General Compute has raised $100 million in Series A funding at an $800 million valuation, betting AI infrastructure’s future is specialised inference chips from Nvidia’s rivals, not more Nvidia GPUs. The round, led by Blackbird Ventures and Abstract Ventures, was unlocked by a deal with a large customer worth $100 million in annual recurring revenue, according to TechCrunch. The startup was valued at just $60 million after a May seed round, making the jump to $800 million in five months one of the sharpest repricings of the year.
A Neocloud That Zigs Where Others Zag
General Compute calls itself a neocloud, a company that rents out computing power for AI, but its strategy is deliberately different from the giants of that sector. While most neoclouds chased Nvidia GPUs for training models, General Compute focused on chips built for inference, the work of actually running models for users in production. Its first big bet was a $400 million deal to purchase new chips from SambaNova, described as the first debt financing of inference chips, and it has since added Positron and Cerebras to its portfolio of chip suppliers.
Instead of building huge data centres from scratch, General Compute rents space in other companies’ data centres, uses smaller power sources, and converts former bitcoin-mining sites into places that run AI models. It will not compete with hyperscalers any time soon, but when it has chips, it will have somewhere to plug them in.
Chief executive Finn Puklowski expects to deploy 75 megawatts of computing capacity in 2027, with access to 200 megawatts of power, more than the company can probably fund and equip. Turning that power into revenue still requires buying and installing a lot of expensive silicon.
Announced In The Worst Week For Neoclouds
The timing could hardly have been worse, at least on the surface. CoreWeave, the publicly traded neocloud that rents out computing power for AI, has seen its stock fall to a 12-month low. Elon Musk, meanwhile, has been flexing $40 billion in debt for buying computing capacity and predicting that more than 200 neoclouds are going to see what he called guppy-level death and consolidation. It is not, as TechCrunch put it, exactly the ideal week to announce a Series A for a neocloud.
Puklowski is unfazed. He is bullish on CoreWeave’s chances, and thinks Musk is just talking his book to drive down the prices of the GPUs he wants to buy. Fears of a chip-led collapse are massively overstated, he said. “I have people calling me up every single week asking me to do TPUs, GPUs, any chip, to get them more compute.” Just as important, he said, Anthropic’s bullish IPO filing, which showed first-half revenue of about $16.3 billion, is settling fears for today.
That confidence is backed by a customer, even if the customer is unnamed: TechCrunch does not name the $100 million customer, so the round’s anchor is a contract the public cannot verify. Blackbird principal James Palmer said his firm’s confidence comes in part from its investment in Baseten, an inference cloud company whose growth it has watched closely over the past 10 months.
Why Inference Chips, And Why Now
The bet rests on a shift in where AI compute is used. Training happens once; inference, running a model for millions of users daily, happens constantly and is becoming the largest workload in computing. Inference chips from Cerebras, SambaNova and Positron promise faster, more efficient tokens than general-purpose GPUs, but deploying them is not as simple as renting capacity on a major cloud.
Alternative silicon needs specialised data centres, cooling, networking and a custom software stack. General Compute handles that complexity, providing the bare metal, raw servers with no software layered on top, that developer-facing inference clouds can build on.
SambaNova chief executive Rodrigo Liang put the test bluntly: “Everyone in this market is selling speed. The real question is whether you can make money doing it.” General Compute’s deployments are expected to go live in early 2027.
The AI buildout is increasingly funded by debt and leases, not just equity, as in our report on Moody’s warning over AI’s $1 trillion buildout and our report on SpaceX’s $40 billion Apollo chip lease. Puklowski is candid: capital is going to be a full-time job for me, both on debt and equity, he told TechCrunch.
Why It Matters
This matters because it is a test of whether the AI infrastructure market has room for specialists, or whether only hyperscalers and the largest neoclouds survive. If General Compute can turn alternative silicon into a profitable, reliable inference cloud, it will prove that inference is a distinct market with its own economics, not just a smaller version of training. That would be good news for Cerebras, SambaNova and Positron, which all need distribution partners that can take them to scale and navigate the complexity of the physical world, as Palmer put it.
If it cannot, the $800 million valuation will look like a product of a moment when any company with a large contract and a power story could raise at a steep multiple. The 13-fold jump from May leaves little room for a slow start.
Background
General Compute was founded by Finn Puklowski and builds bare-metal inference infrastructure for developer-facing cloud companies. Its earlier $400 million SambaNova purchase was financed with a debt facility, and its first major deployments are expected in early 2027.
Source: TechCrunch report
0 Comments