In recent weeks, several of the world’s largest derivatives exchanges have announced plans to develop futures and related contracts linked to the costs of artificial intelligence infrastructure — primarily compute capacity and, in some proposals, API-token pricing. The initiatives aim to provide standardized hedging tools for the pronounced volatility in AI-related expenses.
Key announcements and participants
- On May 12, CME Group said it would launch compute-capacity futures in collaboration with Silicon Data, a GPU market data provider backed by trading firm DRW. The contracts are intended to track the rental cost of Nvidia GPUs available in the open market.
- Rival index provider Ornn has disputed some Silicon Data claims; Ornn’s Compute Price Index (OCPI) is based on actual, settled transaction data and was available on the Bloomberg Terminal before CME’s announcement.
- On May 19, Intercontinental Exchange (ICE) announced a parallel initiative and named Ornn as its index partner.
- In January 2026, Architect Financial Technologies — founded by former FTX US president Brett Harrison — said its AX exchange would offer perpetual futures tied to GPU and memory prices, also relying on Ornn reference data.
- Reuters reported on May 28 that the Shanghai Futures Exchange is exploring a different variant: futures linked to AI tokens, the per-unit pricing metric used by companies such as OpenAI and Amazon to charge for model API access.
Why two approaches: hardware versus tokens
U.S. exchanges are primarily targeting the hardware layer — the direct cost of compute represented by GPU and memory rental rates. The Shanghai proposal would instead track how AI providers price access at the application level via tokens or API calls. Both approaches attempt to give participants in the AI supply chain a standardized hedging instrument, but they address different exposures and require different calculation logic.
Volatility and concrete data
According to Ornn, spot rental rates for Nvidia Blackwell-generation GPUs rose 48 percent between mid-February and mid-April 2026, from $2.75 to $4.08 per GPU-hour. Token pricing varies substantially by model and package: OpenAI’s flagship GPT-5.5 model was priced at $5 per million input tokens and $30 per million output tokens for API access as of late April, roughly twice the cost of the previous generation, while older and budget versions remain available at lower rates.
Design and regulatory hurdles
Designing GPU futures poses structural challenges: unlike commodities such as gold or oil, GPU architectures depreciate quickly as new chip generations arrive, so any index following a given hardware generation needs a regular roll mechanism to handle technology transitions. The methodology of index providers — notably whether they rely on actual settled trades versus surveys or indicative quotes — is central to regulatory acceptability and price integrity. Ornn says it executed the first cleared compute swap in December 2025, which it presents as evidence of OCPI’s validity.
Both CME’s and ICE’s proposed contracts await regulatory approvals in the United States. The Shanghai token-based initiative is reportedly at an early stage, with no timetable for regulatory submission; Baocheng Futures estimated in research that China is unlikely to launch compute-capacity futures within the next three to five years.
Liquidity and market depth questions
These potential GPU- and token-based futures would start from a much less mature liquidity base than established commodity markets. It is unclear whether sufficient institutional demand will emerge to sustain active two-sided trading over time. Smaller, crypto-adjacent platforms have already taken early steps, but ICE’s and CME’s entries — with clearing infrastructure and institutional client bases — could materially raise the market’s maturity.
Why it matters for companies and investors
Terry Duffy, chief executive of CME Group, summed up the rationale: “As the backbone of the digital economy, compute is the new oil of the 21st century.” Kush Bavaria, co-founder of Ornn, likewise argued that compute has grown into a trillion-dollar market yet lacks the pricing and risk-transfer infrastructure that supports other major commodity markets.
If these products reach the trading floors, they could enable firms to hedge, forward-price, or transfer the risks associated with AI compute costs, fundamentally changing how AI infrastructure is financed and evaluated. That outcome depends on regulatory approvals, trustworthy indexes based on transaction data, and sufficient liquidity — none of which are guaranteed today.
Companies’ finance chiefs, asset managers and data-center operators should watch these developments closely.
This article is not investment advice or a recommendation.
