Hyperbolic raises US$30 million to turn fragmented GPU capacity into a single market for AI compute
Vertex Holdings06 Nov 2026The Vertex Ventures China portfolio company is building an orchestration and trading layer that lets AI teams use compute wherever it sits.
Spending on AI hardware keeps rising, but buying more GPUs solves only part of the problem. Capacity is spread across more than 300 GPU cloud providers, each with its own hardware, networking, software and operating standards. AI teams face complex procurement and rigid contracts, and often pay for capacity that sits idle between workloads. Providers, for their part, need to turn their investment in hardware, power and data centres into steady revenue.
Hyperbolic, a San Francisco-based AI infrastructure company, has raised US$30 million in Series A1 funding to close that gap. The funding will go towards three things: Forge, its unified control plane; its multi-cloud orchestration services; and what the company describes as the first Global Compute Grid.
From fragmented clusters to one platform
Forge standardises mixed bare-metal GPU clusters into validated infrastructure. Teams get a single dashboard for provisioning, virtualisation, networking and operations across providers. Hyperbolic’s managed Kubernetes and inference services run on top of it, and a unified API gives access to models hosted by Hyperbolic and by external providers. AI teams get more useful work from each GPU hour, and the providers supplying the capacity see it used more fully.
The approach draws on the background of founder and CEO Jasper Zhang. He completed his PhD in mathematics at UC Berkeley in two years and then worked in market making at Citadel Securities. Hyperbolic treats compute as a resource to be matched, priced and traded, not simply rented out.
“AI teams should be able to access compute wherever it is, run their workloads, and turn unused capacity into revenue through one platform. We’re bringing orchestration and liquidity together so customers have more flexibility in how they use compute, while providers get more productive use and better returns from their infrastructure,” said Zhang.
More than 250,000 AI builders now use the platform, including teams at Modal, Higgsfield, Daytona, HeyGen and Pika. Higgsfield, for example, scaled up on on-demand H100 capacity and then moved to a reserved contract priced 35% below the on-demand rate.
Compute as a market
Hyperbolic’s next phase goes beyond orchestration. The Global Compute Grid is designed to let suppliers and customers buy, sell and resell eligible GPU capacity. Users would draw on compute as they need it, much like electricity, and sell unused capacity back to the grid. The company plans to add spot instances, dynamic pricing and market-making, and is exploring ways for customers and suppliers to hedge against swings in compute prices.
Over time, Hyperbolic expects AI agents to buy compute directly, within budgets and policies their operators set. Programmable access of this kind could also make AI accelerators, and the revenue they generate, easier to value and finance as an asset class.
Vertex’s perspective
Vertex Ventures China first invested in Hyperbolic at its Series A in 2024, when the platform served around 40,000 developers. That figure has since passed 250,000. Over the same period, the company has moved from a GPU marketplace to a platform that coordinates capacity across providers. Its next step is a grid where that capacity can be traded.
Looking ahead
As AI workloads multiply and software agents begin to buy compute on their own, the layer that matches supply with demand will matter more. Hyperbolic aims to be that layer, connecting the hardware already deployed with the teams that need it.
/f/233941/1500x900/90f8a040d7/hyperbolic-vh-website-banner.png)