Spectre lines suggest otherwise. How much leverage does an $8.6 billion IPO buy CXMT with Micron, Samsung and SK Hynix watching closely?

NotablePhysical AI · Reuters

Chinese robot maker Unitree seen worth over $7.4 billion yuan after IPO, Citic says

Unitree Robotics

Citic pegs Unitree's post-IPO value above 50 billion yuan after retail investors piled in at 8,000 times oversubscription. The Shanghai listing puts a public price on a robotics supply chain that still leans on single-sourced vendors for its core sensing stack. Our graph tracks Unitree's dependency on Hesai and RoboSense LiDAR across its humanoid and quadruped lines.

Monday, August 3

High impactPhysical AI · DIGITIMES

FCC robot restrictions cast uncertainty over Unitree and China's consumer robotics exports

Unitree RoboticsD-Robotics

New FCC restrictions on foreign-made robotics and power inverters have unsettled Chinese exporters like Unitree, just days after its blockbuster Shanghai IPO. Watch what the Covered List designation means for Unitree's NVIDIA-dependent compute stack and its freshly raised 6.1 billion yuan in capital.

High impactAI Compute · Reason

FCC Bans Foreign-Made Robots, Including Roombas

Amazon.com, Inc.SZ DJI Technology Co., Ltd.

The FCC's new ban on foreign-made robots over 4.4 pounds sweeps in Roombas and beyond, threatening companies that already sourced components abroad. Amazon and DJI both sit in the crosshairs, but their exposure looks nothing alike. Which of the two faces a $1.5 billion hit to its 2026 revenue?

High impactAI Compute · Reuters

EXCLUSIVE: CXMT plans second chip plant in Beijing and is in talks on its funding, sources say

CXMT

CXMT is reportedly building a second Beijing DRAM fab, with financing talks already underway before its record IPO. The company already runs three 12-inch fabs feeding Dell, HP, Huawei and Alibaba. The Yizhuang project's funding structure and capacity target haven't been confirmed publicly.

High impactAI Compute · Bloomberg

ASML Supplier Zeiss Says It Can Handle Demand for Key AI Parts

ASML

Zeiss says it has enough capacity to keep pace with surging AI chip demand, easing worries about a bottleneck feeding ASML's EUV lithography machines. But ASML's bill of materials leans on a razor-thin set of German suppliers. See how concentrated ASML's sourcing really is around Zeiss before taking the reassurance at face value.

High impactAI Compute · Bloomberg

CXMT's $1 Trillion Target Gap Flags High Stakes in Chip Battle

CXMT

CXMT's Shanghai debut popped 466% on day one, leaving analyst price targets nearly $1 trillion apart. Behind the volatility sits a sprawling HBM customer list spanning Dell, HP, Alibaba, Apple, and Huawei. Which of those relationships carries the real concentration risk if the polarized bet on CXMT goes wrong?

ContextPhysical AI · EE Times

Renesas Tackles Memory Bottleneck with MRDIMM Update

Renesas ElectronicsRambus Inc.

Renesas rolled out Gen 3 DDR5 MRDIMM chipsets, promising up to 16,000 MT/s and 25% more bandwidth than Gen 2 for AI and HPC servers. Watch how this platform push against Rambus's IP model collides with Renesas's own Kumamoto plant restart timeline starting August 5, 2026.

Week of July 27 23 stories ▾

Sunday, August 2

High impactAI Compute · Fortune

CXMT's blockbuster IPO will test whether China's memory makers are ready for the spotlight: 'It does not yet mean China is broadly catching up' | Fortune

CXMT

CXMT's Shanghai IPO exploded 466% in a day, raising $8.6 billion and instantly making it China's biggest chipmaker debut of 2026. But the hype hides a messier supply picture. Trace who's actually buying CXMT's HBM chips right now, including one name that sits surprisingly close to SK Hynix itself.

