General Compute raises $15M to build inference neocloud on SambaNova chips

General Compute, a new AI inference neocloud, has raised $15 million in seed funding at a $60 million valuation, led by FUSE VC with participation from Carya Venture Partners and Village Global Ventures. The startup is betting on specialized inference chips over traditional GPUs to deliver faster, cheaper AI processing.

Founded by CEO Finn Puklowski and CTO Jason Goodison, General Compute is building its infrastructure around SambaNova's upcoming SN50 chips, claiming to have $300 million worth on order and positioning itself as the first neocloud to deploy them. SambaNova's architecture promises 600-700 tokens per second—roughly three times the output of standard GPUs—using more flexible, memory-rich designs optimized specifically for inference workloads rather than training.

A key advantage of the SN50 chips is that they are air-cooled rather than water-cooled, consuming less power and enabling installation in existing data center facilities without costly infrastructure upgrades. General Compute is pursuing colocation deals with traditional data center providers and crypto miners looking to repurpose their infrastructure.

The startup launched its cloud offering last week, claiming the fastest inference speeds for MiniMax 2.7, a powerful open-source LLM. Puklowski aims to compress hour-long coding agent workloads into five to ten minutes, and make real-time audio agents for customer service more economical.