# CME Compute Futures

> Source: https://aiwiki.ai/wiki/gpu_compute_futures
> Updated: 2026-08-17
> Fact-checked: 2026-08-17
> Categories: AI Hardware, AI Infrastructure, Data Centers, Finance AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "CME Compute Futures." aiwiki.ai, 17 Aug 2026. https://aiwiki.ai/wiki/gpu_compute_futures
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**CME Compute Futures** are a planned family of financially settled futures contracts tied to Silicon Data benchmarks for hourly rentals of [Nvidia](https://aiwiki.ai/wiki/nvidia) H100 and B200 graphics processors. New York Mercantile Exchange, Inc. (NYMEX), a CME Group exchange, announced two contracts: Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures. NYMEX filed both products with the US Commodity Futures Trading Commission (CFTC) on August 11, 2026 as a voluntary request for formal approval. The CFTC filing register listed both as "Approval Pending (45)" on August 17. CME's October 5 target date remains conditional on that review.[1][2][15][16]

The contracts are intended to turn changes in standardized [GPU](https://aiwiki.ai/wiki/gpu) rental rates into a cash-settled financial exposure. They do not reserve, deliver, or guarantee access to physical computing capacity. As of August 17, 2026, they had not begun trading, so no trading volume, open interest, bid-ask spread, or adoption record existed for the products.[1][2][15][16]

| Field | H100 contract | B200 contract |
|---|---|---|
| Official title | Silicon Data H100 Rental Index Futures | Silicon Data B200 Rental Index Futures |
| Exchange | NYMEX | NYMEX |
| Product code | `GPU1` | `GPU2` |
| Proposed rulebook chapter | 1045 | 1047 |
| CFTC filing status, August 17, 2026 | Approval Pending (45) | Approval Pending (45) |
| Contract size | 730 GPU-hours | 730 GPU-hours |
| Quotation | US dollars and cents per GPU-hour | US dollars and cents per GPU-hour |
| Minimum price movement | $0.01 per GPU-hour | $0.01 per GPU-hour |
| Value of one tick | $7.30 | $7.30 |
| Proposed settlement series | `SD-H100`, US on-demand | `SD-B200`, US on-demand |
| Settlement type | Financial | Financial |
| First listed month | October 2026, planned | October 2026, planned |

## Announcement and status

CME Group and Silicon Data announced their partnership on May 12, 2026. The initial release said the companies planned to introduce compute futures later in the year, pending regulatory review, and that the products would use Silicon Data's daily benchmarks for on-demand GPU rental rates. It did not yet identify the individual contracts or their sizes.[3]

NYMEX supplied those details in Special Executive Report 9785 on August 11. The notice states that the exchange intends to list the contracts effective Sunday, October 4, for trade date Monday, October 5, 2026, pending all relevant US Commodity Futures Trading Commission review periods. CME's dedicated product page gives the same October 5 date and labels it pending regulatory review.[1][2]

The same-day CFTC filing, NYMEX Submission 26-370, is a voluntary request for product approval under Regulation 40.3, not a product self-certification under Regulation 40.2. The CFTC register listed the H100 and B200 products as "Approval Pending (45)" on August 17.[15][16][17] Regulation 40.3 provides an initial 45-day review period. The Commission may extend that period by as much as 45 additional days for novel or complex issues, an incomplete submission, or an untimely response, and it may extend the period for any duration agreed to by the exchange.[19] Neither register entry recorded an approval as of August 17.[15][16]

The distinction between an announcement, a regulatory filing, and a launch is material. Before the target date and a successful regulatory outcome, the products remain prospective. CME and Silicon Data describe possible benefits for AI builders, cloud providers, infrastructure owners, and financial institutions, but those descriptions are intended uses. They are not evidence that a liquid market or an effective hedge already exists.[2][3][17]

## Contract design

Both contracts represent 730 GPU-hours and are quoted in dollars per GPU-hour. A quoted price of `P` therefore corresponds to a reference contract value of `730 x P`. The minimum price change is one cent per GPU-hour, producing the $7.30 tick value published by NYMEX.[1]

