More than 300 companies now rent out AI GPUs, each with its own hardware, networking, software, and contract terms. For an AI team, that means a procurement project every time it needs capacity, and often paying for GPUs that sit idle between jobs. For the providers, it means expensive hardware, power, and data centers waiting for customers. A former market maker at Citadel Securities thinks this is a trading problem. Jasper Zhang says his time there showed how liquidity and price discovery connect supply with demand. His company is Hyperbolic, based in San Francisco. On October 27, it announced a $30 million Series A1 led by Walden Catalyst, the firm founded by Intel CEO Lip-Bu Tan and former Samsung president Young Sohn. Zhang is the CEO. He's 31 and finished his UC Berkeley PhD in two years before joining Citadel Securities. His co-founder Yuchen Jin is the CTO, and the team also comes from Google Cloud, Lambda, VMware, Supermicro, and Point72. It started as an open-access AI cloud. You can rent H100, H200, or B200 GPUs by the hour with no quota limit and no long-term contract, reserve dedicated clusters, or call an OpenAI-compatible inference API and pay per token. The company says more than 250,000 builders use it, including teams at Modal, Higgsfield, Daytona, HeyGen, and Pika. Higgsfield says it was moved to a reserved contract 35% cheaper than on-demand pricing. The new money goes toward making it the layer between those 300 clouds. Forge, a control plane launched in June, standardizes bare-metal GPU clusters from different providers into one dashboard. Flux, released October 21, runs inference across multiple clouds. Managed Kubernetes entered beta on October 13. Next is the Global Compute Grid, which is planned but not live. Suppliers and customers would buy, sell, and resell GPU capacity, with spot instances, dynamic pricing, and market-making strategies. It is also exploring ways to hedge compute price risk. Zhang's thesis is that AI agents will become the largest consumers of compute, and they won't sit through sales calls or sign three-year contracts. They need a programmable way to find capacity, so the company wants agents to acquire it within set budgets. Dell Technologies Capital, General Catalyst, Vista Equity Partners, Cambium Capital, and Salience Capital also joined the round. Total funding is $50 million, including a $12 million Series A in December 2024 co-led by Polychain Capital and Variant. The vision is for compute to work like electricity, drawing what you need and selling back what you don't.