General Compute raises $15M to build inference-focused neocloud

General Compute has raised a $15 million seed round at a $60 million post-money valuation. The startup is building a neocloud—a company that rents out AI processing power—specializing in inference, which is the phase of AI where a model is actively generating responses rather than being trained. The funding round was led by FUSE VC, with participation from:

• Carya Venture Partners
• Village Global Ventures

The company is positioning itself as the first neocloud to deploy SambaNova’s SN50 chips, with $300 million of the hardware on order. SambaNova is an Intel-backed chipmaker focused on inference that has fallen a bit out of the Silicon Valley conversation. SambaNova claims its architecture outperforms GPUs and other specialized chips built by Groq or Cerebras. According to SambaNova, its chips will generate 600 to 700 tokens per second, compared to about 250 tokens per second for GPUs. Because these chips are air-cooled and consume less power, they can be installed in existing data centers without requiring new infrastructure.

The launch comes amid a broader industry push for specialized inference infrastructure. Recent transactions highlight this shift, including Nvidia’s $20 billion transaction with Groq in December, Cerebras’ $57 billion IPO last week, and a $113 million Series B raised by OpenRouter last week. General Compute launched its cloud offering last week, claiming it is already the fastest at running the model MiniMax 2.7.

Joe Hasselmann, a venture investor and founder of Evercrest Capital Partners, who invested in Groq in 2021, made General Compute one of his firm’s first investments. Hasselmann noted that as much as General Compute is making a bet on SambaNova, SambaNova is also making a bet on General Compute, requiring high-growth customers to deploy its hardware. Finn Puklowski, CEO of General Compute, emphasizes that faster inference is critical as workloads transition. “If you use ChatGPT and it gives you 50 tokens per second, that’s still a heck of a lot faster than we can read, Now that things have moved to agent-to-agent, where agents are out there reading on our behalf or pinging databases, they need to go faster,” Puklowski said.

Why it matters

General Compute’s $15 million seed round highlights the growing demand for specialized inference infrastructure, with the company betting on SambaNova’s chips to solve cooling and power constraints in data centers.