Five months ago, Baseten raised $300 million at a $5 billion valuation. That felt like a serious bet at the time. Now the startup is closing a $1.5 billion round at a valuation of up to $13 billion. No new AI model announced, no flashy launch event — just infrastructure for running AI models faster and cheaper than the competition.
Baseten's business is inference: taking trained AI models and executing them at scale. Customers including Cursor, Mercor, and OpenEvidence report cutting their compute bills by up to 30% compared to closed-source APIs like OpenAI's. The company doesn't build foundation models — it bets that open-source alternatives will win on cost, and that the real money will go to whoever runs those models most efficiently.
One quarter was enough to triple annualized revenue, from $200 million to $600 million. The $1.5 billion round is co-led by Altimeter Capital, Conviction, Spark Capital, Sands Capital, and Wellington Management. The deal features a split valuation: some investors entered at $11 billion, others at $13 billion. That structure typically signals fierce demand for allocation, not hesitation about the business.
The broader picture: open-source model training costs are dropping fast, but inference demand is growing exponentially. Every AI agent, every coding assistant, every search query needs compute to run. That gap — between a trained model and a live user request — is where Baseten operates. The tripling revenue suggests the gap is much more profitable than the industry assumed.
