The global data center GPU market is projected to grow from $138 billion in 2026 to $624 billion by 2034, and the five largest hyperscalers will spend a combined $700 billion on AI infrastructure this year alone.
Compute is following the same arc as every commodity before it: opaque bilateral contracting, then standardized benchmarks, then exchange-traded derivatives. The liquidity fragmentation risk is real: three competing futures products with no dominant index risk leaving the market too thin to serve its purpose.
The road from bilateral contracting to exchange-traded compute
In December 2025, Ornn executed the first cleared compute swap: a bilateral agreement between two counterparties to exchange GPU capacity at a future price, settled through a central counterparty. It was a modest transaction. But it marked the moment compute capacity crossed from private contracting into financial market infrastructure.
Six months later, three global exchange operators have announced plans for regulated compute futures markets. The pace reflects a structural gap: AI infrastructure has attracted hundreds of billions in capital without the pricing transparency or risk-management tools that every other major commodity market takes for granted.
As we wrote in July, compute futures are becoming an institutional asset class. What has changed in the weeks since is the speed of market formation.
Three exchange operators, three benchmark strategies
CME Group moved first. On May 12, it announced a partnership with Silicon Data, a GPU market intelligence platform backed by DRW, to launch compute futures based on Silicon Data's daily rental-rate indices. The contracts will reference the Silicon Data H100 Rental Index and cover A100, H100, and B200 GPUs. CME Chairman Terry Duffy called compute "the new oil of the 21st century."
One week later, ICE, the parent company of the New York Stock Exchange, announced its own GPU compute futures with Ornn, a startup whose Compute Price Index (OCPI) is built from printed transactions rather than surveys. OCPI covers H100, H200, B200, and RTX 5090 hardware and is already published on the Bloomberg Terminal. Ornn's CEO Kush Bavaria described compute as "a trillion-dollar market that still lacks the pricing and risk-transfer infrastructure every other major commodity relies on."
AI Infrastructure Spending Is Not Slowing
Amazon, Microsoft, Google, and Meta alone will spend $325 billion on AI infrastructure in 2026. The five largest hyperscalers are on track for nearly $700 billion. · Silicon Data, CME Group, 2026
On May 28, Architect Financial Technologies acquired a U.S. Designated Contract Market and announced the American Innovation Exchange, the first CFTC-regulated venue purpose-built for compute and AI supply chain derivatives. Unlike CME and ICE, which are grafting compute onto existing futures infrastructure, Architect built the exchange around the AI economy from the ground up, with cross-margining across compute, energy, and metals contracts.
On July 1, ICE announced a second compute futures product, this time with NATIVX, whose COIL Index tracks tokenized, energy-normalized compute and connectivity. The COIL approach strips out regional power cost disparities, a feature that matters because electricity is the single largest variable cost in GPU operation. ICE will list these contracts alongside its existing natural gas and power futures, creating an integrated hedging environment.
The same day, Architect partnered with Compute Desk to launch ComputeConnect, the first U.S. exchange-for-physical network for compute, linking CFTC-regulated derivatives to physical GPU capacity delivery. This allows futures holders to convert their positions into actual hardware access, a feature oil and gas markets have had for decades but compute has never had.
The companies building the plumbing
Behind the exchange announcements, a layer of benchmark and index providers is emerging. These companies are the unsung infrastructure of the emerging market: they determine which price becomes the reference rate for the entire industry.
Silicon Data, founded by former DRW trader Carmen Li, publishes daily GPU rental-rate indices distributed through Bloomberg and Refinitiv. Its H100 Rental Index tracks the hourly cost of renting an NVIDIA H100, which ranged from $2 to $4.50 per hour at peak demand in early 2026. Silicon Data raised its profile with the CME partnership and has since launched a GPU Forward Curve service for forward pricing visibility.
Ornn raised $33 million in seed funding on June 26 to build financial markets for AI. Its OCPI index, built from printed transaction data rather than indicative quotes, is the basis for ICE's first compute futures product. Ornn operates its own exchange venue for compute risk transfer and has attracted over 400 data center operators and AI companies to its platform.
NATIVX, based in San Juan, publishes the COIL Index with sub-indices for training (COIL-T), inference (COIL-I), graphics (COIL-G), and connectivity (COIL-CO). The energy-normalization feature makes it the only compute benchmark that adjusts for regional power costs, which can account for 30-60% of total GPU operating expense depending on location.
The turning point: compute as a financial asset
Bilateral swaps for compute existed before mid-2026. What changed is that three exchange operators simultaneously decided the market was ready for standardized, regulated, exchange-traded products. That trilateral commitment creates a self-reinforcing dynamic: each announcement validates the asset class for institutional buyers who would not participate in an opaque bilateral market.
The Economist described the trend in June as "how to turn compute into a financial asset." The mechanics are becoming clear: standardized benchmarks enable banks to lend against compute capacity, futures enable operators to hedge GPU rental exposure, and exchange-for-physical delivery enables actual hardware to settle derivative positions. Each layer makes the next possible.
BlackRock CEO Larry Fink has stated that a new asset class will emerge around compute futures. DRW founder Don Wilson, whose firm backed Silicon Data and The Compute Exchange, called compute "the largest commodity in the world" and noted that exponential data center spending growth has been "hampered by the lack of a hedging vehicle."
Regulatory approvals: CFTC review of CME, ICE, and Architect futures. The first to clear sets the benchmark precedent.
Volume depth: which exchange attracts the first 10,000 contracts in open interest will likely become the dominant venue
Index convergence: if OCPI, COIL, and Silicon Data indices converge, the market standardizes early; if they diverge, liquidity fragments
Physical delivery mechanics: ComputeConnect's EFP model is untested at scale. The first failed delivery will test the market's credibility.
Energy cross-hedging: ICE's integration with power and natural gas futures is a structural advantage over standalone compute exchanges
The liquidity fragmentation risk
The most important structural risk in the compute derivatives market is the standard problem of every emerging financial market: too many products competing for too little initial liquidity.
Three exchange operators — CME, ICE, and Architect — are launching similar products referencing different indices in the same window. The same GPU market cannot support three liquid futures contracts. In oil markets, ICE Brent and CME WTI coexist because they reference different crude grades. In compute, the underlying asset (GPU rental capacity) is fungible across all three benchmarks. Traders will gravitate to the most liquid contract, but with liquidity split three ways in the early months, none may reach the volume needed for efficient pricing.
The risk is not theoretical. BGC Group launched a Compute Infrastructure Markets division on June 19 to support secondary trading of compute and memory capacity. FalconX executed the first OTC compute forward trade on May 28. Each new venue adds optionality but dilutes the pool of early adopters.
Ornn CEO Kush Bavaria acknowledged the challenge: the company's strategy is to make OCPI the settlement-grade reference that ICE's futures clear against, betting that exchange sponsorship will concentrate liquidity on one benchmark.
What happens next
The timeline is tight. CME, ICE, and Architect all expect regulatory review to complete by late 2026. If approvals land within weeks of each other, the compute futures market could have three competing products at launch. If one clears significantly earlier, that exchange gets a first-mover window to establish the benchmark.
If liquidity remains fragmented, the intermediate outcome is an OTC market layered on top of thin exchange-traded products. That is better than no market at all, but far from the deep, transparent market that AI infrastructure investors need. The winner-take-most dynamic that played out in oil futures (CME WTI and ICE Brent as the two global benchmarks) will likely repeat, but the time to dominance in compute will be measured in months, not decades.
For institutional investors evaluating the space, the key metric is which exchange clears the first 10,000 contracts per day. Not which one announces first. That exchange will set the global reference price for the era of compute as a financial asset.