High impactPhysical AI · Therobotreport

Google DeepMind says Gemini Robotics 2 enables full body control

Apptronik

Google DeepMind says its Gemini Robotics 2 model now controls a humanoid's entire body, not just its arms, unlocking whole-body coordination and dexterity. The demo runs exclusively on Apptronik's Apollo 2. What does deeper whole-body control demand from the SharpaWave hand's 22 degrees of freedom, and who supplies the parts to match it?

High impactAI Compute · Indiatimes

AI keeps consumer prices high in 'RAMaggedon' chip crunch

CXMT

RAM prices have jumped five to six times year-over-year, with Samsung warning the shortage could stretch through 2028 as AI demand devours HBM supply. Amid this, a fourth-tier Chinese DRAM maker just staged an $8.6 billion Shanghai IPO. Our graph maps CXMT's supply lines into Dell, Apple, Huawei, and even SK hynix.

High impactPhysical AI · Yahoo

Nvidia (NVDA) Teams Up With Kawasaki To Build AI Powered Shipyard Robots

Kawasaki Heavy Industries

Nvidia and Kawasaki Heavy Industries are building a digital shipyard powered by physical AI and digital twins, pushing GPU compute into welding, painting, and inspection lines. See how Kawasaki's existing precision-robotics ties to Toyota factor into its new dependency on Nvidia's compute stack.

High impactPhysical AI · SCMP

Mixed feelings: can China's companion robots win a place at home?

UBTech Robotics

China's push to bring companion robots into homes is gaining traction, with UBTech's U1 racking up 13,361 pre-orders ahead of its September delivery date. But behind the sleek marketing sits a supply chain leaning on sole-sourced Harmonic Reducers from Leaderdrive. Can that single dependency survive a mass-market scale-up?

NotableAI Compute · Nasdaq

Mark Zuckerberg Is Betting Up to $145 Billion on AI Infrastructure in 2026

Meta Platforms

Meta has raised its 2026 AI infrastructure capex guidance to as much as $145 billion, doubling down on Zuckerberg's compute buildout even as free cash flow craters. Where does the custom 'Iris' chip fit into Meta's plan to wean itself off merchant GPU suppliers amid a supply crunch?

Saturday, August 1

NotableAI Compute · Indiatimes

Tata Electronics, ASML explore making chip equipment components in India

ASML

Tata Electronics and ASML are reportedly discussing local production of chip equipment components in India, expanding on their May MoU for the Dholera fab. Track which subassemblies actually shift to India versus which stay locked with ASML's core European partners like Carl Zeiss SMT and VDL.

NotableAI Compute · Indiatimes

India enters the story as China begins to rewrite the chips playbook

ASML

China's push into homegrown lithography is colliding with India's first steps into chipmaking, as Tata Electronics joins the global supply chain. Both moves circle back to one Dutch chokepoint. How exposed is ASML's supplier and customer web, from German optics makers to Chinese foundries, now that DUV machines are being built outside its control?

Friday, July 31

High impactAI Compute · Reuters

Linde to invest $1 billion in Arizona after winning semiconductor supply deal

Linde

Linde is committing $1 billion to Arizona fabs after landing a deal with a major chipmaker, part of a broader $800 million Taiwan buildout too. The redacted customer's identity narrows to one of three global fab operators. Linde's build-own-operate model for these gas supply units creates a lock-in structure worth examining.

NotablePhysical AI · Morningstar

Rheinmetall Wins U.K. Contract for Howitzer Weapon Systems

Rheinmetall AGKNDS

Rheinmetall has landed a U.K. contract tied to the RCH 155 howitzer program, arriving as Germany trims munitions budgets at home. See how the Artec joint venture structure with KNDS actually splits revenue between prime integrator and subcontractor roles on this 72-unit order.

NotableQuantum · Reuters

FTC approves IonQ acquisition of SkyWater semiconductor ‍foundry

IonQ

The FTC cleared IonQ's $1.8 billion purchase of SkyWater, its longtime chip foundry partner. The deal turns a supplier contract into an owned subsidiary. With Aria production now in-house, what does this mean for IonQ's leverage over customers like Microsoft and Amazon who depend on its output?

