A young cloud provider that avoids building around NVIDIA alone has raised a large early round. General Compute announced on October 27 that it closed a $100 million Series A round, led jointly by Blackbird and Abstract, with Fuse Venture Partners, Village Global and a long list of other funds and angels joining. According to TechCrunch, the deal puts the company's post-money valuation at $800 million, compared with $60 million after a seed round in May.
The company calls itself a heterogeneous inference cloud. It buys racks of accelerators, places them in colocation data centers and handles power, cooling, networking and software, then sells the result either as dedicated bare-metal servers or under a service-level agreement that guarantees a number of tokens per second. Its website describes a split design in which GPUs handle prefill, the step that reads a prompt, while ASICs from Cerebras, SambaNova, Positron AI and d-Matrix take over decoding, the step that writes the answer. Customers do not have to buy or operate any of that hardware themselves.
Scale is still ahead of it. The announcement says the first major deployments go live in the first quarter of 2027 and that more than 50 megawatts of capacity sit in the pipeline; the company website separately lists more than 30 megawatts. Chief executive Finn Puklowski told TechCrunch he expects to deploy 75 megawatts during 2027 and may have access to around 200 megawatts of power, which he said is more than the company can probably fund and fill with hardware.
Financing is central to the model. General Compute says Upper90 recently agreed to lend it as much as $400 million to pay for its buildout. TechCrunch described a $400 million arrangement tied to buying new SambaNova chips and called it the first debt financing of inference chips. The two accounts may describe the same facility from different angles, but the public materials do not spell out how they fit together. TechCrunch also reported that an agreement with one large customer, valued at $100 million a year in recurring revenue, cleared the way for the round; the company's release does not mention that contract.
The timing is awkward for the sector. TechCrunch noted that CoreWeave's shares had dropped to a 12-month low and that Elon Musk had predicted a shakeout among smaller GPU clouds. Puklowski called fears of a collapse overstated. He argues that demand for any available chip remains strong and that Anthropic's IPO filing has calmed investors for now. Blackbird principal James Palmer told the publication that his firm's view was shaped partly by its stake in inference provider Baseten and by demand for open-weight models.
Chipmakers have an obvious interest in companies like this. Cerebras, SambaNova and Positron each supplied supportive statements for the announcement, and all three need partners able to finance, house and operate their systems at volume. General Compute's approach of renting space in existing facilities and using converted bitcoin-mining sites, as TechCrunch described it, avoids the delays of building large campuses but also caps how quickly it can grow.
Several figures remain company claims. The release promises results up to 20 times faster in one passage and up to 40 times faster than GPUs in its subheading, without published benchmarks behind either number. The valuation, the $100 million customer contract and the 75-megawatt target come from TechCrunch's reporting and interviews rather than from the release. Puklowski also told TechCrunch that raising further debt and equity will remain a full-time job, a reminder that the hardware still has to be financed before any of the capacity exists.