Trading is planned on CME Globex, with trades also eligible for submission for clearing through CME ClearPort. The contracts use a first-in, first-out matching algorithm. NYMEX plans to list 36 consecutive monthly contracts, and trading in an expiring month is scheduled to end on the last business day of that month. The block-trade threshold is five contracts with a 15-minute reporting window.[1][17] The proposed rules also allow $0.005-per-GPU-hour increments for inter-commodity spreads executed as simultaneous transactions on CME Globex.[17]

NYMEX classifies the settlement type as financial. The CFTC uses financial settlement and cash settlement as equivalent terms: an expiring position is settled by a cash amount based on an index or price rather than by delivery of the underlying commodity.[8] A holder therefore cannot use the contract to obtain an H100 or B200 instance, a cluster reservation, a service-level agreement, or a place in a [data center](https://aiwiki.ai/wiki/data_center). The instrument addresses an index-linked price exposure only.

Proposed Chapters 1045 and 1047 define each month's final settlement price as the arithmetic average of the applicable Silicon Data on-demand settlement price on every business day during that contract month. The H100 contract uses `SD-H100`; the B200 contract uses `SD-B200`. Both specify the United States as the index geography.[17] The index-derived contract value at final settlement is therefore 730 times that monthly average.[17]

The filing's position-control appendix sets a reportable level of 25 contracts for each product. The initial spot-month limits are 7,500 H100 contracts and 5,500 B200 contracts, effective at the close three business days before the last trading day. The single-month accountability levels are 15,000 and 11,000 contracts, respectively, while all-month accountability levels are 22,500 and 16,500.[18] These are filed terms for products awaiting approval, not evidence of live positions. The public filing does not specify initial or maintenance margin, or a contingency rule for unavailable index observations.[17][18]

## Reference indices

Silicon Data publishes daily rental-price benchmarks for several accelerators. Its public methodology overview says it draws observations from cloud providers, colocation markets, brokered cluster sales, and private rental platforms. It says the data are standardized for machine specifications, rental terms, platform performance, cluster scale, interconnect, and geography, then processed with outlier removal and quality-control checks.[4][5][6]

For [NVIDIA H100](https://aiwiki.ai/wiki/nvidia_h100), Silicon Data publishes a neo-cloud reading under ticker `SDH100RT` and a separate hyperscaler reading. Both are described as standardized hourly rental rates. For [NVIDIA B200](https://aiwiki.ai/wiki/nvidia_b200), it publishes a single standardized series under ticker `SDB200RT` while provider-segment coverage develops.[4][5]

The proposed rulebook chapters now identify the legal settlement series as the US on-demand `SD-H100` and `SD-B200` indices.[17] Silicon Data's public product pages instead display `SDH100RT` and `SDB200RT` ticker labels.[4][5] The filing does not state that the public-page tickers and the legal settlement identifiers are interchangeable, so the labels should not be conflated.

Silicon Data's methodology is also not static. A December 2025 index notice describes a restatement that removed a divisor adjustment and changed its pricing model and provider weights. The company estimated that change would reduce the H100 index by 4 to 6 percent. A provider expansion effective in April 2026 had an estimated negative effect of 3 to 7 percent without restating prior values.[7] These disclosed changes illustrate why index governance, provider coverage, and revision policies matter to a cash-settled contract.

US rules place continuing duties on a designated contract market that lists a cash-settled product. Regulation 38.253 requires the exchange to monitor both the price and the continuing appropriateness of the settlement-index methodology, and to make a good-faith effort to address threats of manipulation, disruption, or distortion.[20] Appendix C to Part 38 identifies acceptable-practice considerations for a third-party cash index, including reliability, commercial acceptance, public availability, timeliness, calculation safeguards, any needed licensing, and information sharing where practicable.[21] Regulation 38.253 imposes continuing duties; Appendix C supplies guidance. Neither is a finding that the pending contracts satisfy those standards. NYMEX's public submission asserts that the reference indices are broad and supported by robust trading in the underlying cash market, but its supplemental market information is redacted under a request for confidential treatment.[17]

## Hardware scope

The products do not treat all computing resources as interchangeable. H100 is a Hopper-architecture data-center GPU, while B200 is part of the Blackwell generation.[13][14] Separate contracts allow the prices of the two rental markets to move independently. They do not imply that one GPU-hour on each device provides the same performance, memory capacity, energy use, networking capability, or value for a particular workload.

