Ex-Citadel quant’s Hyperbolic builds a GPU marketplace to hedge AI demand risk

CoreWeave shares are sinking, and they are taking a good chunk of the AI trade down with them. That leaves investors with an awkward question: how do you protect yourself if the AI compute boom cools off?

Jasper Zhang thinks he has an answer. Zhang, a former quantitative researcher at Citadel Securities, now runs Hyperbolic, a startup building a GPU marketplace designed to help investors hedge against a drop in AI demand.

A marketplace for the chips everyone wants

Hyperbolic is an AI cloud and GPU marketplace based in San Francisco. Zhang co-founded it with Yuchen Jin.

The basic model works like an Airbnb for graphics processors. Data centers sitting on GPUs they aren’t fully using can list that spare capacity on Hyperbolic. Developers who need compute can rent it, either on demand or through reserved arrangements.

Pricing is refreshed every week. As of October 2026, on-demand rentals of an H100 SXM started at about $3.19 per GPU-hour, according to the research findings.

The company says its prices run up to 50-75% below what traditional providers charge. The platform serves between 200,000 to 250,000 engineers and developers hunting for GPU supply.

The growth curve is steep. Hyperbolic passed an annualized revenue run-rate of more than $80 million by September 2026. In mid-2025, revenue was close to zero.

That kind of ramp usually requires a mountain of venture capital. Hyperbolic has raised approximately $20 million in total, including a $12 million Series A round in December 2024. For a company now talking about an $80 million run-rate, that is a relatively lean balance sheet.

Turning compute into a tradeable asset

The marketplace is only the first layer. Zhang’s bigger pitch is that GPU capacity should be treated as a hedgeable asset class, similar to how other commodities are bought, sold and protected against price swings.

Hyperbolic submitted comments to the Commodity Futures Trading Commission (CFTC) in October 2026 about the listing of compute derivatives contracts. Among its proposals is physical settlement for GPU-related derivatives.

Physical settlement means a contract ends with delivery of the real thing rather than a cash payment. In this case, the buyer would receive actual GPU time, not just a check reflecting price changes.

The company has also commented on compute market-making. Zhang has also floated the idea of customers reselling capacity they bought but don’t end up using. If a company reserves GPUs for a project that gets shelved, it could flip that capacity to someone else rather than eating the cost.

On the supply side, Hyperbolic charges a 10% platform fee on rental income. It also hands out point rewards to suppliers for idle GPU capacity, an incentive meant to coax more spare hardware onto the platform.

Why CoreWeave’s slide matters here

The timing is not subtle. CoreWeave shares fell to approximately $59 in mid-to-late October 2026, a decline of more than 50% from their 2025 peaks.

The stock hit multi-month lows as concerns grew over the company’s rising debt. That happened even while demand for AI computing remained robust.

The disconnect points to a structural tension that Hyperbolic is trying to exploit. GPU infrastructure is financed over long horizons, often with significant borrowing. AI workloads, by contrast, are bought and consumed in the short term, and needs can change quickly.

Hyperbolic’s approach targets that gap without building new data centers. By reallocating existing capacity and enabling resale, the company aims to squeeze more use out of hardware that is already plugged in.

What this means for investors and the AI compute market

For investors, the immediate appeal is obvious. Today, betting against AI demand mostly means shorting stocks like CoreWeave, which bundles compute exposure with debt levels, management decisions and broader market sentiment. A dedicated compute derivative could isolate the one variable many investors actually care about: what GPU time is worth.

Weekly price refreshes and a large developer base give Hyperbolic a starting point for price discovery. Reliable reference prices are a prerequisite for any derivatives market, and Hyperbolic is positioning its marketplace as a source of them.

Still, the obstacles are real. The CFTC comments are just that: comments. No compute derivatives contracts have been described as listed, and regulators will shape what eventually gets approved. Physical settlement also raises practical questions, since GPUs vary by model, location and performance, which makes standardizing a deliverable unit harder than with a barrel of oil.

The things to watch are concrete. Look for the CFTC’s response to proposals on compute derivatives, whether any exchange moves to list such contracts, and whether Hyperbolic’s revenue run-rate keeps climbing as CoreWeave’s stock tests investor patience.