CME and ICE split over the benchmark that will settle GPU futures

Why it matters

Compute futures need a trusted reference price before they can become useful hedging tools. The benchmark that wins early exchange support, lender use and trader confidence could shape how GPU capacity is financed, hedged and valued.

CME and ICE are setting up a standards fight over how GPU prices get settled. CME plans compute futures tied to Silicon Data's benchmarks, while ICE plans contracts that reference Ornn's pricing index, with both offerings pending regulatory approval.

The contest is less about who can name a contract first than which reference price becomes acceptable to traders, lenders and data-center buyers. Compute futures need a settlement value that market participants trust. CME is aligning with Silicon Data, a GPU benchmark provider. Intercontinental Exchange is aligning with Ornn, a compute market-infrastructure startup that publishes a pricing index for GPU compute.

Two routes to a compute contract

Axios reported that, pending regulatory approval, CME plans to launch compute futures tied to Silicon Data's benchmark, and Intercontinental Exchange plans GPU compute futures tied to Ornn's pricing index. The setup creates two potential settlement references for the same emerging market.

That reference choice matters. Benchmarks can be constructed from rental quotes, private transactions or executed trades, and each basis can produce different settlement values. The market-structure problem is how to turn heterogeneous GPU capacity into a number sturdy enough to support margining, hedging and collateral decisions.

Silicon Data's benchmark case

CME is aligning with Silicon Data. The company publishes GPU indices and a forward curve designed for pricing and settling GPU-linked futures, swaps and structured products. For CME, that framework offers a benchmark aimed at standardizing how varied compute contracts show up in a single settlement reference.

ICE's Ornn bet

ICE is aligning with Ornn. The startup publishes a pricing index for GPU compute and is building market infrastructure around it. Ornn has one distribution advantage before any contracts launch: Axios reported its benchmark is already integrated with Bloomberg Terminal and other data providers, giving traders a way to see GPU prices through tools they already use. Axios also noted that lenders can use Ornn for benchmarking, while buyers and sellers of compute can use it to hedge.

The standard decides the market plumbing

Different benchmarks can create different settlement outcomes even when they describe the same broad market. Rental quotes, private cluster deals and executed spot trades can diverge because of contract length, geography, networking, power availability, utilization risk and the buyer's credit. If ICE and CME launch with incompatible references, customers may have to choose which version of GPU pricing they treat as the market.

That choice can spill into financing. A lender valuing a GPU-backed loan needs a reference for collateral and residual value. A data-center operator reserving capacity needs a forward view of compute costs. An AI lab locking in future training capacity needs a hedge that matches its exposure closely enough to justify margin and basis risk. The benchmark that gains early acceptance could become the language those parties use in term sheets, risk models and settlement clauses.

Compute is a difficult commodity

Axios cited a Goldman Sachs estimate that roughly $7.6 trillion will be invested globally from 2026 through 2031 in compute, power and data centers. That spending creates price exposure for cloud providers, AI labs, data-center owners, lenders and investors.

Compute still resists the usual commodity playbook. Axios notes that GPU capacity cannot be stored, so unused compute disappears. New Nvidia chip generations can improve price-performance and change the value of older chips, forcing any benchmark to track an asset that depreciates as hardware improves.

Regulatory approval remains the gate for both exchange offerings. Until then, the contest is being fought through data definitions, customer integrations and the credibility each benchmark can build before the first contracts trade. The early winner may determine how Wall Street quotes GPU risk long before compute futures become liquid.