NotableAI Compute · Morningstar

Linde to Invest $1 Billion to Support Major U.S. Semiconductor Facility Expansion

Linde

Linde will pour $1 billion into a new Phoenix facility to serve one of the world's largest semiconductor manufacturers. The customer behind the deal is not named in the announcement. Our brief maps Linde's existing gas-supply ties to Micron, TSMC, Intel, and GlobalFoundries to narrow down who just locked in a multi-year supply chokepoint.

NotablePhysical AI · Forbes

Humanoid Robot Patents: China Most, U.S. Best?

Apptronik

China leads humanoid robot patent filings by a wide margin, but the U.S. reportedly holds the higher-quality claims. A new FCC ban on Chinese humanoids reshapes who wins the domestic market. Trace how Apptronik's Apollo supply chain, from NVIDIA chips to Inspire Robots components, stacks up against unmapped Chinese morphology IP.

Thursday, July 30

High impactAI Compute · Business Standard

Samsung sees chip shortage extending to 2028, signs long-term supply deals

Samsung Electronics Co., Ltd.

Samsung now expects chip shortages to persist through 2028 and is locking in multi-year supply deals with major buyers. Its chip division just posted a 250-fold profit jump while mobile posted a rare loss. What do those long-term contracts with the top five data centre firms actually lock in on pricing and capacity share?

High impactPhysical AI · Thenextweb

China threatens retaliation over the US robot ban

AgiBot

Beijing is vowing to hit back after Washington's new robot import ban, which sweeps up humanoid makers alongside vacuums and delivery bots. The timing hammers AgiBot's fresh Hong Kong IPO filing. Watch what AgiBot's Yuanzheng line owes to Ningbo Yunsheng for rare-earth magnets before betting on that valuation window.

NotableSpace · Nikkei Asia

SpaceX's supply chain clampdown and China's product power

SpaceX

SpaceX is reportedly auditing its supply chain to purge Chinese nationals and equipment, naming Hikvision as a flagged security risk. Track how this squares with SpaceX's own vendor map, including its reliance on Hon Hai Precision Industry for AI infrastructure, before assuming the decoupling is as clean as it sounds.

NotablePhysical AI · Yahoo Finance

SK Innovation Co Ltd (XKRX:096770) (Q2 2026) Earnings Call Highlights: Battery Turnaround Gains ...

SK InnovationSK On

SK Innovation's Q2 2026 earnings call points to a battery turnaround, with SK On's operating profit climbing on cost cuts and Asia volume gains. The Blue Oval SK joint venture with Ford has now fully unwound. What does SK On's full control of the Tennessee plant actually mean for its margins going forward?

NotablePhysical AI · Yahoo

Satellite maker K2 Space secures $6.8 billion valuation in new funding round

Anduril Industries

K2 Space just hit a $6.8 billion valuation, with Anduril tapping the company to build satellites for the Golden Dome missile-defense program. SES is also locked in, betting its meoSphere constellation on K2's Torrance production line. Our graph maps the single-point-of-failure risk building across K2's satellite bus supply chain.

ContextAI Compute · CNBC

The real story at Apple is the memory crunch -- and there's little room for error

AppleMicrosoft Corporation

Apple's price hikes on MacBooks and iPads are just the opener, with iPhone 18 pricing looming in September amid a chip supply warning. Watch what Thursday's gross margin guidance reveals about whether Micron's HBM squeeze is a manageable cost or a structural ceiling on Apple's pricing power.

ContextSpace · Asianetnews

Rocket Lab's Third Deal With Japanese Satellite Firm Gets Retail Buzzing - But RKLB Stock Heads For Worst Ever Month

Rocket Lab

Rocket Lab just locked in its third launch deal with iQPS in nine months, even as RKLB shares slide toward their worst month ever. Retail investors are cheering the recurring business, but repeat contracts cut both ways on a launch manifest. What does Rocket Lab's dependency on customers like Iridium and Globalstar reveal about its Electron cadence risk?