That distinction also limits the meaning of the generic phrase "compute futures." The announced CME products cover rental-price indices for two specific [AI infrastructure](https://aiwiki.ai/wiki/ai_infrastructure) components. They do not reference CPUs, custom AI accelerators, inference tokens, electricity, complete servers, or a performance-normalized unit of computing.[1][4][5]

## Intended hedging use

CME presents the products as a possible way for buyers of compute to manage rising rental costs and for capacity providers to manage falling rental revenue.[2][3] In a simplified case, a buyer whose future H100 bill moves closely with the designated H100 index could take a position intended to gain value when the index rises. A provider with index-linked rental revenue could use the opposite exposure. The actual hedge ratio would depend on the size and behavior of the cash exposure.

This mechanism cannot lock in physical availability. A cash payout may offset part of a price increase, but it cannot supply GPUs if a region or provider has no capacity. It also cannot make a rented instance conform to the index's standardized configuration. The CFTC calls the risk that the relationship between a cash price and its futures hedge changes unexpectedly **basis risk**.[8]

Potential sources of basis risk include provider type, region, GPU form factor, host configuration, memory, interconnect, cluster scale, support terms, rental duration, and availability guarantees. Silicon Data says it normalizes several of those variables, but a benchmark cannot be assumed to match every [cloud computing](https://aiwiki.ai/wiki/cloud_computing) invoice or provider revenue stream.[4][6][8]

An August 2026 working paper by Federico M. Bandi and Yinan Su highlights a second difference from many physical commodities. Compute service is non-storable: a GPU-hour left unused today cannot be carried into a future month. The authors argue that ordinary cash-and-carry pricing does not directly connect today's spot rental rate to a later compute-futures price. They also distinguish a financial contract from a term rental, which includes actual access to capacity. Their empirical analysis uses synthetic forward curves because traded compute-futures returns did not yet exist, and its findings are preliminary rather than measurements of the CME products.[10]

## Market context and open questions

CME called the project a "first-in-class" market in its May announcement. That wording is the company's positioning, not proof that it will be the first active venue.[3] Intercontinental Exchange and Ornn announced a separate suite of dollar-denominated, cash-settled GPU compute futures on May 19, subject to regulatory approval.[11] ICE and NATIVX announced another planned family based on an energy-normalized COIL Index on July 1, subject to completion of regulatory processes.[12] Axios also reported a proposed OneChronos compute market awaiting federal approval.[9]

These initiatives use different index concepts and cannot be treated as interchangeable. Their existence shows that compute derivatives are a developing category, while their conditional status makes any ranking by launch order premature.[9][11][12]

Several questions will remain open unless and until the CME products receive approval, begin trading, and produce an observable market record:

- How will corrections, methodology changes, or unavailable index observations be handled?
- What initial and maintenance margins will apply at launch?
- Will trading remain active across several monthly maturities?
- How closely will the futures track the rental exposures of actual buyers and providers?

