TheChek News Flow

The Takeaway

Goldman Sachs and JPMorgan are exploring ways to trade on the cost of computing power, according to people familiar with the matter. That includes trading futures contracts tied to rental prices for graphics processing units, among the scarcest resources of the AI boom, which exchanges plan to list later this year.

The discussions show how the hundreds of billions of dollars pouring into data centers and chips are reshaping financial markets. For banks financing the AI buildout, futures could provide a way to manage the risk of a compute glut down the road and help their clients hedge their own compute needs.

The banks’ exploration is still early and they may not move forward immediately, while the compute trading market is nascent and could face regulatory questions. But the idea is not entirely foreign to big banks, which already trade power and other commodities related to AI infrastructure. Compute trading could be a natural next step, some market participants say.

More broadly, firms including exchanges and data providers are racing to transform GPU rental pricing into a formal financial market. The goal is to make a major cost of AI infrastructure behave more like a commodity, with prices that can be tracked and hedged, as short-term costs swing and longer-term supply and demand remain difficult to match up.

“There is so much money flowing through it that everyone wants protection for themselves in a downside scenario,” said Warren Hosseinion, head of capital markets at Compute Labs, which is raising money to buy GPUs and rent them out on behalf of clients. “They’re incentivized to get this out in the quickest and most compliant manner possible.”

Goldman Sachs and JPMorgan declined to comment.

Big traditional market operators CME Group and Intercontinental Exchange have both announced plans to offer contracts later this year, pending regulatory approval from the Commodity Futures Trading Commission.

The market is also drawing startups including Architect Financial Technologies, founded by former Jane Street, Citadel Securities and FTX.US executive Brett Harrison, which is building a compute futures exchange it plans to launch in the third quarter, also pending regulatory approval.

Compute futures would resemble existing commodity futures, letting AI companies that rent or sell GPU capacity bet on whether rental prices will rise or fall. Such futures will likely also attract speculators like hedge funds looking to wager on where the market is headed.

The exchanges developing compute contracts haven’t yet outlined exact contract lengths, though mature futures markets often offer trading on prices months ahead. The trades would be settled in cash, meaning traders pay or receive money based on where prices land, and no actual hardware capacity changes hands.

Beyond the exchanges, building an active market would also require brokers and trading firms. That would help clients trade the contracts and keep activity moving, a role some firms are already positioning themselves to play.

That includes Chicago-based trading firm DRW, which already has a team that trades spot compute, meaning access to servers for a period of time, and has agreed to be a trading partner for CME’s compute futures. DRW expects to be active in the futures market once it launches, betting that the increasing capital being spent to secure capacity is creating a new trading opportunity.

“As spending on compute continues to grow, market participants need better tools to manage risk, improve planning certainty and support long-term investment,” said Don Wilson, DRW’s founder and CEO.

Others are also preparing to bring customers into the market. StoneX, a brokerage that helps clients set up and clear futures contracts, said it plans to provide clients access to compute futures. Crypto prime broker FalconX and crypto trading firm Wintermute also said they plan to trade the contracts, reflecting the interest from firms already accustomed to volatile digital assets.

The plans of these firms have not been previously reported.

“We’ve seen interest from miners,” large companies that supply compute, as well as crypto-native funds interested in the speculative aspect, said Ravi Doshi, global co-head of markets at FalconX.

The compute futures could provide financial protection for a variety of firms involved in the AI buildout. Wall Street banks, for instance, may want protection against falling GPU rental prices or values, which could hurt the worth of their AI infrastructure loans, leases or related collateral. By taking a short futures position, a bank could make money if compute prices decline, helping offset losses elsewhere.

Meanwhile, regional banks, especially West Coast ones, have clients such as AI startups that don’t have enough GPU capacity currently. Those banks could buy futures to help their clients hedge, said Carmen Li, CEO of pricing data provider Silicon Data and GPU marketplace Compute Exchange. “We are heavily engaged with all the banks,” Li said, without commenting on any specific banks.

Cloud providers such as CoreWeave that rent GPU capacity may also have a use for the market. They face uncertainty over what they can charge when their existing contracts expire, and a short position could protect them if future rental prices fall.

Still, the emerging compute market faces hurdles that could slow any broader rollout. For a futures market to take off, there typically needs to be a reliable price benchmark, which determines who pays what in settling contracts. And to make the contracts useful to compute renters or sellers, the benchmark needs to be a good approximation of real-world costs.

But compute pricing is inherently difficult to standardize. Rental rates can vary even for the same type of GPU, depending on hardware configurations, electricity costs and other factors, while major compute deals are often negotiated privately, making real-world pricing data hard to gather. To address that, exchanges are lining up independent data providers, with CME saying it will use Silicon Data, whereas Intercontinental Exchange plans to use data from Ornn.

That complexity also presents a potential regulatory hurdle—before exchanges can list compute futures, the CFTC is expected to examine factors including the reliability of benchmarks and whether there are enough safeguards to prevent traders from gaming the index, industry participants say. The CFTC didn’t respond to a request for comment.

While U.S. exchanges are still working through that process, similar products are already appearing in offshore markets. Prediction market Polymarket and crypto exchange Lighter have recently listed compute futures contracts. Polymarket, for example, allows users to bet on where rental prices for Nvidia H100 GPUs will be at the end of June, using Ornn’s index as the benchmark.