Baseten Raises $1.5B at Up to $13B Valuation, Betting on Open-Source AI's Rise

Baseten, a startup that provides inference infrastructure for open-source AI models, is closing a $1.5 billion fundraising round in a dual-tier structure — some investors buying in at an $11 billion valuation, others at $13 billion. The round is co-led by Altimeter Capital, Conviction, Spark Capital, Sands Capital, and Wellington Management, with the latter making its first-ever investment in the AI inference market through the deal.

The company's core pitch is simple: enterprise customers are increasingly fed up with the cost of relying on closed-source giants like OpenAI and Anthropic, and open-source models from the likes of DeepSeek, Moonshot AI's Kimi, and Nvidia's new Nemotron family have closed the quality gap enough to handle a wide range of real-world tasks. Baseten sits between those models and the computing power needed to run them, having built an inference software layer on top of capacity sourced from 20 different cloud providers. That lets customers like Cursor, Mercor, and OpenEvidence run, optimize, and fine-tune open-source models on their own data — without having to build the plumbing themselves.

The cost savings can be dramatic. CEO Tuhin Srivastava cited one customer who performed a specific task at just 30% of what it would have cost using a closed-source model. Most Baseten customers use a hybrid approach — open-source for routine workloads, frontier models only where peak performance is essential — and even partial adoption generates meaningful savings at scale.

Baseten's rise reflects a broader inference boom. Cerebras, which designs chips purpose-built for inference, hit a nearly $50 billion market cap after a blockbuster IPO in May. Fireworks AI raised at a $4 billion valuation in October, and autonomous-coding startup Factory reached $1.5 billion in April. Wellington Management's involvement — an institution known for long-horizon, lower-risk bets — signals that inference infrastructure is being viewed as durable infrastructure, not speculative momentum.

The strategic question hanging over the space is how fast open-source models continue to improve. Wellington investor Oz Nur framed it as a tiering play: open-source models, even if a few months behind the frontier, are already good enough for a large slice of enterprise use cases. As that slice expands, so does Baseten's addressable market — and its leverage against the closed-source incumbents.