Tuesday, July 28

High impactPhysical AI · Forbes

United States Bans Chinese Humanoid And Quadruped Robots, Citing National Security

Agility Robotics

The FCC just banned new Chinese humanoid and quadruped robot model authorizations, wiping out manufacturers who previously shipped 87% of all humanoid robots as of January 2026. US challengers like Agility Robotics stand to inherit the opening. But with Digit v5 targeted for release and suppliers like Kollmorgen and Schaeffler already in the chain, how ready is Agility to actually fill that void?

NotablePhysical AI · Reuters

Trump administration bans new Chinese humanoid robots, to protect US AI buildout

Unitree RoboticsNVIDIA Corporation

Washington just banned new Chinese humanoid and quadruped robot imports, aiming to shield the US AI buildout. The move lands days after market leader Unitree's blockbuster Shanghai IPO. Trace what a frozen US market does to Unitree's NVIDIA Blackwell chip pipeline and its web of Chinese component suppliers.

Monday, July 27

High impactAI Compute · TechCrunch

Amazon's trying to launch a global satellite cellphone network in 2028

Amazon.com, Inc.SpaceX

Amazon just bought Globalstar's satellite network and its Apple direct-to-device deal, aiming for a 5,105-satellite mobile constellation by 2028. But Globalstar has long relied on SpaceX for launch and service support. What happens to that supplier relationship now that Amazon owns the customer?

Week of July 20 5 stories ▾

Sunday, July 26

NotableAI Compute · The Verge

Amazon wants to launch 5,105 new direct-to-device satellites.

Amazon.com, Inc.

Amazon filed with the FCC to launch 5,105 direct-to-device satellites, nearly doubling its original LEO fleet and dropping the dish requirement entirely. The filing also ties this constellation to the Globalstar system already embedded in iPhones. Our graph tracks which manufacturing chokepoints could throttle a build-out this size.

Thursday, July 23

NotableAI Compute · TechCrunch

Nvidia is sending GPUs to the Moon

NVIDIA Corporation

Lunar Outpost plans to put Nvidia's Jetson chips on its next moon rover, controlling lidar in what would be the first GPU on the lunar surface. It's a trial run for edge compute in extreme conditions, not yet a proven system. Track how Jetson's radiation and thermal test compares to Nvidia's existing space-grade rival.

Tuesday, July 21

NotableAI Compute · CNBC

Intel's foundry lands first named customer under CEO Lip-Bu Tan, as Fortinet signs on for security chips

Intel FoundryIntel Corporation

Fortinet becomes Intel Foundry's first publicly named customer since CEO Lip-Bu Tan took over, committing to a next-gen security chip. The deal lands as Intel ramps Panther Lake production and pursues bigger wins elsewhere. Which process node is Fortinet's SP6 chip actually built on, and what does that reveal about Intel's advanced-node customer pipeline?

NotableAI Compute · Reuters

China's Zhongji Innolight to raise $8 billion in Hong Kong listing, sources say

Zhongji InnoLight

Zhongji Innolight is set to raise about $8 billion on the Hong Kong exchange, one of Asia's largest listings this year. The optical-modules maker counts NVIDIA, Meta, Amazon, and Alphabet among its buyers, and the brief maps its concentrated exposure to their AI datacenter capex cycles.

Monday, July 20

NotableAI Compute · Trendforce

[News] Tata's India Fab Reportedly to Launch at 90nm Instead of 28nm; Commercial Production Delayed to Mid-2028

Tata Electronics

Tata's Dholera fab is starting at 90nm instead of 28nm, with commercial production now slipping to mid-2028. The node downgrade reframes India's chip ambitions from leapfrog to catch-up. Trace how Tata's sole-source dependence on Powerchip (PSMC) for technology transfer intersects with its Apple and Qualcomm supply commitments.