Compute Is Becoming a Commodity

CME Group and Silicon Data have announced plans for Compute Futures, and my first thought was: finally, a way to hedge compute. The contracts are not trading yet. Listing was targeted for October 5, 2026, but the CFTC extended its review, so the earliest date is now November 9.

The two contracts, GPU1 and GPU2, are tied to Silicon Data indexes of hourly neocloud rental rates for NVIDIA’s H100 and B200 GPUs, the hardware behind much of the industry’s training and inference. In other words, they give us a way to track one of the biggest operating costs for AI companies: computing resources.

I’ve spent a good part of my career thinking about hedging. I’ve worked with commodities and other assets where I had to accumulate, hold, or sell physical inventory. And one thing has always seemed obvious to me: hedging is an incredibly useful tool. It’s surprising how many businesses still leave themselves fully exposed to market prices when they don’t have to.

The mechanics are simple.

Imagine you’re selling grain. You have physical inventory sitting in storage, and you’re worried that prices might fall before you sell it. What do you do? You short the market. If the price of grain goes down, your physical inventory loses value, but your futures position makes money. You’ve hedged your exposure.

Now take the opposite case. You’re systematically buying grain, and it takes you a month to build up your physical position. During that month, prices can move against you. The worst-case scenario is that prices go up, and you end up paying more for the same amount of grain. So you go long futures. If the market rises, your futures gains offset the higher purchase price.

That’s linear hedging. Pretty straightforward.

Options make things even more interesting. For example, I can use a collar to reshape the distribution of my portfolio’s outcomes: raise the floor and cap the upside. A zero-cost collar can do this without an upfront premium, although there’s an obvious trade-off: I’m giving up some upside in exchange for more predictable cash flows.

And predictable cash flows matter. They can make a business easier to finance, improve its risk profile, and help it stay within credit covenants. Sometimes reducing volatility is more valuable than maximizing potential revenue.

This is how commodity businesses manage price risk every day.

So why shouldn’t AI companies do the same with compute?

The same hedging logic applies to both sides of the compute market.

Let’s start with companies buying computing capacity.

Imagine you’re an AI startup. Your models run on cloud infrastructure, and you’re consuming GPU capacity on a regular basis. Your exposure is pretty clear: if compute prices go up, your operating costs go up too.

You can hedge that exposure by going long Compute Futures.

Essentially, you’re locking in a financial hedge against future price increases. These contracts are cash-settled, so you’re not actually reserving GPUs or securing physical capacity. You’re taking a position that should gain value if the benchmark price of compute increases.

Suppose market prices go up. Your cloud provider charges you more, but your long futures position generates a profit that offsets some or all of that increase.

What if prices go down? Your futures position loses money, but your actual compute costs decrease. You’re effectively trading some potential savings for greater price certainty.

Of course, there’s an important detail here: the futures benchmark won’t necessarily match the exact rate you’re paying your cloud provider. There’s basis risk, just as there is in commodity markets. But that doesn’t make the hedge useless. It means you need to understand how closely your actual exposure tracks the underlying benchmark.

Now let’s look at the other side: compute providers.

Suppose you own GPUs and rent out computing capacity to customers. You have a different problem. You’re worried that market rental rates will fall. If they do, you’ll have to rent out your capacity at lower prices, and your revenue will take a hit.

Your hedge is the opposite: you short Compute Futures.

If compute prices fall, your short futures position generates a profit, helping offset the reduction in rental revenue. If prices rise, you lose money on the futures position, but your underlying business earns more from renting out GPUs.

Same instrument, opposite exposure, opposite position.

This is exactly the kind of symmetry that makes derivatives so useful. Whether you’re buying or selling a commodity, you can use the same market to manage your exposure.

And compute is becoming a commodity-like resource in this sense. AI companies increasingly depend on it, providers compete to sell it, and its price directly affects business economics.

I think we’re only beginning to see what financial markets can do for this industry. Once listed, Compute Futures will make it possible to apply decades of experience from commodity markets to a new and increasingly important asset class.

Once the market opens, I’m personally looking to get involved and start hedging compute exposure for clients. I think there’s a real opportunity here, particularly for companies that have meaningful and recurring compute expenses or GPU rental revenues.

So, if you’re building an AI company and buying compute, or running infrastructure and selling it, get in touch.

I’d be happy to discuss how to hedge your exposure. My hedge fund is looking to work with companies on both sides of this market.