Baseten Seeks $1.5B Raise at Up to $13B Valuation, Betting on Open-Source Model Inference Infrastructure
Claire Weston
AI inference infrastructure company Baseten is closing a $1.5 billion round at up to $13 billion valuation — a bet that open-source models are eroding the pricing power of closed-source giants, and enterprises need a cheaper way to run them.
What does Baseten actually do?
Baseten builds the "connective tissue" for open-source models — the models themselves are free, but running them requires compute. Baseten layers inference software on top of 20 cloud providers, giving customers the tools to run, optimize, and fine-tune open-source models.
In plain terms = the model is a free engine, but you still need the drivetrain and chassis to drive it. Baseten sells the chassis.
Current customers include AI coding tool Cursor, hiring platform Mercor, and medical AI company OpenEvidence.
Why can it raise this much right now?
The core logic: the quality gap between open-source and closed-source models keeps narrowing, and enterprises are looking for ways around the high costs of OpenAI and Anthropic.
This means → open-source is no longer "cheap but far worse." It is good enough and far cheaper — and that opens a large market.
The round is co-led by Altimeter Capital, Conviction, Spark Capital, Sands Capital, and Wellington Management, using a two-tier valuation structure — some investors enter at $11 billion, others at $13 billion.
Why does Wellington's entry matter?
Wellington Management had never invested in AI inference before this round.
This means → traditional institutional capital now views AI inference as a standalone, long-duration bet — not just a supporting act for foundation-model companies.
Wellington investor Oz Nur's thesis: even if open-source models always trail frontier models by a few months, they already serve the majority of use cases. Enterprises can reserve closed-source budgets for tasks that truly need peak performance.
How much can enterprise customers actually save?
Baseten CEO Tuhin Srivastava cited a customer case: switching to open-source models cut the cost of a specific task to 30% of the closed-source alternative.
In plain terms = the same job that cost a dollar now costs thirty cents.
Customers today commonly run a hybrid of open-source and closed-source models. Even migrating only some tasks delivers a meaningful cut in total AI spend.
What does the open-source model landscape look like now?
The most popular open-source models currently come from China — DeepSeek and Moonshot AI's Kimi.
On the U.S. side, Nvidia recently launched an open-source model family called Nemotron.
This reflects an accelerating supply side, with both the U.S. and China pushing hard. For a "picks-and-shovels" company like Baseten, more models and fiercer competition among them actually makes its business stronger.
Content is for reference only, not financial advice.