General Compute Bets $300M on SambaNova as the Next AI Chip Leader

General Compute, a new cloud provider for deploying AI models, has bet on SambaNova as the next champion in the race for specialized chips. The company ordered $300M in SambaNova SN50 processors for inference—a volume that positions it as the first major buyer of these chips among neo-cloud providers.

An Unusual Bet on a Safe Choice

General Compute just raised $15M in a seed round from FUSE VC at a $60M valuation. Founders Finn Puklowski (CEO) and Jason Goodison (CTO) chose not to compete for scarce Nvidia GPUs, instead developing their own strategy. Rather than fighting for capacity with Groq or Cerebras, the company bet on the Intel-backed startup SambaNova—less famous but promising.

SambaNova: Tripled Processing Speed

The technical numbers are compelling. SambaNova SN50 generates 600–700 tokens per second for inference, approximately three times faster than standard GPUs (around 250 tokens/sec). The architecture promises to outperform not just GPUs, but also competing specialized chips from Groq and Cerebras—though in the narrow scenario of inference, where speed is needed but training new models is not required.

• Tokens/sec: 600–700 vs 250 on GPU
• Air cooling instead of liquid cooling
• Power consumption one level below competitors
• Compatibility with existing data center infrastructure

Air Instead of Water

Operational advantages are often underestimated. SambaNova chips use air cooling, not liquid cooling—critical for data centers that aren't prepared for costly infrastructure upgrades. Installing a chip in an existing server rack that doesn't require liquid cooling is a very practical solution for cloud operators.

Why This All Matters

The inference stage in AI is a separate challenge from the training phase. When a model is already trained, you need to minimize latency in user responses and maximize throughput. Against this backdrop, Cerebras recently went public at a $57B valuation, and Groq was acquired by Nvidia for $20B in December. In other words, the market believes that specialized chips for inference are not a side item but a serious investment.

What This Means

General Compute chose SambaNova not because it's the most hyped brand. The company is investing in architecture that works better for a specific task—cloud inference. If SambaNova reliably delivers on its promised numbers, it could shift which chips operators choose. It's the same move CoreWeave made with Nvidia, but in the opposite direction—with a less established but potentially more efficient player.