CME Group plans to list futures tied to Nvidia GPU rental prices on Oct. 5, creating one of the first standardized U.S. derivatives markets for AI computing capacity. The proposed NYMEX contracts would settle against Silicon Data indexes tracking hourly rental rates for Nvidia’s H100 and Blackwell B200 accelerators, CME said in an Aug. 11 announcement.
The contracts arrive as GPU capacity is increasingly treated as a financeable asset rather than a simple hardware purchase. Data-center operators, cloud providers, lenders and chip buyers are looking for ways to manage the risk that rental revenue and resale values can move sharply as newer processors enter the market.
CME first outlined plans with Silicon Data on May 12 to launch compute futures within 2026. Its later announcement named the two planned products as Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures. Each would track hourly rental pricing for a particular model, avoiding an attempt to turn different chips into a single unit of computing power.
That model-specific approach reflects how uneven the GPU market has become. An H100, designed for the previous generation of large-scale AI training, is not directly interchangeable with the newer B200, while availability, cloud configuration, geography, power costs and networking can all affect the price a customer pays to rent capacity.
Futures markets target compute-rental risk
A liquid futures market could give data-center owners a way to hedge future revenue from GPU rentals and offer large buyers a reference price for expected compute costs. It could also create a more transparent benchmark for lenders financing GPU fleets, where loan underwriting has often depended on private contracts and assumptions about resale values.
The products remain subject to the practical challenge that undermined earlier attempts to create tradable infrastructure markets. Bandwidth futures attracted interest during the telecom boom of the late 1990s but failed to sustain durable liquidity after the sector’s collapse. Enron completed a widely cited first bandwidth trade in December 1999, tied to a monthly DS-3 contract on Global Crossing’s New York-to-Los Angeles route, but later participants including Williams, Dynegy and El Paso exited the market, while platforms such as RateXchange closed.
GPU contracts could face a similar test: whether a benchmark can reflect a fragmented physical market closely enough for both buyers and sellers to use it. Silicon Data’s indexes track rental prices daily, while the CME contracts would reference hourly rental prices. That gives the market a clearer underlying measure than an index based solely on advertised cloud rates, but it also leaves the contracts exposed to changes in chip specifications, supply arrangements and demand for particular AI workloads.
ICE announced a competing effort on May 19 with data provider Ornn to develop futures based on the Ornn Compute Pricing Index, or OCPI. Ornn’s coverage spans enterprise accelerators including the H100 and H200, along with consumer hardware such as Nvidia’s RTX 5090. Ornn has described OCPI as drawing on executed transactions rather than posted offers, and the index appeared on Bloomberg terminals in April 2026.
Bilateral and on-chain deals have arrived first
Trading tied to compute-price benchmarks has already begun outside listed futures markets. FalconX said on May 27 that it completed what it described as the first over-the-counter compute forward price swap, referencing Ornn’s H100 forward price. Robert Leshner, co-founder and chief executive of Superstate, was named as the counterparty.
Polymarket said on June 2 that it executed its first institution-sized on-chain block trade between FalconX and AneraLabs. The transaction settled against Ornn’s OCPI index, was recorded on Polygon, and had a value in the six figures in U.S. dollars, according to Polymarket.
Those transactions are small beside conventional commodity derivatives markets, but they show how compute-price references are reaching both institutional over-the-counter structures and blockchain-based settlement systems. The immediate use case is less about speculative trading than transferring exposure to a costly and volatile operational input.
Kalshi added another layer on July 14 by publishing an AI compute forward curve covering Nvidia’s B200, H200 and A100 chips. The platform’s other compute-related contracts also referenced the H100 and RTX 5090, according to the supplied material. Forward curves can show how market participants expect rental prices to change across future dates, though early curves in a thin market may be sensitive to limited trading activity.
Financing depends on residual values
The move toward derivatives coincides with a much larger financing effort around AI infrastructure. Nvidia said on Aug. 10 that it had signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR intended to mobilize more than $500 billion in third-party capital for AI data centers, chip plants and associated power infrastructure.
Nvidia has presented GPUs as infrastructure that could be pledged as collateral. In later public descriptions, Nvidia Chief Executive Jensen Huang discussed a case-by-case residual-value support mechanism of up to 25% on certain financings. Electron Economics’ review of Nvidia’s Aug. 10 release found that the original announcement did not explicitly use the terms “residual value” or “backstop.”
That distinction matters for lenders because GPU values have already shown substantial depreciation. Silicon Data and GPUSmith data cited in the supplied material showed H100 units that were priced near $40,000 at the end of 2023 trading in a roughly $12,000 to $22,000 range by mid-2026, with some auctions reaching as low as $8,200. Financing structures that assume stable collateral values would therefore depend heavily on the precise terms of any support arrangement, maintenance requirements and the remaining commercial usefulness of the hardware.
Credit markets have also reacted to the debate over AI infrastructure funding. ICE Data Services showed Nvidia’s five-year credit default swap spread rose 14 basis points to an intraday 82 basis points on July 27, the largest intraday increase since the contract began actively trading in November 2025, before later retreating. An Investing.com analysis placed the spread near 77.5 basis points on Aug. 10.
China develops spot-market infrastructure
China has been building compute-trading infrastructure on a separate track. The Shanghai computing power trading platform began trial operations in April 2023, released version 2.0 in December that year and became the Shanghai node of the national computing power platform in May 2025.
The China Academy of Information and Communications Technology said the national platform reached full interconnection at the China Computing Power Conference in August 2025, linking ten provincial-level subplatforms. CSI Commodity Index Co. followed on Dec. 24, 2025 with an intelligent-computing supply index series containing 16 indexes, including national, regional and computing-hub measures.
Shanghai’s municipal government added compute futures to its policy agenda in a document issued on May 28 and made public on June 2. The document called for research and development preparation for power futures and compute futures. Separate reporting described an early Shanghai Futures Exchange concept linked to AI token usage rather than hourly GPU rental prices, though no regulatory approval timetable had been set.
The emerging market is therefore forming around several competing definitions of compute: hourly GPU access, forward delivery prices, transaction-based indexes and potential token-based measures. CME’s planned H100 and B200 contracts would give the market its clearest standardized U.S. test yet, while also exposing whether demand for hedging can keep pace with the rapid replacement cycle of AI hardware.
Explore how AI copy trading can complement GPU futures when managing algorithmic trading and AI infrastructure risk.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
