rewrite this title The new oil? Inside the effort to turn AI computing power into a tradeable commodity

For decades, companies have turned to futures markets to manage uncertainty. Airlines hedge fuel costs. Farmers hedge crops. Manufacturers hedge metals.

Now a startup wants to bring that same financial machinery to artificial intelligence.

Silicon Data, a company that tracks pricing across cloud providers and GPU marketplaces, has partnered with CME Group to launch what could become the world’s first futures contracts tied to the computational power needed to run AI, allowing companies to hedge against fluctuations in the cost to train and run AI models. The contracts are still awaiting regulatory approval.

Early signs suggest investor interest is quickly emerging. Within days of Silicon Data’s announcement with CME Group, asset managers including ProShares and Rex Shares filed proposals for exchange-traded funds tied to the proposed contracts, including leveraged and inverse products.

Founder and CEO Carmen Li believes the market could eventually rival some of the world’s largest commodity markets.

“I think it will be larger” than oil futures, Li said in an interview, adding that energy demand tied to running artificial intelligence will eventually surpass all other energy uses, combined.

Like jet fuel

The idea stems from a simple observation: AI companies increasingly depend on compute in the same way airlines depend on jet fuel.

Most companies don’t own the high-end graphics processing units, or GPUs, that power modern AI systems. Instead, they rent access through cloud providers and a growing ecosystem of so-called neoclouds. As demand for AI infrastructure surges, the cost of that compute can fluctuate, making it difficult for businesses to forecast expenses.

“Right now we’re at a high point of uncertainty,” said Seoyoung Kim, a finance professor at Santa Clara University. “A lot of people don’t know how much computing power they’ll need in the next year, and a lot of suppliers of that computing power right now don’t know how many GPUs and to what capacity they should order and the manufacturers, like Nvidia, they don’t know how much they should produce.”

Silicon Data has built a series of GPU price indexes that track the hourly rental cost of specific chips across providers. The company hopes those benchmarks can serve as the foundation for a futures market, much as West Texas Intermediate crude oil underpins energy derivatives.

Like any futures market, compute contracts will need both buyers and sellers. Companies worried about rising compute costs would seek protection from higher prices, while providers with large amounts of capacity could hedge against the risk of prices falling.

Silicon Data’s benchmarks have already begun appearing in high-profile corporate disclosures. SpaceX, for example, referenced the company’s GPU rental-rate data in its prospectus to go public.

Speculators coming in

Not everyone in the market would be looking to hedge risk. As with other futures markets, compute contracts would also draw speculators — traders with no direct need for GPU capacity but a view on where compute prices are headed.

Proponents argue that speculators play an important role in building liquidity and improving price discovery. Critics counter that speculation can amplify volatility and disconnect prices from underlying demand.

“Speculators are a very important piece of the ecosystem as well,” Li said. “You need natural hedgers. You need market makers. You need speculators. They have opinion. They want to express their opinion, which is perfectly fine.”

The Harvard MBA said traders who believe they have insight into future supply and demand dynamics should be able to express those views through the market, helping establish prices for the broader industry.

The ProShares and Rex Shares filings for ETFs are contingent on regulatory approval of the futures market. Still, they suggest some investors already view AI compute as a potentially tradable asset class rather than simply a technology input.

Benchmarking AI compute cost

Unlike a barrel of oil, AI compute is not a standardized physical commodity. Silicon Data said there are more than 50 different configurations of Nvidia’s H100 chip alone, with prices varying based on processors, memory, networking, utilization rates and data center location.

For the proposed futures market to work, traders need confidence that a single benchmark can accurately represent those variations.

“What we do is normalize the prices coming to our platform every day to a base H100 case,” Li said. “It’s a very complicated normalization step, even before the index calculation step.”

Kim, the Santa Clara finance professor, noted that standardization has always been a challenge for futures markets. Corn futures, for example, specify the exact grade of corn that can be delivered under a contract. Compute markets face a similar task: defining precisely what buyers and sellers are trading.

The Commodity Futures Trading Commission “is going to want to know exactly what the product is,” Kim said. Contract specifications, settlement procedures and benchmark construction are all likely to face scrutiny before the market can launch, she said.

— CNBC’s Charlotte Morabito contributed to this story.