AI Compute Futures (M-E)

Commodity or Gimmick

Aug 11, 2026

In May 2026, the Chicago Mercantile Exchange (CME Group) announced a partnership with market intelligence firm Silicon Data and proprietary trading firm DRW to launch a “first-in-class” futures market for compute power. The aim is “to transform volatile compute costs into a predictable, manageable and tradable asset class”.

If the CME is a hammer, then AI Compute is a nail. Their attempt to index, standardize, and trade a swiftly evolving input technology carries a lot of unknowns and pitfalls.

Call me skeptical.

The Pitch

In principle, the creation of a new secondary market & standard contract specs for an in demand input material would be right up the alley of the CME exchange and other clearing businesses.

The input material in this case is “Compute”. “Compute” refers to the utilization of physical hardware capacity in a datacenter. Compute is required to process data, perform calculations, process AI tasks, and run software applications.

Compute capacity is the output function of all those datacenters and NVIDIA chips being bought up and deployed with hyperscalar CAPEX bonds.

Datacenter = offshore oil rig

Compute = crude oil

CME argues in addition to the usual liquidity and hedging benefits of futures contracts, an AI compute market will create:

A Compute Forward Curve — a forward curve is a snapshot of what the market believes the price of an asset will be at various points forward in time (get it?). CFOs can better plan their upcoming needs if expected costs are more readily known.
Through rigorous standardization and outlier-filtering, the data on overall available compute resources would be synthesized into a daily benchmark. (Like SOFR or electricity pricing I guess) These benchmarks provide a single, standardized hourly rate expressed in United States Dollars per GPU-hour

The Customers

A contract would be a cash settled at expiration based on the GPU reference indices mentioned above. In other words, you either pay or receive cash; not compute resources at expiration.

The natural short side would be hyper scalars, cloud providers and possibly chip foundries trying to sell their compute at the highest price possible. By selling compute futures, a cloud service can lock in a guaranteed return on its hardware investment at the moment of purchase, effectively hedging its revenue stream

The natural long side would be AI foundational labs, enterprise software developers, and researchers looking to use compute to run their training sets and calculate the inference results with the frontier models they produce.

The question would boil down to adoption by these players.

TRADERDADS STORE

The Risks for CME
In my view, the two biggest risks to CME launching a new contract type like AI Compute is something called the Semi-Fungibility Problem and private contract commitments.

Semi Fungibility

Gold is fungible. It’s the same material regardless of source (assuming Good Delivery quality)

Compute is not as simple. An hour of NVIDIA H100 computation in a massive data center in Virginia is not interchangeable with an hour of H100 computation in a small facility in Iceland due to the harness surrounding it. What is the bandwidth attached? Memory available? Cooling system? Vibration and fault tolerance controls?

The differences only grow from there. A Blackwell architecture chip is not the same as the Hopper architecture chip. TPUs are different than GPUs and on and on.

With so many localized variables creating quality variances, I simply cannot see high end frontier model builders omitting these factors in determining their compute choices.

Private Contracts

As a result of this persnickety crowd, to date, we’ve been seeing long-term, fixed-price and bespoke commitments for compute. The largest players in the industry know it’s smart to exit the volatile spot market in favor of predictable costs and resources. They simply achieved it privately. Until these contracts expire or no longer serve a purposes, few if any of these customers will need to employ the futures contracts being designed.

Meta (Facebook) is developing a cloud business under the banner “Meta Compute” to sell its excess AI compute power. The new line of business only moved forward with a sugar daddy of a customer lined up. AI developer Anthropic is set for a compute lease deal worth up to $10 billion proving even new players(buyers & sellers) in the compute market are opting for private contracts over a standardized open market structure the CME will be offering.