Behind the rapid growth of artificial intelligence lies an enormous demand for computing power. At the heart of the infrastructure supplying that power are specialised computer chips known as graphics processing units, or GPUs, which are used to train and run AI models.
The cost of accessing that processing power has become significant enough that demand is emerging for something familiar to commodities markets: a derivatives market that allows buyers and sellers to manage the risk of future price changes.
Several US exchanges have announced plans to launch compute futures in partnership with companies that track prices in the underlying market. The first of these is scheduled to begin trading in October, subject to regulatory review.
Relatedly, the Commodity Futures Trading Commission in August issued a request for comment on compute derivatives, saying it was the “first step” towards establishing “clear rules of the road," and asking for feedback on the underlying cash market, liquidity, market oversight and manipulation.
So, what exactly is being developed, and can computing power really become a standardised commodity that can be bought and sold on exchanges?
What is compute?
At its simplest, compute is the processing power and hardware infrastructure that machines need to train and run AI models. Much of the discussion around compute markets concerns the capacity supplied by GPUs, which are particularly well suited to training and running AI models.
A company does not necessarily have to own these chips. It can rent access to GPU capacity from cloud providers and specialist infrastructure companies. That makes it possible to put a price on the use of a particular type of processor, commonly expressed as the cost of renting one GPU for one hour.
Not all GPU-hours are equal, however. The price can vary depending on the type and configuration of the GPU, its location, and the terms under which capacity is rented. That variability is one of the central challenges in creating a derivatives market for compute.
What are compute futures and how do they work?
The basic idea is to create a futures contract whose value is tied to the cost of renting computing capacity.
Contracts announced by CME Group and Silicon Data, a GPU market intelligence and benchmarking company, in August illustrate the model.
CME plans to list futures based on Silicon Data's H100 and B200 Rental Indices, which measure the hourly rental cost of two models of Nvidia GPUs. Each contract represents 730 GPU-hours, roughly equivalent to one month of continuous capacity for a single GPU. At the time of writing, CME's planned launch on 5 October remains subject to regulatory review.
ICE is taking a different approach, working with two partners to list two types of compute futures. One set of contracts will reference Ornn's Compute Price Index, which tracks live-traded spot prices for GPU compute across major hardware types. ICE has said the contracts could reference indices covering GPUs including the H100, H200 and B200.
The other set is based on the COIL index developed by NATIVX, which takes a different approach by normalising compute prices for energy costs, helping make prices from locations with different power costs more comparable.
The most recent entrant is Nodal Exchange, which announced a partnership with Compute Desk, a provider of GPU rental price indexes, to launch its own set of compute futures. Although Nodal is smaller than CME and ICE, it has more than half of the open interest in the US electricity futures market, and that gives users a way to capture correlations between power and compute.
Architect Financial Technologies also plans to offer futures and options on compute costs tied to multiple GPU vendors and models through its American Innovation Exchange, a newly-established exchange.
The contracts will be cash-settled, rather than physically delivered. A trader holding such a contract to expiry would not receive racks of computer equipment or the right to use capacity in a data centre. The contract would instead settle financially against a reference price.
That makes the quality of the reference benchmark crucial. A futures hedge is only useful to a commercial participant if movements in that benchmark bear a sufficiently close relationship to the price the participant actually pays or receives for compute.
Why create a futures market for compute now?
The immediate answer is the rise in artificial intelligence. The expansion of AI has made computing capacity an increasingly important operating input. Companies developing AI models need access to GPUs, while cloud providers, specialist "neoclouds" and data-centre businesses are investing in the infrastructure supplying that capacity.
Agentic AI is especially important to this trend. Rather than answering a single prompt, agentic AI can carry out multiple steps to complete a task, potentially increasing the amount of computing capacity required.
The rapid rise in demand for AI coincides with a scramble to build data centres and outfit them with the latest generation of GPUs. In the US, investment in information-processing equipment reached an annualised $752 billion in Q2 2026, overtaking residential investment, according to Bureau of Economic Analysis data.
Goldman Sachs estimates that global AI investment will exceed $1 trillion in 2026 while McKinsey & Company projects that $6.7 trillion in capital spending on data centres will be required worldwide by 2030 to keep pace with demand for computing power. McKinsey breaks this down as $5.2 trillion for AI workloads and $1.5 trillion for traditional IT.
