The circular AI trade, mapped

The AI buildout looks like the purest demand story in markets: the hyperscalers can’t buy chips fast enough. But look upstream, at the cap tables instead of the income statements, and the demand starts to look more circular. The companies selling the compute are also funding the companies that buy it. Nowhere is that clearer than in humanoid robotics.

The Figure cap table. Figure AI closed a more-than-$1B Series C at a $39B post-money valuation in September 2025, with NVIDIA, Microsoft, and Amazon all participating Reported · 0.8. Those are the three most consequential names in AI compute. And what do Figure’s robots run on? NVIDIA: Figure 02 performs all inference on dual onboard NVIDIA RTX GPUs and trains its models on NVIDIA H100s Reported · 0.9. So NVIDIA holds equity in Figure, and Figure’s hardware budget flows back to NVIDIA. The investor is the supplier.

This is vendor-financed demand. A supplier taking a stake in its customer is an old playbook, but at this scale it changes how you read the numbers. When the seller funds the buyer, near-term revenue looks like organic growth, and the concentration risk hides in the equity footnotes rather than the order book. The question a hedge fund should be asking: how much of "AI compute demand" is genuinely independent, and how much is financed, directly or through cloud credits and strategic rounds, by the same handful of firms that also supply it?

The concentration underneath. Set aside who funds whom, and the supply is still dangerously thin. NVIDIA's top two (unnamed) customers were roughly 39% of Q2 FY2026 revenue Reported · 0.8, widely understood to be hyperscalers. NVIDIA, in turn, is sole-sourced on its two most critical inputs: TSMC is its only foundry, per NVIDIA's own 10-K Confirmed · 0.9, and SK hynix is the sole supplier of 12-layer HBM3E for the GB300 Reported · 0.8. Single points of failure, stacked three deep, feeding a customer base partly funded by the seller.

What the graph shows. In the Tettares supply graph, NVIDIA now carries three confirmed buyers, Amazon, Microsoft, and Figure, while TSMC and SK hynix each flag "high" customer-concentration on the Herfindahl measure. The ownership layer flags the NVIDIA, Microsoft, and Amazon stakes in Figure as strategic. That last pattern, a supplier holding equity in the company it sells to, is the lock-in that never appears on either party's income statement. It is exactly the relationship our analytics are built to surface.

The contrast: Tesla isn't in the loop. Tesla designs its own AI silicon (AI5 and AI6, fabbed by TSMC and Samsung under a $16.5B Samsung deal Reported · 0.9) and does not depend on NVIDIA for Optimus's brain the way Figure does. So two models of humanoid are emerging side by side: vertically financed (Figure, sitting on its suppliers' cap table) and vertically integrated (Tesla, building its own chips). Their supply chains, and their risk profiles, diverge from there.

What we're watching upstream

• Disclosure of how much hyperscaler AI capex is tied to vendor financing or strategic investment, versus arm's-length demand.
• Any new NVIDIA strategic stake in an AI-compute customer: the loop widening.
• TSMC and SK hynix capacity allocation, the real bottleneck behind the demand narrative.
• Figure's path to revenue. A $39B valuation underwritten partly by its own suppliers eventually needs a customer that isn't also an investor.

The downstream story is "AI demand is insatiable." The upstream question is how much of that demand the suppliers are funding themselves. When the seller owns a slice of the buyer, growth and risk can look like the same number. Read the cap table.

Sources

• Figure AI Series C (NVIDIA, Microsoft, Amazon): Figure AI
• Figure 02 onboard NVIDIA GPUs: NVIDIA blog
• NVIDIA customer concentration (~39% of Q2 FY2026): CNBC, Aug 2025
• TSMC sole foundry for NVIDIA: NVIDIA FY2025 Form 10-K (SEC)
• SK hynix sole 12H HBM3E for GB300: TrendForce, Mar 2025
• Tesla / Samsung $16.5B chip deal: CNBC, Jul 2025