General Compute Raises $100M Series A for AI Inference

General Compute, a San Francisco start-up that rents out computing power for running AI models, has raised $100M in Series A funding. The company said in a blog post and a press release on 27 October that the round was co-led by Blackbird and Abstract, with Fuse Venture Partners, Village Global and others taking part.

TechCrunch reported that the round values General Compute at $800M post-money, up from $60M after a seed round in May. The company has not published a valuation.

A cloud built on Nvidia’s rivals

Most AI clouds are built around Nvidia GPUs. General Compute says it will “finance, deploy, and operate GPUs and purpose-built inference chips from Cerebras, SambaNova, and Positron AI”, and send each job to the hardware that runs it best. Customers can buy dedicated bare-metal servers or a “guaranteed tokens-per-second” service level, without owning the machines.

The company buys the hardware, places it in rented space in other operators’ data centres and runs it, handling “cooling, networking, energy and software”. Co-founder and chief executive Finn Puklowski said in the release: “The era of one-size-fits-all AI hardware is over.”

The round follows a $400M debt facility from Upper90 that the company announced earlier, and supply deals with all three chipmakers. Its first major deployments go live in the first quarter of 2027, it said, “with more than 50 MW of capacity in the pipeline”. Puklowski gave TechCrunch a larger figure: he expects to deploy 75 MW in 2027 and to have access to 200 MW of power.

According to TechCrunch, a deal with a large customer worth $100M in annual recurring revenue paved the way for the round. The customer was not named.

What the backers and chipmakers said

James Palmer of Blackbird said in the release that “a year or two ago, the bottleneck was silicon. Now it’s power, and both power and silicon supply chains are slow to scale.” David Kwon of Abstract said that within months of their first meeting the founders “had a big partnership with Cerebras and a debt facility of up to $400 million”.

The chipmakers also backed the company publicly. SambaNova chief executive Rodrigo Liang said: “Everyone in this market is selling speed. The real question is whether you can make money doing it.”

Why a cloud built on rival chips leans on debt

General Compute is selling distribution to chipmakers that lack Nvidia’s reach. Cerebras, SambaNova and Positron need someone to buy their systems, install them and sell the output to AI companies. General Compute takes on that job, but it has to pay for the hardware up front, which is why a debt facility sits alongside the equity. Puklowski told TechCrunch that raising “capital is going to be a full-time job for me, both on debt and equity”.

That model carries two risks. The first is concentration: if one chipmaker falls behind, the cloud is left with hardware customers no longer want. The second is demand. TechCrunch describes a single customer worth $100M a year but gives no figure for the rest of the company’s revenue, so it is unclear how much of the business rests on that one buyer. Other neoclouds face the same pressure, as CoreWeave’s push into Copenhagen and Asia shows.

There is a wider trend here. Inference, the work of running trained models, is drawing money towards chips designed for it, from Unconventional AI’s deal to buy NeuReality to start-ups building their own capacity, such as SWI Compute’s planned UK AI factory.

Two sets of capacity numbers to watch

The company’s own release speaks of more than 50 MW in the pipeline, while its chief executive told TechCrunch he expects 75 MW in 2027. Neither figure has been tied to named sites or a named customer. The Q1 2027 go-live is the first test, and the next debt raise will show whether lenders will keep funding hardware from Nvidia’s smaller rivals. TSN tracks this and other rounds on its AI funding rounds page.