Against that backdrop, the cost of accessing GPU capacity can vary considerably, creating an incentive for more transparent pricing and tools to manage future price risk.
"For years, two companies buying the exact same GPU capacity could pay wildly different prices with no way to know who got the better deal. They will now have a benchmark to check that against," said Carmen Li, CEO of Silicon Data.
"Compute futures give the market something it's never had: a public, tradable reference price for the resource every AI system runs on...that turns compute from something enterprises negotiate blindly into a market they can actually plan around."
Who would trade compute futures?
An AI developer expecting to rent large amounts of GPU capacity in future could take a futures position that gains value if the relevant compute benchmark rises, helping to offset a higher rental bill.
A provider investing in GPUs faces the opposite risk: it has invested in hardware whose earning power depends partly on future rental rates. It could use futures to reduce its exposure to falling prices.
This creates a need for hedging on both sides of the market, although neither would necessarily get a perfect hedge.
An AI company might expect to need a particular GPU configuration in London six months from now, while its futures contract settles against a broader index of on-demand capacity, meaning its own rental costs might move differently from the benchmark.
That difference is basis risk: the possibility that the price being hedged and the futures price do not move together closely enough for the hedge to offset the exposure fully.
Financial firms could also trade compute futures without needing GPU capacity themselves, taking on price risk and potentially adding liquidity to the market.
How is the price of compute determined?
There is no single universal market price for compute. Prices vary across different GPUs, providers, locations and rental arrangements, so benchmark providers are trying to turn those disparate prices into representative reference rates.
Silicon Data says its methodology collects GPU rental prices and makes adjustments designed to compare different rental arrangements on a more consistent basis. Ornn takes a transaction-based approach, with its Compute Price Index built from traded GPU compute prices.
The aim is not to make every GPU rental identical. It is to create a trusted reference price against which a futures contract can settle.
The challenge is to make the benchmark broad and robust enough to command confidence while ensuring it still reflects the economic exposure participants actually want to hedge.
How do you turn compute into something standardised enough to trade?
This is perhaps the most difficult question facing the emerging market. A Nvidia H100 is not a Nvidia B200. Even within a particular processor model, price can depend on how the hardware is configured, where it is located and the terms under which the capacity is available.
Another issue is that the cost of compute varies by the location of the data centre and the cost of power at that location. In the US, power prices vary considerably across the country because of differences in weather, energy inputs and other factors.
One solution is to create separate benchmarks for individual types of GPU, but that presents another trade-off: greater precision could mean spreading trading across more contracts, making it harder to concentrate liquidity.
If a benchmark is standardised too aggressively, the futures price may bear too little relation to the compute users actually buy. Standardised too little and the market risks splintering.
Compute does not have to become completely uniform to support a futures market. The harder question is whether the differences can be reduced to a manageable basis risk that commercial users are willing to accept.
What could stop a compute futures market from taking off?
A number of hurdles need to be overcome. One is the benchmark. Can the industry establish a representative and robust price for an underlying market that is fragmented across different types of hardware, providers and transactions?
Another is basis risk. Even if the benchmark is reliable, it needs to move closely enough with the actual compute costs faced by businesses for a futures hedge to be useful.
Liquidity presents another hurdle. A successful market will require commercial participants with genuine price exposure as well as intermediaries willing to make markets. There is also a circular problem: hedgers may hesitate to use a futures contract until it is liquid, while liquidity may struggle to develop until hedgers use it.
And then there is obsolescence. Today's heavily used GPU could eventually be displaced by newer hardware, forcing the market to migrate towards new benchmarks and potentially splitting liquidity between generations.
None of those obstacles necessarily prevents a market developing, but together they help explain why strong demand for AI computing power does not automatically translate into a successful futures market.
Could compute become an established derivatives market?
Conversely, compute does not need to become perfectly fungible for a futures market to succeed. Commodity markets already manage differences in grade, location and delivery. What matters is whether the industry can establish trusted reference prices around which enough commercial risk and liquidity can concentrate.
If it can, compute futures could give AI companies and infrastructure providers a way to manage price risk and create something the compute economy currently has only in limited form: a transparent market view of what computing capacity may be worth in the future.
For now, however, the market remains at an early stage. The exchanges are arriving. The benchmarks are being built. The harder test is whether a liquid market will follow.
