General Compute raises $100M for a non-Nvidia inference cloud

General Compute raised a $100 million Series A to build an inference cloud on chips that are not Nvidia GPUs. The company buys hardware from SambaNova, Cerebras and Positron, puts it in leased space and old bitcoin-mining sites, and sells bare metal to the cloud companies that serve fast inference to developers. You will probably never sign a contract with General Compute. But the tokens your agents burn may run on its racks, and every new supplier like this pushes the price of fast inference down.

What actually happened

TechCrunch reported the round on October 27. As reported:

• The round. $100 million, led by Blackbird Ventures and Abstract Ventures, at an $800 million post-money valuation. The company was valued at $60 million after a seed round in May.
• The trigger. The round followed a deal with a large customer worth $100 million in annual recurring revenue. The customer is not named.
• The chips. A $400 million deal to buy SambaNova chips, which TechCrunch describes as the first debt financing for inference chips. Positron and Cerebras are now suppliers too.
• The capacity. General Compute expects to deploy 75 MW in 2027 and says it has access to 200 MW of power, more than it can currently fund and fill.
• The model. No giant new data centers. Smaller power sources, leased space, converted mining sites.

No pricing was given. The ARR figure and capacity plans are company statements reported by TechCrunch, not audited numbers.

Why it matters for your business

Inference is turning into a commodity with many suppliers. A year ago fast tokens meant one or two providers. Now specialized chip makers need distribution, and companies like General Compute exist to give it to them. As Blackbird's James Palmer put it to TechCrunch, Cerebras, Positron and SambaNova "all need distribution partners who can take them to scale."

For an operator, that means two things.

Prices will move, mostly down, and not evenly. The provider that is cheapest for your workload this quarter may not be next quarter. If your code calls one vendor's SDK directly, you cannot take the better deal without a rewrite.

Supply is still fragile. These are young companies financed with debt. The capital is "a full-time job," says CEO Finn Puklowski. If a provider stumbles, you want to switch in an afternoon.

So build for the swap. Put model calls behind one internal interface. Prefer OpenAI-compatible endpoints, which most inference clouds expose. Keep a small eval set per task, so a switch is a test run and not a leap of faith. And log cost per task, not cost per month, so you can see when a cheaper route appears.

Sources: TechCrunch.