What Are Compute Futures?

Compute futures are financial derivatives contracts tied to the price of GPU rental capacity. Like oil futures or interest rate futures, they allow one party to lock in a price today for compute that will be delivered - or settled in cash - at a future date. The buyer hedges against rising GPU costs; the seller hedges against falling demand or price.

In October 2026, CME Group and Silicon Data are launching the first standardized compute futures contracts on NYMEX - specifically H100 and B200 rental index futures. ICE and Ornn have announced a competing product covering five GPU models. These launches mark the formal financialization of GPU compute: the moment a raw infrastructure resource gains a derivatives market.

Why Compute Futures Are Forming Now

Three conditions produced this market. First, the GPU rental market reached sufficient scale: approximately $52 billion in 2026, projected to reach $198 billion by 2031. Second, prices became volatile enough to create a hedging need. H100 rental rates have moved substantially since Blackwell arrived, creating risk for AI companies planning multi-month compute budgets. Third, a standardized reference price emerged - the Silicon Data GPU Rental Index and the Ornn Compute Price Index (OCPI) - giving futures markets a credible settlement benchmark.

The parallel to energy markets is instructive. Oil futures emerged once oil became a standardized commodity traded across many buyers and sellers, with prices observable in a liquid spot market. Compute is following the same path, compressed into years rather than decades.

How Compute Futures Work

CME compute futures are cash-settled contracts - no GPUs change hands at expiration. Instead, at contract expiration, positions are settled in dollars against the relevant GPU rental price index. Each contract represents one month’s rental cost for the underlying GPU at the index price.

The settlement index is the Silicon Data GPU Rental Price Index - a daily measure of actual GPU rental rates across cloud providers, weighted by transaction volume. The index captures spot and on-demand pricing, not reserved or committed capacity.

For an AI company, this creates a practical hedging mechanism: if you expect to need 1,000 H100s for three months starting December, you can buy H100 futures today to lock in the rental rate. If prices rise, your futures gain offsets higher actual rental costs. If prices fall, you give back the futures gain but pay less for actual compute.

CME vs. ICE: Two Competing Products

Two major exchanges have announced compute futures, each with a different index methodology and model coverage.

The key methodological difference: Silicon Data’s index captures rental offers and transactions across cloud providers. Ornn’s OCPI is built on printed transactions only - actual executed deals, not offers - which Ornn argues produces a more accurate settlement benchmark for institutional participants.

The Forward Curve: What the Market Implies

Before the formal futures market launched, Silicon Data published indicative forward curve data. The H100 curve shows approximately 13% backwardation over 36 months - meaning futures prices are below today’s spot price, implying the market expects H100 rental costs to fall. The B200 curve shows approximately 8% backwardation - shallower, reflecting the newer supply cycle and less certain depreciation trajectory.

Backwardation in commodity markets typically means the spot resource is more valuable today than in the future - often because supply is expected to increase. In GPU compute, this reflects anticipated supply expansion as Blackwell and Rubin architectures add capacity, and the expected generational displacement of H100 by B200 and beyond.

Who Trades Compute Futures?

The primary participants in compute futures will likely be:

• AI companies and hyperscalers hedging compute cost exposure for multi-month training runs
• GPU cloud providers (CoreWeave, Lambda, Crusoe, Nebius) hedging revenue from reserved capacity commitments
• Quantitative trading firms arbitraging spot vs. futures price relationships
• Hedge funds and macro investors taking directional views on AI infrastructure demand
• ETF issuers - the Roundhill Compute ETF (GPUX) and similar products may reference compute price exposure
• Private credit lenders using futures to hedge collateral value on GPU-backed loans

Market Access

CME compute futures will trade on NYMEX through registered futures commission merchants (FCMs) - standard institutional futures brokers. Access requires a futures trading account with a registered broker. Retail access may emerge through ETFs and structured products over time, but the initial market is institutional.

ICE contracts will be accessible through ICE’s standard brokerage infrastructure, with index data distributed via Bloomberg Terminal.

What This Market Does Not Cover

Compute futures, in their initial form, cover GPU rental price risk only. They do not cover hardware ownership or resale value. They do not cover software, networking, or storage costs. They do not cover specific providers or regions - they settle against a composite index. A company that has a fixed-price committed contract with a specific provider gets no basis risk protection from an index-settled futures contract.

These limitations are features of early-stage commodity markets, not permanent constraints. Energy derivatives started with crude oil and expanded over decades to natural gas, electricity, emissions, and weather. Compute derivatives will expand as the data infrastructure matures.

Frequently Asked Questions

What are compute futures?

Compute futures are cash-settled financial contracts tied to GPU rental price indexes. CME Group and Silicon Data plan to launch H100 and B200 futures on October 5, 2026, on NYMEX.

How are compute futures settled?

At expiration, CME compute futures settle in cash against the Silicon Data GPU Rental Price Index. No GPUs change hands. The settlement value equals the index price at contract expiration.

Who can trade compute futures?

Institutional participants with futures trading accounts through registered FCMs. Access requires a standard futures brokerage relationship. Retail participants may gain access through ETFs over time.

What is the difference between CME and ICE compute futures?

CME (with Silicon Data) covers H100 and B200, launching October 5, 2026. ICE (with Ornn) covers five models including H200, B300, and RTX 5090, with transactions-based settlement distributed on Bloomberg.