Over sixteen days in May, three of the biggest names in financial markets quietly moved to do the same thing: build a futures market for computing power. CME Group, the world's largest derivatives exchange, partnered with a data firm called Silicon Data. The owner of the New York Stock Exchange, Intercontinental Exchange, teamed up with a startup called Ornn. And a third company, Architect — run by a former FTX US president — announced it would launch its own exchange from scratch.

None of it made front-page news — we flagged it ourselves only briefly in this weekend's Daily Update. But taken together, it's one of the more telling developments of the year: the AI industry's core input is being turned into something you can trade, hedge, and speculate on — the way the world already trades oil, wheat and natural gas. If that holds, it changes how the entire business of AI is financed.

What a "compute future" actually is

A futures contract is an agreement to buy something at a set price on a future date. Airlines use them to lock in jet-fuel costs so a price spike doesn't wreck a quarter. Farmers use them to guarantee a price for a harvest months away. The point is to manage risk: to turn an unpredictable future cost into a known one.

Compute futures apply that idea to renting GPUs — the chips that train and run AI models. And the problem they're built to solve is real. According to the research firm SemiAnalysis, the price of renting an Nvidia H100 on a one-year contract climbed nearly 40% in five months, from a low of around $1.70 per hour in October 2025 to roughly $2.35 by March 2026. Spot prices swing even more wildly — depending on the provider and how long you commit, an H100 today can cost anywhere from about $1.50 to over $11 an hour, with the market median somewhere near $3.

For a lab planning a months-long training run, that volatility is brutal. A compute future lets it lock in a price now; if GPU rental rates rise, the gain on the contract offsets the higher bill. Cloud and data-centre operators want the mirror image — predictable revenue to justify the enormous sums they're spending on chips and power. As CME's chairman Terry Duffy put it, "Compute is the new oil of the 21st century… Investors need a trusted futures market to provide transparency, liquidity and effective risk management."

Three bets on the same idea

What's striking is that the three efforts are structurally different — they don't even agree on how to measure the price of compute.

CME and Silicon Data will base their contracts on Silicon Data's "Silicon Index," a daily benchmark of H100 and A100 rental rates captured across cloud providers and recalculated every business day. Silicon Data is backed by the trading firm DRW, whose founder Don Wilson predicts "compute will become the largest commodity in the world." The contracts are expected to be cash-settled to that index, with a launch planned for later this year pending regulatory review.

ICE and Ornn take a different tack. Their contracts reference the Ornn Compute Price Index, which Ornn says is the first compute benchmark built only from printed transactions — actual executed trades, rather than surveyed rates. The contracts are dollar-denominated and cash-settled, and would cover a wider spread of chips: H100, H200, B200 and even the RTX 5090. The survey-based-versus-transaction-based distinction is the crux of a genuine debate about which method better reflects the true price of compute.

Architect is doing the most ambitious thing of the three: building its own exchange. Rather than list contracts on someone else's venue, it acquired a regulated exchange licence to create what it calls the American Innovation Exchange, settling to indices from a provider called Compute Desk that track Nvidia's Hopper and Blackwell chips. It's a direct challenge to both CME and ICE.

The financial world is already piling in around them. Within days of CME's announcement, the fund managers ProShares and Rex Shares filed for ETFs tied to the proposed contracts — including leveraged and inverse products — before a single one has launched.

Why now

The timing isn't a coincidence. The whole industry is short of compute, and everyone expects that to continue. Days before any of these deals went public, BlackRock chief executive Larry Fink told the Milken Institute conference that "a new asset class will be buying futures of compute," arguing the US simply doesn't have enough chips, memory and power for the AI workloads coming. He turned out to be describing something that was already weeks from launch.

That's the real signal here. You don't build a futures market for something you think is a passing shortage. You build one when you believe scarcity is a permanent, structural feature of the economy — durable enough to underwrite a whole layer of financial machinery on top of it.

The case against

It's worth being honest about why this might not work, because the objections are serious.

The biggest is fungibility. A barrel of crude oil is a barrel of crude oil. A "GPU-hour" is not nearly so uniform: its real value depends on the chip model, how the cluster is wired together, the maturity of the software, storage, even geography. Two clusters with the same nominal H100-hours can deliver meaningfully different output. Critics argue that makes compute a "fake commodity" — a marketing proxy that hides too much variance to price cleanly. Notably, Silicon Data's own chief executive, Carmen Li, has publicly acknowledged that fungibility challenges complicate trading.

Then there's obsolescence. Oil doesn't go out of date; GPUs do, fast. The H100 is already being succeeded by the H200 and Blackwell chips, which raises a hard question about what a contract dated a year out is even referencing. And there's the basic liquidity problem: no one yet trades a standardised "1,000 H100-hours, US-East, third quarter" contract, and a market only works if enough buyers and sellers show up.

The counter-argument from believers is historical: electricity, bandwidth and liquefied natural gas were all once dismissed as too heterogeneous or too illiquid to trade — and functioning markets formed around them anyway. Whether compute follows the same path, or stays too messy to standardise, is the open question.

What it signals

Even if some of these specific contracts fail, the direction is clear enough to take seriously. The people who price risk for a living have looked at AI compute and concluded it's becoming a tradable commodity — scarce, volatile, and central enough to the economy to deserve its own market. The executives talking it up claim it could eventually rival oil futures in size; those are sales pitches, not modelled forecasts, and should be read that way.

But strip out the hype and a quieter truth remains. When Wall Street starts building futures markets around a resource, it's making a bet that the resource matters, and will keep being scarce, for a long time. That, more than any single contract, is what this says about where AI is heading.