San Francisco–based startup SF Compute has secured $40 million in equity financing to expand its marketplace for AI computing power
The company’s approach could help prevent systemic risk in the fast-growing AI sector. The new funding round values the company at $300 million and marks a growing push by investors to rethink how AI infrastructure is financed and consumed.
The funding news was shared with the Wall Street Journal. SF Compute’s pitch arrives amid rising concern that AI developers, who typically charge customers based on usage, are increasingly on the hook for large, inflexible compute contracts that could leave them with expensive idle hardware. SF Compute argues that this mismatch has put pressure on startups’ balance sheets and could trigger failures if market conditions tighten.
AI’s expensive
AI developers must secure access to graphics processing units (GPUs)—the chips that power modern AI models—often signing deals that lock them into 12- to 36-month commitments. Although their own customers may use a service only sporadically, the startup must pay for peak capacity whether it is fully utilized or not.
Investors have begun warning that the structure is creating hidden liabilities inside the AI ecosystem. If a downturn or a slowdown in demand arrives, many AI startups risk being saddled with costs far exceeding revenue.
This is the backdrop in which SF Compute positions itself. By enabling buyers to resell unused capacity in a flexible marketplace, the company says it can reduce the likelihood of overbuilding data-center resources and limit exposure to fixed charges.
A marketplace model
SF Compute sells long-term compute contracts financed by outside investors, then gives buyers the option to sublease their spare capacity through a liquid, real-time marketplace.
The company likens its model to Airbnb—managing rather than owning the underlying asset. Airbnb doesn’t own homes, and SF Compute doesn’t own GPUs.
The platform takes roughly 10 percent from each transaction. While CEO Evan Conrad declined to disclose revenue figures or user counts, the company says it manages more than $100 million in hardware, encompassing several thousand GPUs.
AI infrastructure funding surges
The Series A round was led by DCVC and Wing Venture Capital, joined by previous backers Electric Capital and Jack Altman’s Alt Capital.
The deal arrives during a major surge in cloud and AI infrastructure investment: PitchBook reports that $11.76 billion has flowed into cloud-computing startups so far in 2025, more than double the total for all of 2024. Companies such as Groq, Lambda, and Nscale have been among the year’s biggest fundraising winners.
This momentum reflects the industry’s scramble to secure compute access as model training and inference demands rise. However, it also heightens the risk of oversupply—potentially leaving the market with unused clusters and unsustainable financing structures.
SF Compute argues that a marketplace that recirculates unused compute could reduce pressure on data-center operators and curb speculative overbuilding.
New hires
SF Compute now employs about 30 people. Recent executive hires include Eric Park, former CEO of cloud-computing provider Voltage Park, as chief technology officer. The startup also brought on Alan Butler, previously with Sun Microsystems and most recently vice president of partnerships at Lambda Labs, as chief business officer.
The hires underscore how intensely competitive the GPU-provisioning market has become, as dozens of companies scramble to secure chips and offer differentiated infrastructure services.
Protection against an AI downturn?
While SF Compute seeks to reduce systemic risk, its model is not insulated from broader industry cycles. If demand falls, sellers could outnumber buyers, pushing prices down—and decreasing revenues on a percentage-based model.
Still, Ali Tamaseb, a general partner at DCVC, who joined SF Compute’s board, argues that marketplaces tend to persist even in recessions, In an event like the dot-com bubble bursting, for example, he said SF Compute “would probably lose 80% of its value but it wouldn’t die, because it’s a marketplace.”
AI is infiltrating offices, factories, and call centers faster than many workers anticipated. A new McKinsey analysis suggests the shift could be more significant than initially thought.