Traders seek physical GPU hedges

CoreWeave shares have fallen sharply, dragging down a large portion of the artificial intelligence investment complex. This drop has forced investors to ask a direct question: how can they hedge against a decline in AI demand?

Jasper Zhang, a former Citadel Securities quantitative researcher, believes he has a solution. He leads Hyperbolic, a 50-person financial technology startup. The company built a spot marketplace for tradable GPU computing power. This tool lets traders buy and sell GPU clusters, the physical units that power AI workloads.

The platform operates through software called Forge. Forge virtually logs into clusters around the world. It makes technical adjustments so that mismatched clusters work together. This process allows traders to trade a standardized unit of compute while retaining control over the underlying hardware.

Zhang describes the access as being equivalent to being inside the data center. He views the technology as the first step toward physically settled GPU futures. Under such contracts, owners would gain complete control over a cluster with known specifications. This structure offers a direct way to hedge by trading the hardware itself.

Current GPU architecture varies widely based on assembly methods. Factors like networking, memory, and security all influence performance. Every data center is different. Financing agreements often prevent builders from sharing exact prices. This makes survey-based indices inaccurate because they average setup costs while missing outlier expenses.

In futures markets, this discrepancy is known as basis risk. Zhang aims to solve this problem. He is 31 years old and arrived in the United States with several Chinese math medals. He completed a math doctorate at the University of California, Berkeley, in two years before joining Citadel Securities to build trading models.

Hyperbolic has raised $20 million since launching in 2022. The company is currently in the middle of another funding round. It expects to reach an annualized run rate of $150 million by the end of the year. As of October 26, the company reports about 250,000 users. An estimated $200 million worth of annual GPU compute is connected to its platform.

Other large players are developing similar tools. CME Group plans to offer a cash-settled futures product tied to GPU pricing. Silicon Data maintains the index for this product. Silicon Data is backed by DRW, a major proprietary trading firm based in Chicago.

Don Wilson, founder and CEO of DRW, noted the high levels of risk in the sector. He said there is a ton of demand and capital flowing into the ecosystem. He compared the new product to cash-settled electricity and oil futures. The CME product is under review by the Commodity Futures Trading Commission. The commission extended that review period to November 9.

Hyperbolic’s head of finance, Hong Chan Kim, said cash-settled futures may not reflect reality. Kim is a hedge fund veteran who worked at Point72, Balyasny Asset Management, and Citadel LLC. Kim stated that physical usage will tether the connection between reality and derivatives.

Carmen Li, CEO of Silicon Data and Compute Exchange, explained that indices are normalized global numbers. She said such indices provide financial hedges but do not help with procurement. Compute Exchange is laying the groundwork for physically settled futures.

Zhang argues that smaller startups need a virtual pipeline to physical hardware. He noted that CoreWeave’s debt situation would be less concerning if providers had more ways to monetize available GPU capacity during a slowdown. Zhang said cash-settled futures can help with falling prices, but they do not put idle GPUs to work.