Baseten Raises $1.5B Series F at a $13B Valuation

Baseten, the San Francisco company whose platform runs other companies' AI models in production, has raised a $1.5 billion Series F at a valuation of up to $13 billion, it announced on June 22. The round — among the largest AI-infrastructure raises on record — was led by Altimeter Capital, with Conviction and Spark Capital co-leading and Sands Capital and Wellington Management joining alongside existing backers including IVP and Greylock.

The capital came in two tranches, priced at $13 billion and $11 billion, and marks a roughly 160% jump in Baseten's valuation in under five months: the company had closed a $300 million Series E at a $5 billion valuation only in early 2026. Baseten says it now handles more than one billion inference calls a day across 87 compute clusters on 18 different clouds, with revenue up roughly 20-fold year over year.

"AI inference is going to be one of the largest, if not the largest market in AI in the world," said Altimeter partner Apoorv Agrawal. Inference — the work of actually running a trained model to answer a prompt, generate code or analyze a document — is where an AI product's cost and latency live, and it scales with every user rather than every training run. As model quality converges, serving those models quickly and cheaply has become its own battleground, contested by hyperscalers, GPU clouds and specialists like Baseten.

Baseten's pitch is that it abstracts that problem away: customers deploy open-source or custom models and Baseten handles the GPUs, autoscaling, routing and reliability across a patchwork of providers. "We have to diversify our compute sources," chief executive Tuhin Srivastava said, adding that the company has "been very supportive of open-source models." The new money will fund its infrastructure, engineering and research teams, and Baseten has said it intends to triple its headcount in 2026.

The raise underscores how the AI build-out is splitting in two. Capital is flowing not only to the labs that train frontier models, but to the unglamorous layer that serves them at scale — the place where rising usage turns directly into compute bills. With a $13 billion price tag, investors are betting that owning the serving layer, and the freedom to route across clouds and open models, is a durable business in its own right.