This is the kind of move that sounds boring—“remove a product”—but it’s actually a shot across the bow. If you’re building markets around AI, the U.S. government is signaling it wants to decide where the line is before you get comfortable on the other side of it.
Based on public reporting, the U.S. Commerce Department ordered Kalshi to remove an AI-compute futures product. The basic idea behind that product wasn’t hard to understand: create contracts tied to AI compute resources, like GPU rental prices, so people can manage risk. Kalshi has been working on event contracts and “forward curves” aimed at pricing that risk. At the same time, other parts of the federal government, including the CFTC, are exploring how to oversee these kinds of compute-related derivatives as prediction markets expand into AI.
That’s the fact pattern. Here’s my read: the government isn’t just worried about a weird financial novelty. It’s worried about letting a financial market form around something that is starting to look like critical infrastructure.
Compute isn’t a cute hobby input anymore. It’s closer to electricity for a certain class of companies. If you can’t get it at a price you can afford, your product roadmap slips, your customers leave, and your investors get jumpy. So the urge to hedge is real. If you run a team training models and your GPU costs swing wildly, you’d love a way to lock in expectations. If you’re a cloud provider or a broker in the middle, you’d love a clearer price signal. In a normal world, turning that uncertainty into a tradable contract is what markets do.
But we don’t live in a normal world with AI right now. We live in a world where compute is tied to national security, export controls, and who gets to build what, and how fast. And once you let a market like this get liquid, you don’t just get “risk management.” You also get speculation, leverage, and incentives that have nothing to do with “helping builders plan.”
Imagine a small AI startup that barely survives month to month. If it can hedge compute prices, maybe it avoids a brutal surprise bill and keeps hiring. That’s the happy story. Now imagine a well-capitalized trading shop that figures out how to push narratives about upcoming GPU shortages, then uses that attention to take positions. Or imagine someone with inside knowledge of supply constraints—orders, shipments, policy changes—using a contract like this as a clean way to profit. Even if nobody does anything illegal, you’ve now built a scoreboard for a resource the government is actively trying to control.
That’s why this Commerce Department angle matters. Commerce isn’t the agency people think of when they think “derivatives.” So when it shows up, it suggests the concern isn’t just “is this gambling” or “is this a regulated financial product.” It’s also “does this create a side-channel around compute policy.” If the government is trying to restrict certain flows of advanced compute, the last thing it wants is a widely traded contract that effectively broadcasts scarcity, expectations, and workarounds in real time.
There’s another tension here that people who love markets tend to wave away: a market doesn’t just reflect reality. It can shape reality. If enough players believe GPU prices will spike, they start behaving like they will. They hoard capacity, lock deals early, shift budgets, and suddenly the spike becomes more likely. Price signals are powerful, but they’re not neutral. In a hot sector, they can turn into self-fulfilling chaos.
At the same time, I don’t fully buy the idea that banning or yanking these products is automatically “responsible.” There’s a paternalistic streak in U.S. regulation where the default response to something new is to stop it first and think later. If compute derivatives are going to exist—and I suspect they will, because the demand is real—then pushing them out of visible, regulated venues doesn’t make the risk go away. It can just move it to darker corners where oversight is weaker and the products are sloppier.
There’s also a fairness issue hiding under the surface. Big companies already have ways to manage compute risk: long-term contracts, preferred access, relationships, custom pricing. Smaller teams don’t. A well-designed, well-regulated market could actually level the field a bit by giving more people a predictable way to plan. Killing the idea outright protects incumbents by default.
So the stakes are messy. If the government is too strict, it slows down legitimate risk management and cements power in the hands of the few players who already control supply. If it’s too loose, it invites a speculative layer on top of a scarce resource—and we’ve seen how that movie goes in other markets when financial players show up early and aggressively.
What I want—and what I’m not sure we’re getting—is a clear principle for where the boundary is. Is the problem “prediction markets tied to AI are morally icky”? Is it “anything that looks like a compute derivative belongs under a specific regulator”? Or is it “compute is too strategic to allow public contracts that could influence behavior”? Those are three very different positions, and they lead to very different rules.
If we’re going to treat AI compute like a strategic asset, then we should say that plainly and design rules that match that reality, instead of playing whack-a-mole with products after they ship.
So what should the line be: should markets be allowed to price and trade AI compute risk in public, or is compute now important enough that we should keep financial speculation away from it entirely?