# CME sets Oct. 5 launch for H100 and B200 compute futures, pending review > Source: > Published: 2026-08-12 12:04:52.434198+00:00 # CME sets Oct. 5 launch for H100 and B200 compute futures, pending review - CME plans to list Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures on October 5, pending regulatory review. [[1]](https://www.cmegroup.com/markets/energy/power/compute-futures.html) - The contracts are designed to settle financially against GPU rental-rate benchmarks, giving users exposure to price changes rather than access to physical GPUs or reserved cloud capacity. [[2]](https://www.silicondata.com/blog/gpu-futures) - CME identifies AI builders, cloud providers and institutional investors as potential users, but has not publicly named liquidity providers or disclosed final position limits in the materials reviewed. [[3]](https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html) CME Group plans to launch two futures contracts tied to the rental cost of H100 and B200 GPUs on October 5, creating a standardized financial instrument for a narrow slice of the market that supplies AI training and inference. The launch remains subject to regulatory review. [[1]](https://www.cmegroup.com/markets/energy/power/compute-futures.html) The contracts will reference Silicon Data’s H100 and B200 rental indexes and settle in cash. A futures position will transfer the financial difference between the contract price and the reference value; it will not deliver a GPU, server or guaranteed block of cloud capacity. [[2]](https://www.silicondata.com/blog/gpu-futures) ## Two GPU-specific benchmarks CME’s August 2026 product notice names the products Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures. CME’s compute-futures page identifies October 5 as the planned first trading date. [[1]](https://www.cmegroup.com/markets/energy/power/compute-futures.html) Silicon Data describes its benchmarks as measures of on-demand rental rates for specific GPU models. The company says the planned contracts are cash-settled and denominated in dollars per GPU-hour, with separate products for different hardware generations rather than one blended measure of computing power. [[2]](https://www.silicondata.com/blog/gpu-futures) That design gives the contracts a more specific reference than a broad AI-infrastructure index, but it also narrows the hedge. A firm using H100 futures would be hedging the benchmarked H100 rental rate, not the full cost of operating an AI cluster. Power, cooling, networking, storage, staffing, software, financing, taxes, data transfer and regional availability remain outside the contract unless they are reflected indirectly in rental prices. ## What the futures can and cannot hedge CME says the intended users include AI builders, cloud-service providers, financial institutions and institutional investors. CME executives have also described possible participation by asset managers, banks, energy firms, hedge funds and professional trading firms. [[3]](https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html) An AI company could use a long futures position to offset part of a future increase in rental costs. A GPU host could use a short position to offset falling rental revenue. Those hedges would still be imperfect if the firm’s actual machines, location, utilization rate, contract terms or availability differ from the benchmark. The product does not reserve capacity. A successful hedge could offset part of a price increase while the buyer still struggles to obtain GPUs, waits for a cloud provider or pays a regional premium. Conversely, a firm with physical access to GPUs may face a basis mismatch if its actual rental economics differ from Silicon Data’s reference rate. Cash settlement avoids the logistics of delivering computing capacity at expiration. Physical delivery would require agreement on hardware configuration, location, networking, uptime, security and software environments. The trade-off is that the futures market remains a financial layer above the operational market: it can transfer price risk, but it cannot manufacture chips or add power capacity. ## Regulatory status and the liquidity question CME and Silicon Data announced the market in May as a launch planned for later in 2026, pending regulatory review. CME’s current product page still carries that qualification. The CFTC’s public designated-contract-market product page reviewed for this report did not show the two compute contracts among the displayed certified products. [[4]](https://www.cftc.gov/IndustryOversight/IndustryFilings/TradingOrganizationProducts?Cate=&Type=Future&page=1) CME’s August notice confirms that the exchange has prepared a formal contract-specification notice, SER-9785. The materials reviewed identify the two indexes and planned launch date, but do not identify a named market-making group for the products. CME’s published liquidity-provider directory lists programs for several existing futures categories and does not list a compute-futures program. [[5]](https://www.cmegroup.com/markets/market-maker-directories.html) The materials reviewed also do not provide a confirmed position limit, accountability level, reportable-position threshold, minimum price fluctuation, contract multiplier or initial listing schedule for the two contracts. CME normally publishes those details in product notices and position-limit tables. Until those terms are available, users cannot determine the notional value of one contract, the margin required to hold it or the size at which a position could trigger reporting or exchange scrutiny. [[6]](https://www.cmegroup.com/market-regulation/position-limits.html) Early liquidity will matter because the underlying market is fragmented. Silicon Data’s index can provide a reference price, but a liquid futures market also needs two-sided participation, clearing access and enough commercial hedging demand to support trading beyond initial speculation. CME has the clearing and electronic-trading infrastructure; it has not disclosed which firms will provide continuous bids and offers for the new contracts. ## Companies mentioned ## Further sources [[1] CME’s compute-futures page gives the planned October 5 launch date, while CME’s… ↗](https://www.cmegroup.com/markets/energy/power/compute-futures.html) [[2] Silicon Data describes the planned products as standardized, cash-settled contr… ↗](https://www.silicondata.com/blog/gpu-futures) [[3] CME and Silicon Data identify AI builders, cloud providers, financial instituti… ↗](https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html) [[4] CME and Silicon Data continue to describe the launch as pending regulatory revi… ↗](https://www.cftc.gov/IndustryOversight/IndustryFilings/TradingOrganizationProducts?Cate=&Type=Future&page=1) [[5] CME’s liquidity-provider directory lists programs across several existing asset… ↗](https://www.cmegroup.com/markets/market-maker-directories.html) [[6] CME says contract-specific position limits and reportable levels are published … ↗](https://www.cmegroup.com/market-regulation/position-limits.html)+1 more The stories that matter, in one email. Free — unsubscribe anytime.