General Compute has raised $100 million in a Series A round that values the inference-focused neocloud at $800 million, a more than thirteenfold jump from the $60 million valuation it carried out of a seed round in May. The round was led by Blackbird Ventures and Abstract Ventures, and it arrives at a moment when the neocloud sector is under intense scrutiny.

The trigger for the raise was a single customer commitment worth $100 million in annual recurring revenue, according to TechCrunch, which first reported the round on October 27. That anchor deal gave investors something the sector's sceptics say is in short supply: a paying customer at scale, signed before the new capital arrived.

General Compute sells bare-metal infrastructure for running AI models. Instead of competing for Nvidia's flagship graphics processors, it builds its cloud out of specialised inference chips from Nvidia rivals — SambaNova, Cerebras and Positron — matching workloads to the hardware that runs them best.

A bet against the GPU queue

The company's strategy is to zig where the rest of the industry zags. While hyperscalers and rival neoclouds chase scarce GPUs, General Compute focused on chips built specifically for inference — the work of running trained models for users, rather than training them. It struck a $400 million deal to buy chips from SambaNova, described as the first debt financing of inference hardware, and has since added Positron and Cerebras as suppliers.

Rather than building huge data centers from scratch, the company rents space inside other operators' facilities, taps smaller power sources, and converts former bitcoin-mining sites into places that run AI models. That keeps its build-out lean and sidesteps the years-long waits for grid connections that have hobbled larger rivals. Co-founder and chief executive Finn Puklowski expects to deploy 75 megawatts of capacity in 2027 and says the company has access to around 200 megawatts of power.

Raising into a neocloud storm

The timing is pointed. CoreWeave, the publicly traded neocloud, has seen its stock fall to a 12-month low, and Elon Musk has predicted that more than 200 neoclouds face "guppy-level death and consolidation" as debt-funded capacity collides with uncertain demand. Puklowski is unfazed, calling fears of a chip-led collapse "massively overstated" and arguing that Musk is talking his own book to drive down the prices of the GPUs he wants to buy.

"I have people calling me up every single week asking me to do TPUs, GPUs, any chip, to get them more compute," Puklowski told TechCrunch. He added that Anthropic's bullish IPO filing is "settling fears for today" across the sector.

Blackbird principal James Palmer, who co-led the round, said chipmakers need distribution partners able to scale in the physical world. SambaNova chief executive Rodrigo Liang framed the test bluntly: "Everyone in this market is selling speed. The real question is whether you can make money doing it."

What still has to be proven

The sceptical case has not gone away. The 13-fold valuation jump and the reported customer contract remain unproven until deployments and revenue arrive, and the model still demands heavy capital — Puklowski himself concedes that "capital is going to be a full-time job for me, both on debt and equity."

Still, the round is a data point against the simplest version of the neocloud-collapse story: even in a week when the sector's biggest name hit a 12-month low, investors put $100 million into a company betting that inference — not training — is where the money will be made, and that non-Nvidia silicon can make it profitably.