Compute just got a spot price

The Ornn Compute Price Index tracks live-traded GPU spot prices across H100, H200, and B200 hardware. ICE, the exchange that sets the price of Brent crude, wants to list derivatives against it. When the people who price oil decide your GPU hours need a futures contract, something has shifted.

Three exchanges are now racing to build these markets. Shanghai Futures Exchange is designing contracts tied to AI token pricing. CME partnered with Silicon Data on an H100 Rental Index tracking training GPU costs. GPU rental prices surged 48% between mid-February and mid-April 2026, and institutions want instruments to hedge that volatility.

The obvious reading: compute is becoming a commodity. Buy it, sell it, hedge it. But look at what else happened this week.

Anthropic closed a $65 billion Series H at a $965 billion valuation, and the investor list included Samsung, SK Hynix, and Micron as "strategic infrastructure partners." The companies that sell Anthropic its memory chips now own a piece of the company buying them. CNBC reported that Mistral is exploring custom chip design alongside a €4 billion push into data centres targeting 1 GW by 2030. And Samsung's investment may extend beyond memory procurement into foundry manufacturing, giving Anthropic a path to custom silicon made by one of the world's three leading chipmakers.

So one group of actors is building markets premised on compute being fungible and tradeable. Another is building supply chains premised on compute being proprietary and vertically controlled. These are opposing bets running on the same timeline.

The commodity paradox

Commodities markets require standardisation. A barrel of Brent crude is a barrel of Brent crude. That fungibility is what makes it tradeable. But the companies spending the most on AI compute are actively trying to make their compute non-fungible: custom chips, proprietary architectures, dedicated foundry relationships. Anthropic doesn't want to rent the same GPUs as everyone else on a spot market. It wants silicon nobody else can buy.

This tension has a precedent. In the early decades of the oil industry, Standard Oil vertically integrated from wellhead to petrol pump while commodity exchanges simultaneously developed crude futures. For a while, both models coexisted. Rockefeller's proprietary supply chain ran alongside a growing spot market. Eventually the commodity market won. Oil became fully fungible, and the vertically integrated giants either broke apart or specialised.

The question is whether AI compute follows the same arc. Right now, the frontier labs are behaving like early Standard Oil: locking up supply chains, investing in custom manufacturing, building infrastructure that cannot be replicated on a spot market. The exchanges, meanwhile, are pricing the generic layer.

I think both will coexist, but they will serve different tiers. Commodity compute, hedgeable and priced on an index, becomes the baseline for inference and smaller workloads. The frontier labs pushing model capabilities will increasingly run on proprietary silicon that never touches a trading floor. For anyone building on AI today, the practical question is which tier you are on. The cost of generic inference is about to get as predictable as electricity. The cost of staying at the frontier is about to get as opaque as ever. The spread between those two prices is where the next decade of competitive advantage lives.

---