AI computing power is no longer just a cloud budget line. CME's reported move toward compute futures shows how quickly scarce infrastructure is becoming a financial product.
CME Group is reportedly preparing a futures market tied to computing power used for artificial intelligence, a sign that the AI boom is moving from chip shortages and data center leases into the machinery of global finance. If the contracts take shape, compute will start to look less like a private procurement problem and more like power, oil or interest rates: something companies try to lock in before prices move against them.
According to Bloomberg, BlackRock Chief Executive Larry Fink said at the Milken Institute Global Conference that buying futures of compute could become a new asset class, because the United States is short of compute, chips, memory and power. That comment landed at exactly the moment CME's reported plans started to make sense. AI demand has made graphics processors, cloud capacity and electricity feel like linked bottlenecks rather than separate markets.
The important detail is what CME actually standardizes. A futures contract needs a clean unit, and compute is messy. A GPU hour on one provider is not necessarily the same as a GPU hour on another. Model type, memory, networking, location, latency, uptime and power availability all change the value. That makes a simple contract based on raw GPU hours harder than it sounds.
The cleanest version would be cash settled against a benchmark index for compute prices, rather than physically delivered cloud capacity. That would avoid the problem of deciding whether a buyer should receive Nvidia H100 hours, B200 capacity, a basket of cloud instances, or some other standardized measure. It would also make the market easier for financial firms to trade without becoming cloud brokers.
Physical delivery would be more useful for companies that genuinely need capacity, but much harder to police. A startup training a model does not just need any compute. It needs compute in the right region, with the right networking, available at the right time and supported by the right software stack. A contract that delivers cheap but unsuitable capacity would not solve the real problem.
This is why the first serious compute futures market may look more like a financial hedge than a cloud reservation system. A buyer exposed to rising AI infrastructure costs could take a position that offsets price increases, while still sourcing actual capacity through Amazon Web Services, Microsoft Azure, Google Cloud, CoreWeave or another provider. That is not perfect, but it is how many commodity markets work. Airlines do not hedge jet fuel because futures deliver every gallon directly to the gate. They hedge because price risk matters.
CME has the scale to make the idea credible. The exchange reported April 2026 average daily volume of 25.9 million contracts, including 192,000 cryptocurrency contracts worth $14.8 billion in notional value. That matters because new markets need trust, clearing, margin systems and enough participants to make prices meaningful. Compute is a new subject, but the exchange infrastructure around risk is familiar.
Startups may not be the first winners
For AI startups, the pitch is attractive. Training and inference costs can decide whether a product works economically, especially when customer demand spikes or a model needs to be rebuilt. A hedge that gives finance teams more certainty could help smaller companies plan budgets, raise capital and commit to customers without praying that GPU prices stay friendly.
But there is a catch. The earliest winners may be the players already sitting closest to the supply. Hyperscalers, GPU cloud providers, data center owners, power developers and financing firms have the capacity, contracts and balance sheets needed to participate from day one. They can use a futures market to monetize scarcity, finance new infrastructure and transfer risk to traders looking for exposure to AI demand.
That does not make the market useless for startups. It just means the benefits may arrive indirectly. If futures improve price discovery, lenders can underwrite data center projects with more confidence, cloud providers can structure longer contracts, and buyers can see a market price instead of negotiating in the dark. Over time, that could make compute procurement less opaque.
The risk is that financialization runs ahead of standardization. Compute is not wheat. It changes quickly, and today's premium accelerator can become tomorrow's middle-tier machine. If contracts are poorly designed, they could create a market that is easy to trade but only loosely connected to the capacity builders actually need.
Still, the direction is clear. AI has turned compute into a strategic input, and strategic inputs rarely stay outside financial markets for long. The next question is whether CME can create a benchmark that reflects real AI infrastructure costs, not just a tradable story about scarcity. If it can, compute will leave the procurement spreadsheet and enter the same conversation as energy, rates and commodities.
Also read: Micron helps turn DRAM into Wall Street's fastest ETF breakout • Daniela Amodei built the world's most valuable AI lab on a literature degree • Microsoft's Kenya AI data center shows the grid limits of cloud expansion
