The San Francisco startup has gone from a $60 million valuation to $800 million in five months, betting that specialized chips will reshape the economics of AI.
As investors question whether the hundreds of companies building AI data centers can survive, one San Francisco startup has just raised $100 million to pursue a different approach.
General Compute announced a Series A co-led by Blackbird Ventures and Abstract Ventures, valuing the company at $800 million, according to TechCrunch. That's more than 13 times its $60 million valuation in May, when it raised a $15 million seed round.
Fuse Venture Partners, Village Global, QED Investors, 645 Ventures and several other investors participated.
Founded in 2025 by Finn Puklowski and Jason Goodison, General Compute rents out processing capacity for running AI models. But unlike the emerging generation of cloud providers buying enormous quantities of Nvidia GPUs, the company is building much of its infrastructure around chips designed specifically for inference.
Its hardware partners include Cerebras, SambaNova and Positron AI.
The premise is that AI companies shouldn't have to rely on one type of processor for every task.
Running a large language model involves two distinct computational stages.
The first, known as prefill, processes the user's input. The second, decode, generates the answer token by token.
Nvidia GPUs can handle both operations, but the hardware requirements are different. Some specialized processors can generate tokens faster or more efficiently for particular workloads.
General Compute wants to combine those architectures.
Its system can use GPUs for prefill and specialized chips for decode, matching different stages of a request to hardware suited to the task.
That could be particularly useful for coding agents, which repeatedly generate code, run tools, examine results and continue working through multiple steps. Faster inference can shorten the time needed to complete those sequences.
The company claims its infrastructure can deliver substantial speed improvements over conventional GPU-only deployments, depending on the workload. Those comparisons are company-reported and should not be treated as universal performance results.
There is another problem General Compute wants to solve: even if a chip is technically superior for a particular application, making it available to customers requires far more than buying the hardware.
Someone has to finance the systems, locate suitable data center space, arrange power and cooling, configure networking and operate the equipment.
General Compute handles that work and sells customers dedicated computing capacity under commercial contracts.
The company has already secured a debt facility of up to $400 million from Upper90 to help finance its infrastructure expansion.
The facility is separate from the new $100 million equity round.
Rather than constructing entirely new data centers, General Compute is leasing space in existing facilities and repurposing locations previously used for cryptocurrency mining.
That approach could allow it to deploy hardware faster than companies waiting for new large-scale data centers to be completed.
The company has also secured a major customer agreement that TechCrunch described as worth $100 million in annual recurring revenue.
That represents a significant commercial commitment, although it should not be confused with revenue already recognized.
General Compute's initial large-scale deployments are scheduled to begin in the first quarter of 2027.
The company says it has more than 50 megawatts of computing capacity in its development pipeline. Puklowski told TechCrunch he expects to deploy approximately 75 megawatts during 2027.
Neither figure represents computing capacity that is already fully operational.
General Compute is entering a market dominated by enormous capital commitments.
Companies such as CoreWeave have built businesses around providing access to GPU infrastructure, while Amazon, Microsoft and Google continue investing heavily in their own computing networks.
Puklowski is taking a different route.
Rather than competing directly to build the largest GPU cloud, General Compute wants to become the distribution and infrastructure partner for chip manufacturers attempting to challenge Nvidia's position.
For Cerebras, SambaNova and Positron, that solves a commercial problem. Designing a powerful chip is only part of building a successful semiconductor business. The manufacturer also needs customers who can access the hardware without purchasing and operating it themselves.
General Compute could become an important intermediary between those chipmakers and the AI companies looking for faster inference.
But the model carries considerable risk.
The business requires expensive hardware, dependable customers and enough utilization to cover financing and operating costs. Chips can also become less competitive as newer architectures arrive.
If inference prices fall faster than General Compute's costs, the economics could deteriorate even as demand for AI computing increases.
The startup's valuation has risen sharply before its first major deployments are scheduled to come online.
Its next challenge is proving that its combination of specialized processors, borrowed data center capacity and long-term customer contracts can deliver profitable computing at scale.
For now, investors are backing the idea that the future of AI infrastructure will involve more than one chip architecture — and that a new generation of cloud providers can build substantial businesses connecting them.
Sources: General Compute announcement · TechCrunch · Data Center Dynamics