For an AI company, securing GPUs can solve one problem while creating another: a commitment to capacity its business may not always need. Experiments end, workloads change, and infrastructure reserved for one project can become an expensive mismatch for the next.
Hyperbolic is betting that making those commitments more flexible could become a business in its own right.
The AI infrastructure company has announced a $30 million strategic funding round led by Walden Catalyst, the venture capital firm co-founded by Lip-Bu Tan and Young Sohn. Participants included Dell Technologies Capital, General Catalyst, Vista Equity Partners, Cambium Capital and Salience Capital, alongside strategic and angel investors.
The funding will help build Forge, Hyperbolic’s infrastructure control plane, and expand its ability to manage workloads across multiple cloud providers. It will also help finance a larger ambition: a global market where customers can acquire GPU capacity and resell eligible capacity they no longer need.
That strategy puts Hyperbolic at the intersection of two businesses with different demands. One involves making complicated infrastructure work reliably. The other involves building a market with enough buyers and sellers to make trading useful.
The company’s wager is that each can make the other more valuable.
“The demand for AI compute will continue to grow, but adding more hardware is only part of the answer,” Tan, Walden Catalyst’s founding managing partner and Intel’s CEO, said in the announcement. “The value of that investment depends on how effectively the capacity is put to work and made available to the teams that need it.”
Hyperbolic describes a GPU cloud ecosystem spanning more than 300 providers, with different hardware, networking, software and operational standards. For buyers, that can mean navigating separate procurement processes, contracts and engineering requirements. For suppliers, owning GPUs and securing power are only steps toward the commercial objective: keeping infrastructure productively occupied.
The resulting mismatch is central to Hyperbolic’s pitch. Buyers want flexibility and predictable performance. Providers want utilization and consistent revenue. A platform that can connect those interests could earn a role in both infrastructure operations and capacity distribution.
Founder and CEO Jasper Zhang’s background helps explain the approach.
According to the company, Zhang experienced the constraints of limited GPU access while conducting research at Berkeley AI Research. He later worked as a quantitative researcher focused on market making at Citadel Securities.
Those experiences connect the two halves of Hyperbolic’s strategy. Accessing compute is an engineering problem, but matching available capacity with changing demand also involves pricing, contract terms and liquidity.
In financial markets, liquidity helps participants enter and exit positions. Hyperbolic wants to bring some of that flexibility to compute, where a customer’s infrastructure commitment can outlast the workload that justified it.
Before capacity can trade efficiently, however, customers need to know what they are buying and how to use it.
That is the role of Forge. Hyperbolic describes the product as a common operating layer for heterogeneous GPU infrastructure, with tools for standardizing clusters, provisioning resources and managing operations across providers. The company also offers managed Kubernetes and inference services, including access to models through a unified API.
These capabilities provide a foundation for the proposed marketplace. A listing has limited value if a buyer must undertake a separate infrastructure project before running a workload on it.
Hyperbolic’s planned Global Compute Grid would build on that foundation with spot instances, dynamic pricing and resale of eligible capacity. The company also plans to develop market-making strategies to connect demand across different contract durations. Tools for managing compute price risk remain exploratory.
For customers, the potential benefit is a way to recover value from some unused commitments. For providers, it is another route to finding demand for available infrastructure. For Hyperbolic, the opportunity is to become both the software platform used to operate capacity and the market used to distribute it.
“AI teams should be able to access compute wherever it is, run their workloads, and turn unused capacity into revenue through one platform,” Zhang said in the announcement.
The qualification around eligible capacity matters. The proposed system would not automatically make every GPU reservation transferable. Its usefulness will depend partly on which commitments can be resold, under what terms, and whether another buyer wants them.
Hardware differences create another constraint.
A GPU-hour on one cluster may not substitute for a GPU-hour on another. Accelerator type, memory, interconnects, storage and software configuration can determine whether capacity is suitable for a particular workload.
Hyperbolic’s comparison to an electricity grid captures the convenience it hopes to offer. Delivering that experience will require translating substantial differences underneath it into choices customers can evaluate and trust.
The market will also need sufficient usable supply and recurring demand. Dynamic pricing can help match buyers and sellers, but it cannot by itself guarantee that the right configuration will be available when a customer needs it.
Hyperbolic says it supports more than 250,000 AI builders, including teams at Modal, Higgsfield, Daytona, HeyGen and Pika. That gives the company a potential audience for its broader platform.
Its longer-term plans extend beyond customers manually selecting GPUs.
Hyperbolic envisions a system in which a user supplies a model or workload, a budget and a deadline. The platform would identify suitable infrastructure, run the job and adjust resource allocation as prices, availability and performance change. Suitable workloads could eventually move between providers with minimal disruption.
The company sees AI agents as prospective customers for that programmable infrastructure.
“They need a programmable way to discover compute, acquire capacity, and run workloads,” Zhang said.
That vision also underpins Hyperbolic’s interest in making AI infrastructure easier to value and finance. More transparent pricing and clearer revenue expectations could, in principle, help suppliers and investors assess infrastructure returns. Whether compute becomes a broadly investable asset class will depend on how predictable and transferable those returns become.
For now, Hyperbolic’s task is more immediate: prove that operating software and a capacity market can reinforce one another.
If it can make diverse infrastructure consistently useful to buyers, it could attract more demand. If that demand helps suppliers earn more from their capacity, it could attract more supply. The $30 million round gives Hyperbolic capital to pursue that cycle. Its next test is demonstrating that GPU capacity can become meaningfully easier to trade while still delivering the performance customers require.
