General Compute Raises $100M at $800M to Run Inference on Cerebras, SambaNova and Positron

General Compute, a San Francisco inference cloud built on non-NVIDIA inference chips, raised a $100 million Series A co-led by Blackbird and Abstract Ventures, the company said in its October 27 announcement. TechCrunch reports the round values the company at $800 million post-money, up from $60 million after a May seed.

The numbers

• $100M Series A, co-led by Blackbird and Abstract, with Fuse Venture Partners, Village Global, Bold Capital, QED Investors, 645 Ventures, Endeavor Catalyst and others, per the press release.
• $800M post-money valuation, per TechCrunch; the company’s release does not state a valuation.
• $400M debt facility from Upper90, previously announced, to finance chip purchases.
• 50+ MW of capacity in the pipeline, with first major deployments live in Q1 2027.
• $100M+ in annual recurring revenue from one large customer deal, per TechCrunch.

The bet: inference without GPUs alone

General Compute finances, deploys and operates GPUs alongside purpose-built inference chips from Cerebras, SambaNova and Positron AI, then routes each workload to the hardware that runs it best. Customers buy either dedicated bare metal or a guaranteed tokens-per-second service-level agreement, without procuring hardware themselves. “The era of one-size-fits-all AI hardware is over,” co-founder and CEO Finn Puklowski said in the release.

No data centers of its own

The company buys the hardware and places it in rented colocation space, handling cooling, networking, energy and software itself. TechCrunch reports it also converts former bitcoin-mining sites and leans on smaller power sources, and that Puklowski expects about 75 MW deployed in 2027 with access to roughly 200 MW of power. “A year or two ago, the bottleneck was silicon. Now it’s power,” said James Palmer of Blackbird.

Why the timing matters

The raise lands in a rough week for neoclouds: TechCrunch notes CoreWeave shares hit a 12-month low as Elon Musk predicted “guppy-level death and consolidation” across 200+ GPU clouds. General Compute’s answer is specialization in inference silicon rather than racing hyperscalers on GPU scale.

What we are watching

None of the big deployments are live yet; the first arrive in Q1 2027. The release’s speed claims are inconsistent, citing both “up to 40x” and “up to 20x” faster than GPUs, with no published benchmark. The test is whether the Cerebras, SambaNova and Positron capacity turns on on schedule and the $100M ARR contract converts into delivered tokens.