General Compute raises $100M Series A on inference-chip bet as neocloud fears mount

General Compute has raised a $100 million Series A led by Blackbird Ventures and Abstract Ventures, valuing the neocloud startup at $800 million post-money — up from a $60 million valuation after its May seed round. The announcement lands in one of the worst weeks the AI-computing rental market has seen in months.

CoreWeave, the publicly traded neocloud that rents out computing power for AI workloads, has watched its stock price fall to a 12-month low. Elon Musk is meanwhile flexing $40 billion in debt to buy computing capacity and predicting that "200+ neoclouds are going to see guppy-level death and consolidation." Against that backdrop, a Series A for a neocloud is either contrarian or reckless. General Compute's investors are betting on the former.

The round was paved by a deal with a large customer worth $100 million in annual recurring revenue, according to the company. That contract matters more than the valuation multiple: it is the clearest evidence so far that the startup's unusual chip strategy produces revenue, not just capacity promises.

Why bet on Nvidia's rivals?

General Compute has built its cloud out of specialized inference chips made by Nvidia's competitors rather than the GPUs (the general-purpose accelerators that dominate AI training) that most rivals chase. The company struck a $400 million deal to purchase new chips from SambaNova — the first debt financing of inference chips on record — and has since added Positron and Cerebras to its supplier portfolio.

The logic is straightforward. Inference (running trained AI models to produce outputs, as opposed to training them) is a price-sensitive, high-volume business where cheaper silicon can win on unit economics. Executives at all three chipmakers praised General Compute to TechCrunch, each emphasizing the startup's focus on making the economics of AI cloud compute actually work.

"Everyone in this market is selling speed," SambaNova CEO and co-founder Rodrigo Liang said. "The real question is whether you can make money doing it."

That quote cuts both ways. Selling speed is what CoreWeave and the hyperscalers do at scale; making money doing it is the question none of them has definitively answered in public financials. Liang is endorsing General Compute's approach, but he is also a supplier with $400 million of chips riding on the startup's success. Take the praise with the appropriate discount.

What does the CEO say about the market panic?

General Compute co-founder and CEO Finn Puklowski is unfazed by the week's gloom. He is bullish on CoreWeave's chances, and he thinks Musk is simply talking his book to drive down the prices of the GPUs he wants to buy.

Fears of a chip-led collapse are "massively overstated," Puklowski 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 says, Anthropic's bullish S-1 (the registration filing a company files before going public) is "settling fears for today." His implication: demand for AI compute at the frontier-lab level remains strong enough to support the supply chain beneath it.

The caveat is obvious. Demand calls to a startup CEO are not contracted revenue, and "settling fears for today" is doing a lot of work in that sentence. Puklowski himself does not claim the financing question is closed.

"Capital is going to be a full-time job for me, both on debt and equity," he told TechCrunch. That admission is the honest one: a $100 million equity round is thin relative to a $400 million chip purchase agreement and the multi-hundred-megawatt ambitions behind it.

How does the company deploy without building data centers?

General Compute's strategy is to zig where others zag. Instead of undertaking the complex, capital-intensive effort of building huge data centers, it uses smaller power sources, rents space in other companies' data centers, and converts former bitcoin-mining sites into facilities that run AI models.

The approach means General Compute will not catapult into the top ranks of neoclouds or compete with hyperscalers any time soon. But it also frees the company from the constraints of building data centers from scratch. When General Compute has chips, it has somewhere to plug them in.

The numbers behind that claim are concrete. Puklowski expects the company to deploy 75 megawatts of computing capacity in 2027, and he suspects he will have access to 200 megawatts of power — "more than we can probably fund and get hardware for." Read that admission carefully: power is not the bottleneck. Capital and hardware supply are. That is a more manageable problem than a power interconnection queue, but it is still a problem.

The plain takeaway: General Compute has traded scale for speed and optionality, and its 2027 targets depend on raising far more money than it has raised to date.

Who are the customers?

The startup has found its key customers in cloud companies that specialize in offering developers high-speed inference. General Compute provides the bare metal — raw servers with no software layered on top — and those customers build their inference offerings on top of it.

That positioning explains why investors see the company as infrastructure for the AI middle layer rather than a CoreWeave competitor. James Palmer, a principal at Blackbird who co-led the round, said his firm's confidence in General Compute comes in part from its investment in Baseten, an inference cloud company.

"We've been proximate to their journey over the last 10 months and observing the wild growth story that [Baseten] has gone on, and the demand for open weight inference, and for companies owning their own intelligence," Palmer told TechCrunch. Open weight models are AI models whose parameters are publicly downloadable, letting companies run them on their own hardware instead of paying an API provider.

Palmer's case for General Compute itself is about distribution for the chipmakers: "Cerebras, Positron, SambaNova, all need distribution partners who can take them to scale, who can navigate the complexity of the physical world."

That is the sharpest framing in the story. Nvidia's rivals have struggled for years not on chip design but on getting their silicon deployed at scale. General Compute is positioning itself as the deployment layer for that hardware — a picks-and-shovels role that does not require winning the model race.

Why this matters now

The Series A arrives at a moment when the market narrative has flipped from compute scarcity to compute glut. CoreWeave's 12-month-low stock price and Musk's prediction of mass neocloud extinction are the bear case in two data points. General Compute's $100 million ARR customer, its $400 million SambaNova deal, and Anthropic's IPO filing are the bull case in three.

Both cannot be fully true indefinitely. If inference demand keeps compounding and non-Nvidia chips hit workable economics, the bare-metal, rented-space, converted-mining-site model is a cheap way to capture it. If Musk's consolidation prediction proves right, a $800 million-valuation startup carrying debt-financed chip purchases will feel it quickly.

Puklowski's own words set the test for the next 18 months: deploying 75 megawatts by 2027 while admitting that capital remains "a full-time job" means the Series A is a waypoint, not a destination.