CME Delays Nvidia GPU Rental Futures Again, to January

CME Group has agreed to delay its planned futures on Nvidia GPU rental prices for another 60 days, taking any approval into January, The Information reported on 6 November, citing a letter the exchange sent to the Commodity Futures Trading Commission on 4 November. It is the second delay for contracts first slated to trade on 5 October.

Silicon Data’s published neo-cloud H100 rental reading, ticker SDH100RT, moved from $2.81 per GPU-hour on 7 October to $2.97 on 5 November, up 5.7% over the period by our arithmetic. The H100 contract would settle on Silicon Data’s H100 Rental Index, the NYMEX filing shows.

Business Pill · Financial Conditions

A short explainer on financial conditions: how the price and availability of money and risk transfer shape what companies can build. It teaches the general idea only and does not discuss CME, Silicon Data or GPU futures.

What Was Reported

According to The Information, the CFTC asked CME to voluntarily delay the introduction of the product, and the exchange agreed, writing that it wanted to give the commission “further opportunity to consider the issues raised as a result of the submission”. CME also wrote that its agreement is “conditioned on other commission registrants not self-certifying, obtaining commission approval, or otherwise listing comparable contracts during the pendency of this extended review period”, the report says.

The Information reports that companies and individuals submitted dozens of letters during the comment period, many urging the CFTC to set conditions the underlying indices must meet before any futures are approved. It adds that Intercontinental Exchange has announced plans for GPU futures but has not set a date. We read The Information’s briefing; we did not see CME’s 4 November letter itself, and we did not contact CME or the CFTC.

The First Delay, in the Regulator’s Words

The CFTC’s Division of Market Oversight extended its initial review of NYMEX’s submission on 21 September, its letter shows, for an additional 45 days “until the end of November 9, 2026”. The letter cites “the need for additional time to analyze the Products in light of novel or complex issues” and a pending Request for Comment on listing compute derivatives, whose comment period closed on 20 October.

If the new 60 days run from the end of that review on 9 November, they end on 8 January 2027, 95 days after the original 5 October trade date, by our arithmetic. The Information gives only “January”.

What the Contracts Are

CME announced on 11 August that it would list “Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures” on NYMEX, tracking indexes of hourly GPU rental costs published by Silicon Data, its release says. NYMEX filed them for approval under CFTC Regulation 40.3(a), a voluntary submission rather than a self-certification, the filing shows.

Each contract covers 730 GPU-hours, roughly a month of one GPU’s rent; prices are quoted in dollars per GPU-hour with a minimum move of $0.01, worth $7.30 a contract; settlement is financial, and monthly contracts would be listed 36 months out, the NYMEX filing says. The H100 contract’s floating price is “the arithmetic average of the Silicon Data H100 Rental Index (SD -H100) on-demand settlement prices” for each business day of the month, the filing says.

At the neo-cloud reading of $2.97 on 5 November, 730 GPU-hours come to about $2,168, by our arithmetic; the filing does not say whether that reading is the settlement series.

The Index Behind It

Silicon Data describes SDH100RT as a daily, like-for-like H100 rate drawn from neo-cloud providers and GPU marketplaces, with hyperscaler prices published as a separate reading, on its index page. Its published 30-day table runs from $2.81 on 7 October to $2.97 on 5 November, with dips along the way, such as $2.85 to $2.81 on 15 October. The same page’s FAQ gives a current price of $2.53 per GPU-hour; we could not reconcile that with the daily table and use the dated table readings.

CME’s Pete Keavey said in August that the contracts would “turn compute into a standardized, tradable commodity”, and Silicon Data chief executive Carmen Li said two companies buying the same capacity “could pay wildly different prices”.

Connecting the Coverage

Related coverage of GPU rental economics: SpaceX rents Nvidia GPUs to Anthropic at about $5.27 an hour, as we reported, and Bernstein puts Nebius at $11.1 million of revenue per megawatt, as we reported.

Nvidia’s guarantees sit against six-year chip lives at AI clouds, as we reported. For how compute futures work, see our explainer. On Friday morning, AI stocks were winning back part of their losses, as we reported.

The Structural Read

As we read the CFTC’s 21 September letter, the regulator tied the first extension to novel or complex issues and to its Request for Comment on compute derivatives; the second delay, as The Information describes it, came at the CFTC’s request.

As we read The Information’s account of CME’s letter, the exchange conditioned its agreement on no other registrant listing comparable contracts during the extended review.

As we read the NYMEX filing, each H100 contract would cover 730 GPU-hours and settle in cash on a monthly average of Silicon Data’s H100 index; the filing does not say which segment reading that average uses.

Three Implications

FOR GPU RENTERS CME’s H100 and B200 rental futures are not trading yet; The Information reports approval is now pushed into January.

FOR INDEX PROVIDERS The Information reports that many comment letters urged the CFTC to set conditions the underlying indices must meet before any futures are approved.

FOR READERS Whether and when the contracts are approved is not established; the 8 January date is our arithmetic.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Supercycle.

The analysis says: “The job of any operator, investor, or builder is to know which layer they are operating inside and which layer the binding constraint is rotating toward next.”

As we read it, a futures market on GPU rent would put a public price on the compute layer itself. The analysis does not discuss CME, Silicon Data or GPU futures.

What Is Not Established

Whether the CFTC approves the contracts, on what conditions, and the date trading starts are not established. We did not see CME’s letter, the comment letters or any CFTC statement on the second delay; the details of it come from The Information. The 8 January date is our arithmetic, not a date anyone has announced.

The Bottom Line

CME agreed to push its Nvidia GPU rental futures back another 60 days at the CFTC’s request, The Information reported, after a first 45-day extension in September. Silicon Data’s neo-cloud H100 reading rose 5.7% between 7 October and 5 November, to $2.97 an hour.