Jump、DRW 投入自营资本与交易团队布局新兴算力期货市场，CME/ICE 合约待监管批准

Jump Trading and DRW are committing traders and capital to an emerging market for financial contracts tied to computing power, as efforts to make the infrastructure behind the AI boom tradable move from theory into practice.

Jump has assembled a dedicated team that will use its own balance sheet to take positions and hold associated risks. The growing team includes Affaan Mustafa and Alejandro Soumah Fernandez, who joinedfrom compute-trading startup Ito Markets Inc., and plans to trade across markets including over-the-counter deals, perpetual futures and prediction bets.

Jump expects to make markets in compute futures planned by CME and Silicon Data, as well as rival contracts from Intercontinental Exchange and Ornn. DRW plans to trade compute through three teams pursuing different strategies, two existing and one new. It plans to trade the Silicon Data products but hasn’t decided which other compute contracts it will trade. CME’s launch remains subject to regulatory approval.

Read more: CME to Create Futures Market for Computing Power Backing AI

Jump Trading Hires Startup Founders to Lead Compute Desk

The push comes as a big selloff in AI infrastructure stocks and bonds raises fresh doubts about whether revenue from computing power can justify the cost of building it. Shares and bonds of CoreWeave, one of the biggest neocloud operators, have slumped as investors question whether future AI revenue can support the debt taken on to build data centers.

Carmen Li, founder of Silicon Data, whose indexes are expected to underpin CME’s planned futures, said the volatility strengthened the case for contracts that could help companies hedge GPU rental prices. “Futures cannot solve weak underlying demand or poor capital allocation. But they can help companies separate market price risk from fundamental business risk.”

Boston Consulting Group estimates the AI compute market could reach $2.3 trillion by 2030. BlackRock chief executive Larry Fink has said futures on computing power could become a new asset class.

The bet for firms such as Jump and DRW is to get there before the market is fully formed. Both moved early into crypto, making markets when trading was fragmented and much of Wall Street was still on the sidelines. Jump sees some of the same conditions in compute. A customer might need a particular Nvidia chip in one region, connected in a particular cluster, while similar GPU capacity sits unused somewhere else.

Dave Olsen, president and chief investment officer of Jump Trading, said the firm expects to “trade against those different pockets of demand and supply.” He added: “We think there’s a real opportunity for us.”

With the industry still dominated by negotiated deals, exchanges and trading firms are seeking to build out a financial market for hedging the price of computing power. CME and Silicon Data have been preparing futures tied to hourly rental prices for Nvidia H100 and B200 chips. Intercontinental Exchange is backing contracts tied to indexes from Ornn. FalconX, a digital-asset prime broker, has executed an over-the-counter swap tied to future compute prices.

Read: Race to Turn AI Compute Into a Commodity Spurs New Crypto Boom

DRW founder Don Wilson co-founded Compute Exchange, where companies buy and sell GPU capacity. He also invested in Silicon Data. Wilson sees transaction data and credible pricing indices as prerequisites for a functioning derivatives market tied to compute. Jump is an investor in both Silicon Data and Ornn.

Read: Odd Lots: Don Wilson on the Future of GPU Trading (Podcast)

Wilson said DRW has so far spent more time on indexes and derivatives than on buying and selling compute directly. Its spot activity, he said, has amounted to “dabbling.”

“Right now, the market is very opaque, and there isn’t good price transparency,” Wilson said.

Olsen expects financial firms to get there first. Hedge funds, dealers and other traders could dominate futures trading for “a prolonged period,” he said, before AI labs, neoclouds and other large compute buyers start using the contracts themselves. Those buyers will “stay on the sidelines until they see the cash settled products track their procurement experience,” Olsen said. “That will take some time.”

The sums involved are enormous. Anthropic has committed more than $100 billion over 10 years to Amazon Web Services technologies for up to 5 gigawatts of computing capacity.

“Continued investment in AI infrastructure depends on someone being willing to bear the risk of future compute prices and customers’ ability to pay,” Wilson said. “Futures give providers a way to manage price risk without relying on multiyear commitments from AI startups.”

In this view, a neocloud could one day sell futures against some of its capacity rather than rely entirely on a years-long agreement with just one customer. An AI lab could use the same market to protect itself against a jump in future compute costs.

Both Jump and DRW say futures products could help curb price and credit risks tied to the compute market. Rental prices for high-end chips can be decidedly volatile on a quarter-on-quarter basis, judging by data from SemiAnalysis, leaving AI startups exposed to potential cost spikes and the lenders that finance data centers vulnerable to bouts of market volatility.

The problem is that the same type of GPU can fetch different prices depending on where it is, who provides it and how it is configured. A futures index could therefore move differently from the specific capacity a company needs. “Right now, that’s a little bit of a leap of faith,” Olsen said.

Unused computing capacity also expires. An hour on a GPU cannot be saved for next month. A barrel of oil can sit in a tank; copper can lie in a warehouse. Olsen said that makes a standard commodity trade — buying the asset, storing it and selling it later against a higher futures price — difficult to reproduce.

Wilson said futures could make the economics of longer-term compute deals easier to assess. A company might buy several years of capacity and profit from reselling the near-term portion while still facing losses later if prices fall. A forward curve could make that remaining exposure more visible and give firms such as DRW another way to manage the risk. He also expects algorithmic trading to become another opportunity if the futures market develops enough liquidity.

For now, the market makers could arrive well before many of the customers the market is ultimately meant to serve.

“There are a lot of unknowns, and that’s the calculated investment choice that you make to try to be in early before it’s clear what all is going to happen,” Olsen said.