CoreWeave, a publicly traded neocloud company providing compute leasing for AI, has seen its stock price fall to a 12-month low; meanwhile, Elon Musk is raising $40 billion in debt to fund compute purchases and predicts “hundreds of neoclouds will face minnow-level deaths and consolidations.” Against this backdrop, announcing a Series A round for a neocloud is poorly timed.
However, Finn Puklowski, co-founder and CEO of General Compute, isn’t concerned. He is optimistic about CoreWeave’s prospects and believes Musk is merely “speaking out to support his position” in an effort to lower the price of the GPUs he wants to buy. He dismisses worries about a chip-driven crash, calling such concerns “greatly exaggerated.” “I get calls every week asking me to provide TPUs, GPUs—any chip—that can give them more computing power.”
Equally important, he said that Anthropic’s optimistic S-1 filing “temporarily eased market concerns.”
General Compute identified a niche market by building its cloud services around specialized inference chips launched by NVIDIA's competitors. The company entered into a $400 million agreement with SambaNova to purchase the new chips—the first debt-financed purchase of inference chips—and subsequently added Positron and Cerebras to its list of chip suppliers.
A deal with a major client generating $100 million in annual recurring revenue has paved the way for the company’s $100 million Series A funding round, led by Blackbird Ventures and Abstract Ventures. General Compute was valued at $60 million following its seed round in May and now has a post-money valuation of $800 million for investors.
The startup’s strategy is “when others go right, it goes left.” While others chase GPUs, it focuses on chips designed for inference; instead of investing in complex engineering to build large data centers, it uses smaller power sources, rents space in other companies’ data centers, and repurposes former Bitcoin mining facilities to run AI models.
This means General Compute won’t quickly rise to become a top player in neocloud or directly compete with hyperscale cloud providers. But it also means it doesn’t have to contend with the constraints of building data centers from scratch. As long as it secures chips, General Compute can always find access to power. Puklowski expects the company to deploy 75 megawatts of computing capacity by 2027 and believes it could secure up to 200 megawatts of power—“which may exceed the scale we can fund or acquire hardware for.”
However, this plan has its limits. Puklowski said the fundraising effort isn’t over—“For me, capital will be a full-time job, whether through debt or equity financing,” he told TechCrunch.
This startup has already identified key customers among cloud companies focused on providing high-speed inference services for developers; General Compute supplies them with bare metal—raw servers without any software layer—to help bring their products to market. James Palmer, partner at Blackbird who co-led the investment, said the firm’s confidence in General Compute is partly rooted in its prior investment in the inference cloud company Baseten.
Palmer told TechCrunch: “Over the past 10 months, we’ve closely followed their growth trajectory and observed Baseten’s explosive growth, along with rising demand for open-source weight inference and proprietary intelligence from the company.” Regarding General Compute, he said: “Cerebras, Positron, and SambaNova all need distribution partners that can help them scale and handle real-world complexity.”
Executives from all three chip manufacturers expressed their appreciation for General Compute to TechCrunch, each highlighting the importance of their companies focusing on making the AI cloud computing economic model viable. Rodrigo Liang, CEO and co-founder of SambaNova, said, "Everyone in this market is selling speed. The real question is, can you make money from it?"