General Compute Raises $100 Million to Run AI Clouds on Chips That Are Not Nvidia's

General Compute has raised a $100 million Series A co-led by Blackbird, and the timing is deliberate. The round lands while the neocloud narrative sits at its gloomiest: CoreWeave has traded down to a 12-month low, and Elon Musk, armed with $40 billion in debt-backed buying power, has predicted that more than 200 neoclouds face guppy-level death and consolidation.

Founder Puklowski says he is not fazed. He told TechCrunch he is bullish on CoreWeave, reads the prediction from Musk as talking his book to drive down the price of the GPUs he wants to buy, and calls fears of a chip-led collapse massively overstated. Anthropic's bullish S-1, he added, was settling fears for today. On Tuesday CoreWeave actually bounced 5.2 percent, and AMD rose 4 percent, as AI names lifted the broader market.

The niche is precise: build the cloud out of specialized inference chips made by the competitors of Nvidia rather than the GPUs everyone else is fighting over. General Compute struck a $400 million deal to buy new chips from SambaNova, a transaction TechCrunch reports as the first debt financing of inference chips, and it has added Positron and Cerebras to its supplier list. It sells bare metal, raw servers with no software layered on top, to cloud companies that specialize in high-speed inference for developers.

Physically, the company avoids the data center buildout trap. It rents space inside the facilities of other companies, uses smaller power sources and converts former bitcoin-mining sites into locations for running AI models. A deal with a large customer worth $100 million in annual recurring revenue helped clear the way for the Series A. The company was valued at just $60 million after a May seed round, and Puklowski expects 75 megawatts of computing capacity deployed in 2027, with access to roughly 200 megawatts, more, he says, than the company can probably fund and get hardware for. Capital, he told TechCrunch, is going to be a full-time job, on both debt and equity.

Blackbird principal James Palmer, who co-led the round, said conviction came partly from proximity to Baseten, an inference cloud company, and from watching demand for open-weight inference and for companies owning their own intelligence. He argues Cerebras, Positron and SambaNova all need distribution partners that can take them to scale. SambaNova chief executive Rodrigo Liang framed the economics bluntly: everyone in this market is selling speed, and the real question is whether you can make money doing it.

What it means: the neocloud market is bifurcating. The giants chase GPU scale for training and prefill, while a long tail specialises in the decode side, the token generation that determines how fast agents respond. Debt-financed inference chips are a new capital structure for that bet, and the wager underneath General Compute is that agent workloads make efficient inference, not raw FLOPS, the scarcest resource in AI.