While Everyone Fought Over GPUs, This Startup Went the Other Way and Raised $100M

Here's a funding story that cuts against the gloomy mood around AI infrastructure. General Compute, a San Francisco startup that rents out computing power for running AI models, has raised a $100 million Series A, co-led by Blackbird Ventures and Abstract Ventures. Its valuation is now $800 million, up from $60 million after a seed round in May.

The timing is bold. The week it announced, CoreWeave's stock had sunk to a 12-month low, and Elon Musk was predicting a wave of deaths and consolidation among smaller "neocloud" providers. CEO and co-founder Finn Puklowski isn't buying the panic. He told TechCrunch the fears are massively overstated and that people call him every week begging for more compute.

What makes General Compute different is its contrarian bet. Instead of chasing Nvidia GPUs like almost everyone else, it builds its cloud on chips designed specifically for inference, the step where a trained model actually answers you. It buys from Nvidia rivals SambaNova, Cerebras and Positron, and it struck a $400 million deal to finance SambaNova chips, described as the first debt financing of inference chips.

It's also skipping the hardest part of the business, building huge data centers from scratch. The company rents space in others' facilities, uses smaller power sources, and converts old bitcoin-mining sites into AI-ready sites. Puklowski expects to bring 75 megawatts of capacity online in 2027.

The raise was made possible by a single large customer deal worth $100 million in annual recurring revenue. General Compute mostly sells raw servers to clouds that specialize in fast inference for developers. Still, Puklowski admits the money hunt never stops, saying raising capital will be a full-time job for him. As SambaNova's CEO put it, everyone sells speed; the real question is whether you can make money doing it.