General Compute raises $100M as neocloud bets on non-Nvidia inference chips

General Compute has landed $100 million, which TechCrunch describes as a Series A, to build out a cloud designed around AI inference rather than training. The awkward part is the timing.

Per TechCrunch, this is not exactly the friendliest week for a neocloud to announce a funding round. CEO and co-founder Finn Puklowski doesn’t seem bothered.

He told the outlet he is bullish on CoreWeave’s prospects. He also suggested Elon Musk is talking his book, trying to push down the prices of the GPUs Musk wants to buy.

How the money is structured

The research findings frame the deal as more than a plain equity check. They describe a July 2026 debt financing package of up to $400 million from Upper90 Capital Management.

The first draw under that facility is $100 million. Part of the collateral backing it is the company’s specialized inference chips.

That matters because lenders have historically been most comfortable with Nvidia gear as collateral. Accepting non-Nvidia silicon is a signal that capital markets are warming to alternatives.

Including the facility, General Compute’s total disclosed capital now reaches up to $415 million. That figure combines the $400 million debt package with an earlier $15 million seed round.

The seed closed in May 2026 at a post-money valuation of $60 million. Two months later, the company’s first debt draw alone is worth more than that entire valuation.

A different hardware recipe

Most neoclouds are built on Nvidia GPUs. General Compute took a different route.

The company runs air-cooled ASICs from SambaNova alongside AMD MI300X GPUs. An ASIC is a chip built for one job rather than many, like a racing bike compared with a mountain bike.

In this case, the job is inference. That is the moment a trained AI model actually answers a question, writes code, or takes an action.

General Compute is targeting workloads where speed is the product. Those include AI agents and voice applications, where a slow response breaks the experience entirely.

The company claims its inference hardware can run up to 16 times faster than GPU-based clouds in certain benchmarks. That is General Compute’s own figure, and the qualifier matters: “certain benchmarks” is doing real work in that sentence.

The air-cooling detail is less flashy but arguably just as important. Many high-end AI systems need liquid cooling, which requires specialized facilities and lengthy buildouts.

Air-cooled hardware can slot into existing colocation data centers. That means General Compute can potentially deploy capacity faster, without waiting on new construction or plumbing retrofits.

Who’s behind it

General Compute was founded in 2025 and is headquartered in San Francisco. Puklowski serves as CEO, and Jason Goodison is the company’s CTO.

The startup sits within the broader neocloud category. These are cloud providers that specialize in AI compute, positioning themselves as alternatives to the hyperscalers.

CoreWeave is the best-known name in that space. Puklowski’s vote of confidence in CoreWeave is notable, given that the two companies are chasing overlapping customers with very different hardware strategies.

What this means for the AI infrastructure market

The biggest takeaway is about who is willing to underwrite what. Upper90 accepting inference chips from SambaNova as partial collateral suggests lenders now see value beyond Nvidia’s ecosystem.

It also reflects a shift in where AI spending is heading. Training a frontier model is a huge but occasional expense, while inference runs every time someone uses that model.

Puklowski’s comments about Musk point to a pricing battle underneath all this. If major buyers are working to push GPU prices down, a cloud less dependent on Nvidia GPUs could be insulated from some of that volatility.

The 16x speed claim is also something to watch closely. Benchmarks chosen by a company tend to flatter that company, and real customer workloads will be the true test.