Startup Led by Former Citadel Quant Pitches GPU Marketplace as a Hedge for AI Risk

With CoreWeave shares sliding and pulling much of the AI sector down with them, investors are asking how best to protect themselves against a downturn in AI. One answer is being developed by a former Citadel Securities quantitative researcher, Jasper Zhang, who now runs the startup Hyperbolic.

Hyperbolic is a 50-person financial technology company building a spot marketplace for tradable GPU computing power. The idea is to give traders a tool for hedging against a decline in AI demand. Investors on the platform can buy and sell GPU clusters, the hardware units that supply AI compute.

Trades run through Hyperbolic's Forge software, which logs into clusters around the world virtually and makes technical adjustments so that mismatched clusters work better together. Buyers can therefore do more than purchase a standardized unit of compute. They can also control the underlying GPUs for their own purposes. Zhang says the experience is as good as being inside the data center itself, and he sees the technology as a first building block for physically settled GPU futures. Upon settlement, such contracts would hand owners full control over a cluster with known characteristics.

The approach targets a problem in the market. GPU performance varies widely depending on how systems are assembled, and networking, memory and security all affect output. Financing agreements also often stop builders from sharing prices, so a survey-based index may average typical setup costs while missing outliers. In futures trading this mismatch is called basis risk, and Zhang hopes Hyperbolic can help address it.

Zhang is 31. He arrived in the United States with several Chinese math medals and finished a math doctorate at U.C. Berkeley in two years before joining Citadel Securities to build trading models. Hyperbolic's head of finance, Hong Chan Kim, is a hedge fund veteran who has worked at Point72, Balyasny Asset Management and Citadel LLC. The company has raised $20 million since launching in 2022 and is in the middle of another funding round. It expects a $150 million annualized run rate by the end of the year, and says it has about 250,000 users and an estimated $200 million of annual GPU compute connected to its platform.

Larger players see the same opening. CME Group plans a cash-settled futures product tied to GPU pricing maintained by Silicon Data, an index provider backed by the Chicago trading firm DRW. DRW founder and CEO Don Wilson said there is heavy demand and capital flowing into the ecosystem, and therefore a lot of risk. He called cash futures a great first building block, noting that electricity and oil futures are also cash-settled.

The CME product is under review by the CFTC, which recently extended its review period to November 9. That extension is likely also holding up another proposed futures contract from Intercontinental Exchange and data provider Ornn, which is backed by Andreessen Horowitz's a16z crypto fund. Kim argued that cash-settled futures may not reflect reality when they launch, and that physical usage is what will tie the derivatives to the real market.

Cash futures are also an imperfect hedge for real-world risk. If AI demand stalls, rent-a-center style providers such as CoreWeave could be left holding billions of dollars of hardware with little way to cash out. Carmen Li, CEO of Silicon Data and of the DRW-backed Compute Exchange, said an index is a normalized global number: it can provide a financial hedge but does not help someone who needs to procure hardware. Compute Exchange is itself laying groundwork for physically settled futures.

Zhang contends the AI sector most needs a virtual pipeline to physical hardware, especially for smaller startups that lack the capital or credit to make big purchases. "CoreWeave's debt situation would be less concerning if providers had more ways to monetize available GPU capacity during a slowdown, especially when long-term contracts carry thin margins," he said. Cash-settled futures can help with falling prices, he added, but they do not by themselves put idle GPUs to work.