GPU compute futures explained: CME, ICE and the price of an H100

GPU compute is becoming something you can trade. CME Group lists H100 and B200 rental futures from 5 October 2026, built on Silicon Data's GPU rental price indices. ICE and Ornn have announced GPU compute futures of their own. This post explains what GPU futures are, why they are arriving now, who would use them, and what they still depend on.

Opengrid is pre-launch and isn't an exchange. We are building the on-demand routing layer, and we cover it at the end.

What a GPU futures contract is

A futures contract is an agreement to buy or sell something at a set price on a future date. Oil, wheat and electricity have traded this way for decades. A GPU rental future applies the same idea to the hourly price of renting a GPU.

The contracts don't hand you a GPU. They settle against a price index, a published reference price for renting a given chip. CME's H100 and B200 contracts are built on indices from Silicon Data, a company that publishes GPU rental price indices. If the index rises, a buyer of the future gains. If it falls, the seller gains.

Why now: the price moves too much to ignore

A hedge is only worth having when the price moves. GPU prices now move by double digits in weeks:

When	What happened
Jun 2025	AWS cut its H100 on-demand price 44% in a single announcement
Jun to Jul 2025	Other providers followed with cuts of 20 to 25% in single moves
Dec 2025 to Jan 2026	Silicon Data's H100 rental index rose from $2.00 to $2.20 in four weeks, up 10%
Oct 2025 to Mar 2026	One-year H100 contract prices rose from $1.70 to $2.35 per GPU-hour, up about 40%

And each chip moves on its own. In one stretch, H100 rose 10% while A100 and B200 stayed flat. For a team with a large training budget, or a GPU cloud with racks of capacity to fill, that is real financial risk. We cover the full price history in how GPU rental prices moved in 2025 and 2026.

Who would trade GPU futures

Three groups have a reason to use these contracts:

• AI teams locking in next quarter's training cost, so a price spike doesn't blow up the budget.
• GPU clouds hedging against falling rental prices on capacity they have already paid for.
• Investors taking a view on AI demand through the price of compute itself.

Money is already moving into this layer. Ornn raised a $33M seed round in June 2026, and Silicon Data has raised a $30.5M Series A backed by CME Ventures, DRW and Jump.

CME, ICE and Ornn

• CME Group lists H100 and B200 rental futures from 5 October 2026, built on Silicon Data indices.
• ICE has announced its own GPU compute futures.
• Ornn publishes the OCPI price index and has announced futures with ICE, alongside a marketplace for dedicated GPU capacity. We compare it with Opengrid in Ornn vs Opengrid.

We aren't covering contract sizes, tickers or settlement terms here. Check the exchanges' own contract specifications for those.

What futures still need: a liquid spot market

A futures market is only as good as the price it settles against. For oil, that price comes from huge volumes of physical barrels changing hands every day. For GPU compute, the spot market underneath is still fragmented:

• 209 GPU clouds, each with its own console, contract and API.
• The same H100 listing from about $1.25 to $11.07 an hour, depending on who you rent from.
• Aggregators reaching about 20 of those clouds, so most supply is outside any single view.

An index can sample list prices. But the strongest reference price comes from real transactions, real jobs placed at real prices across the whole market. The more on-demand compute changes hands through a neutral layer, the better the price that futures settle against.

Where Opengrid fits

Opengrid is building that spot layer. You describe a job, with GPU type, count, region, price ceiling and deadline. Opengrid routes it to whichever provider wins on price, availability, region and reliability. Every routed job produces a real transaction price, not a list price.

That builds in three steps: route real workloads, turn those trades into a live price for every GPU type and region, and eventually let people trade it. For that last step, our plan is to list contracts with a regulated exchange partner, not to become an exchange.

Today, Opengrid is pre-launch. The market view ships first and the routing API is in development.