CME's compute futures will not list on 5 October. On 25 September the exchange reissued its launch notice with one field changed: the initial listing date now reads "TBD", and CME says it "will further advise regarding the specific effective date."
According to The Information, the Commodity Futures Trading Commission wrote to CME on 21 September to extend its review by 45 days, to 9 November, on the grounds that the contracts raise novel or complex issues.
The delay was only possible because of how CME chose to file the contracts. The filing itself, public on the CFTC's website since August, also settles most of what was unclear about how they would work.
Why the CFTC could stop the clock
An exchange can bring a new futures contract to market in two ways, and the difference decides how much say the regulator has.
The common route is self-certification: the exchange certifies that the contract complies with the law, files that the day before, and lists. The CFTC can stay a certification only on narrow grounds. This is how CME and Cboe launched bitcoin futures in December 2017. The CFTC ran what it called a heightened review alongside, but said at the time that the grounds to stay a self-certification were limited, and that none were evident.
The other route is to ask for approval. Here the exchange submits the contract and waits. The CFTC has 45 days, and it can add 45 more if the product raises novel or complex issues. The standard is permissive: the Commission must approve unless the contract's terms violate the law or its regulations. If the period ends without a formal non-approval, the product is deemed approved.
CME took the second route for compute. Its 11 August filing is headed "Voluntary Submission for Product Approval". Forty-five days from 11 August is 25 September, the day CME revised its notice. Another 45 lands on Monday 9 November.
Several reports described 9 November as the new launch date. It is the date the CFTC has to decide by. If the Commission does nothing by then, the contracts are approved by default, and CME would still need to publish a new listing date.
How compute futures settle
The rulebook text is short. For the H100 contract, the settlement value is "the arithmetic average of the Silicon Data H100 Rental Index (SD-H100) on-demand settlement prices as published by Silicon Derivatives Inc. (Silicon Data) for each Business Day during the contract month." The index configuration is listed as Geography: United States. The B200 contract is written the same way.
Averaging changes what it takes to move the price. A contract that settles on a single day's print can be pushed by anyone able to move that one number. A contract that settles on a month's average needs the distortion sustained across about twenty business days, and each day's value is fixed once it is published. An expiry squeeze becomes much harder. Of the two ways the rule could have been written, this is the safer one.
It also matches what a hedger is exposed to. An operator renting GPUs pays across the month, so a contract settling on the month's average price tracks that bill more closely than one keyed to the last day.
Which price
Silicon Data's public page shows two H100 readings, neo-cloud and hyperscaler, nearly three times apart. The filing names neither of them. Settlement uses on-demand prices, configured for the United States, derived from the SD-H100 index. On-demand means capacity rented by the hour with no term commitment, as opposed to reserved capacity, which Silicon Data also tracks. The public page does not say which of its readings, if either, corresponds to that configuration, so the series that would settle the contract is one whose daily values an outside reader cannot currently check.
Silicon Data could change that before launch by publishing the series. It also bears on what the CFTC is reportedly examining.
What the review is weighing
For any cash-settled contract, the question the CFTC asks is whether the settlement price can be manipulated. Its guidance says the price series behind a cash-settled contract should be "reliable, acceptable, publicly available and timely", and that where a third party produces it, the exchange should check that the provider's practices minimise the opportunity or incentive to manipulate it.
CME's filing addresses this in one sentence. It says the index "is sufficiently broad in definition and scope and supported by a robust level of trading in the underlying cash market". The evidence for that sentence sits in an exhibit titled Supplemental Market Information, which CME asked to keep confidential. In the public copy it reads [REDACTED].
Redacting it is normal practice. Market data submitted to a regulator is often commercially sensitive, and the CFTC sees the unredacted version. But it means the claim everything else rests on is the one part of the filing nobody outside can evaluate.
According to the reporting on the letter, the CFTC's concern is how GPU rental prices are set: mostly in private deals across cloud providers, brokers and marketplaces, rather than on a venue where trades print publicly. Silicon Data describes its inputs as observations across cloud providers, colocation markets, brokered cluster sales and private rental platforms. An index built from sources like those depends on what contributors report and how the provider checks it. The more of it rests on executed transactions, the harder it is to move.
What we take from it
We read the delay as the process doing its job. The design is more careful than the launch coverage suggested. Monthly averaging answers the squeeze question well, and hourly on-demand pricing is the right thing for an operator to hedge. CME also chose the slower route, which means the contracts will launch with a formal approval behind them if they launch at all.
That leaves one open issue, the data under the index and whether anyone outside can see it.
If approval arrives alongside a published daily US on-demand series and a stated share of transaction-based inputs, the remaining concern largely falls away and this becomes useful infrastructure. If the contracts are approved by default with the key evidence still redacted, they launch on a benchmark its users have to take on trust, and that is worth knowing before anyone references it in a budget or a loan agreement.
The next fixed point is 9 November, when the extended review ends. By then the CFTC either issues a non-approval or the contracts are approved, formally or by default. CME would then still need to publish a listing date, so the first contract month will not be October. Silicon Data can answer the open question sooner than that, by publishing the US on-demand H100 series before the regulator decides.