Until those points have public answers and observed market data, claims that the contracts create reliable price discovery, deep liquidity, cheaper financing, or consistently effective hedges remain projections rather than established outcomes.[1][2][7][9][15][16]

## References

1. CME Group. *SER-9785: Initial Listing of Two (2) Compute Futures Contracts - Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures*. August 11, 2026. https://www.cmegroup.com/content/dam/cmegroup/notices/ser/2026/08/ser-9785.pdf
2. CME Group. *Compute Futures*. Accessed August 15, 2026. https://www.cmegroup.com/markets/energy/power/compute-futures.html
3. CME Group. *CME Group and Silicon Data Partner to Launch First Compute Futures*. May 12, 2026. https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html
4. Silicon Data. *H100 Rental Price Index*. Accessed August 15, 2026. https://www.silicondata.com/products/silicon-index/h100
5. Silicon Data. *B200 Rental Price Index*. Accessed August 15, 2026. https://www.silicondata.com/products/silicon-index/b200
6. Silicon Data. *GPU Rental Price Indices*. Accessed August 15, 2026. https://www.silicondata.com/products/silicon-index
7. Silicon Data. *GPU Index Announcements*. Accessed August 15, 2026. https://docs.silicondata.com/products/gpu-index-announcements
8. US Commodity Futures Trading Commission. *Futures Glossary*. Accessed August 15, 2026. https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/CFTCGlossary/index.htm
9. Scott Rosenberg. *The push to create a futures market for AI compute*. Axios, August 12, 2026. https://www.axios.com/2026/08/12/ai-futures-compute-cme
10. Federico M. Bandi and Yinan Su. *(Early) AI Compute Asset Pricing*. Working paper, revised August 3, 2026. https://arxiv.org/abs/2607.12156
11. Intercontinental Exchange. *ICE and Ornn to Launch GPU Compute Futures Contracts*. May 19, 2026. https://ir.theice.com/press/news-details/2026/ICE-and-Ornn-to-Launch-GPU-Compute-Futures-Contracts/default.aspx
12. Intercontinental Exchange. *ICE and NATIVX to Launch Energy-Normalized Compute Futures Contracts*. July 1, 2026. https://ir.theice.com/press/news-details/2026/ICE-and-NATIVX-to-Launch-Energy-Normalized-Compute-Futures-Contracts/default.aspx
13. Nvidia. *H100 GPU*. Accessed August 15, 2026. https://www.nvidia.com/en-us/data-center/h100/
14. Nvidia. *Supported GPUs*. NVIDIA Multi-Instance GPU User Guide, accessed August 15, 2026. https://docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-gpus.html
15. US Commodity Futures Trading Commission. *Designated Contract Market Products: Silicon Data H100 Rental Index Futures*. Filing 62544, accessed August 17, 2026. https://www.cftc.gov/IndustryOversight/IndustryFilings/TradingOrganizationProducts/62544
16. US Commodity Futures Trading Commission. *Designated Contract Market Products: Silicon Data B200 Rental Index Futures*. Filing 62545, accessed August 17, 2026. https://www.cftc.gov/IndustryOversight/IndustryFilings/TradingOrganizationProducts/62545
17. New York Mercantile Exchange. *NYMEX Submission No. 26-370: Voluntary Submission for Product Approval*. Filed with the US Commodity Futures Trading Commission, August 11, 2026. https://www.cftc.gov/filings/ptc/ptc08112615405.pdf
18. New York Mercantile Exchange. *NYMEX Submission No. 26-370, Appendix B: Position Limit, Position Accountability and Reportable Level Table*. Filed with the US Commodity Futures Trading Commission, August 11, 2026. https://www.cftc.gov/filings/ptc/ptc08112615406.xlsx
19. Electronic Code of Federal Regulations. *17 CFR 40.3: Voluntary submission of new products for Commission review and approval*. Accessed August 17, 2026. https://www.ecfr.gov/current/title-17/chapter-I/part-40/section-40.3
20. Electronic Code of Federal Regulations. *17 CFR 38.253: Additional requirements for cash-settled contracts*. Accessed August 17, 2026. https://www.ecfr.gov/current/title-17/chapter-I/part-38/subpart-E/section-38.253
21. Electronic Code of Federal Regulations. *Appendix C to Part 38: Demonstration of Compliance That a Contract Is Not Readily Susceptible to Manipulation*. Accessed August 17, 2026. https://www.ecfr.gov/current/title-17/chapter-I/part-38/appendix-Appendix%20C%20to%20Part%2038