General Compute Launches Neocloud, Orders $300 Million in SambaNova Chips
Updated
Updated · TechCrunch · May 28
General Compute Launches Neocloud, Orders $300 Million in SambaNova Chips
2 articles · Updated · TechCrunch · May 28
Summary
$15 million in seed funding backed General Compute’s neocloud launch, with the startup saying its service is already the fastest for running MiniMax 2.7.
The company is betting on SambaNova’s inference chips rather than GPUs, targeting 600-700 tokens per second versus roughly 250 for GPUs, according to CEO Finn Puklowski.
$300 million of SambaNova SN50 chips are on order, and General Compute says it will be the first neocloud to deploy them.
Air-cooled, lower-power chips let the startup place hardware in existing facilities without major upgrades, including through colocation deals with data centers and crypto miners.
The push reflects a broader shift toward specialized inference infrastructure as AI agents and multi-model services make speed and token costs more important competitive factors.
With Nvidia's B200 boasting higher raw output, can SambaNova's specialized chips truly win the AI inference war on efficiency alone?
After a $15M seed round, is a $300M chip order a genius move or a startup's gamble of the century?
General Compute Bets $300M on SambaNova SN50: ASIC-Native Neocloud Targets Agentic AI Inference Revolution
Overview
AI infrastructure is rapidly transforming as the computational demands of AI models evolve. While GPUs have seen huge demand, there is a growing consensus that they are not ideal for the inference phase, which has different requirements than training. In response, a new class of chips is being designed specifically for AI inference. This industry shift is highlighted by major moves like Nvidia’s Groq transaction and Cerebras’ IPO. General Compute’s launch of its ASIC-native neocloud platform aligns with this trend, aiming to deliver more efficient, purpose-built solutions for the next generation of AI workloads.