Why Wall Street wants to trade AI compute like oil

A new boom in tech industrialization changes the face of Silicon Valley’s capital stack.

Before they became tools for speculation, crude futures were a cure for volatile oil prices. In the early 1980s, the New York Mercantile Exchange and the Chicago Board of Trade began trading the contracts, promising the oil market some predictability through hedging.

Now more than 40 years later, the oil trader’s playbook is giving rise to a futures contract custom-built for the AI age, one that would let investors hedge the price of GPU computing power. And it may arrive not long after the Wall Street debut of Anthropic, which is days away from becoming the world’s first major pure-play artificial intelligence lab to go public.

That innovation in finance points to a still-evolving structural shift in Silicon Valley’s financing model. Private credit investors are stepping up—and upstaging VCs—to fund the AI infrastructure boom, alongside other capital-intensive reindustrialization efforts across robotics, autonomous vehicles, energy, defense and advanced manufacturing.

Investment is drifting away from the VC-led software era and is giving way to hard tech, with its unique need for massive, early bets in the form of long-duration capital outlays, which historically have relied more on credit-based financing than equity.

But a lack of price transparency or market benchmarks leaves AI’s lenders exposed if compute prices fall. New GPU price indexes, and the futures built on them, could give them better visibility and help them hedge their risks.

The next phase of growth

“Private credit is playing, and will play, a much more profound role in the next phase of Silicon Valley’s growth because hard tech is in again,” says Marc Andrew, founder of the Private Markets Forum, which hosts events at the intersection of technology and private markets

To feed the ravenous computing demands of Anthropic and other AI labs, private lenders have been lining up unprecedented billions of dollars in financing to buy or rent GPU chips—the hardware needed to train and run models like Claude or ChatGPT.

Because AI-focused data centers are built for massive “compute” power that is replenished every one to three years, the industry effectively treats GPU capacity as a commodity that could trade like oil futures.

Indeed, the nascent market for compute capacity is marked by other problematic quirks—and risks—that recall the gyrations that led to the creation of oil futures contracts.

Uncertainty around oil prices, always a crucial input for so much of the economy, drove many firms to manage their exposure to oil by hedging with futures.

Price disparity

Two parties buying the same GPU capacity often pay very different prices without knowing who got the better deal.

Many of the AI economy’s pick-and-shovel players, such as data-center builders like CoreWeave and Crusoe, have taken on debt using GPUs as collateral. But a dearth of price transparency or market benchmarks for the chips heightens risk in their industry.

Today, GPUs are increasingly viewed as a new asset class. So why not offer a formal futures market to allow firms in the AI ecosystem to manage their own version of that uncertainty?

That is the question being posed by a small contingent of finance engineers who have teamed up with US commodities market operators aiming to launch a compute market in the coming months. CME Group’s derivatives platform for compute futures is awaiting regulatory approval in the US.

CME is also an investor in Silicon Data, a venture-backed startup that pioneered a GPU price index aiming to be the market’s main reference for settling compute. A data rival called Ornn, whose backers include Andreessen Horowitz, is a partner with Intercontinental Exchange on an alternative futures contract. Ornn’s compute pricing data is already distributed through Bloomberg terminals.

When AI giants like Anthropic or OpenAI line up their compute supply, they negotiate directly with cloud providers for multiyear contracts, often resulting in wide pricing disparities.

At least in the early going, futures markets’ main participants will likely be derivatives professionals on trading desks of major private credit firms financing multibillion-dollar GPU supply deals between frontier AI labs and cloud-computing providers. Cloud providers, AI developers, and frontier labs may participate later as they assess the market and use futures to manage compute commitments.

Who holds the bag?

“This incredible AI [infrastructure] is built on the backs of creditors that have financed GPUs, the data centers, the land, the power,” said Wayne Nelms, CTO of Ornn. If there’s a market slump or crash, he added, “a lot of the bag holders, so to speak, will be not necessarily the AI companies that are buying the compute, but rather the financiers that are putting up capital.”

Don Wilson, the founder of trading firm DRW, has said that compute power is in the early days of becoming “the world’s biggest commodity.” DRW is an investor in Silicon Data, which has raised more than $35 million in venture capital, according to PitchBook data.

Those are lofty expectations. GPU-as-a-commodity bulls are making some bold assumptions that are open to debate. Critics counter, for example, that compute hardware lacks the degree of interchangeability and standardization needed for a liquid commodity market.

And then there’s the uncertain outlook for supplying all that land, power and labor required to build out GPU clusters. Anthropic’s IPO prospectus repeatedly cites that concern as one of its top risk factors, and nobody knows how severe those hurdles will be.

Providers of price benchmarks say that price transparency will drive efficiency and fuel participation in the compute trade, as well as the AI infrastructure buildout.

Not having this market infrastructure could pose risks of its own, such as slowing growth and favoring already-established operators over smaller innovators.

Adds Nelms: “My expectation is that a lot of capital and a lot of interest in this market is sitting on the sidelines, waiting to hedge their risks more efficiently.”