Baseten Seeks $2 Billion at Up to $30 Billion, and the Inference Layer Is Getting Lab-Sized Checks

The money is chasing the part of the AI stack that actually runs the models. Baseten, a San Francisco company that operates an inference platform for enterprises, is seeking roughly $2 billion in new funding, with a valuation structured to climb from $26 billion in the first tranche to as much as $30 billion in the second, according to a Bloomberg report published Friday. The company has not announced the round, and the terms could still change.

If completed at the top of the range, the valuation would be more than double the $13 billion Baseten reached in June, when it raised $1.5 billion. Sands Capital, an existing investor, is expected to lead the new round. Bloomberg's report did not name the other participants, and Baseten did not respond to requests for comment.

Baseten sits between model builders and the applications that use them: customers such as Notion and Cursor run open and custom models on its infrastructure rather than operating their own GPU fleets. The company, founded in 2019, has previously said its platform handles more than a billion inference requests a day, a figure that reflects how much production traffic has moved to dedicated serving layers.

The round is the latest sign that private capital has rotated from training compute toward inference. Baseten's closest comparable, Fireworks AI, raised $1.5 billion at a $17.5 billion valuation in July, and Nvidia has backed both companies as it pushes more enterprise workloads onto its chips. The June-to-October jump in Baseten's ask mirrors the broader repricing of anything that touches serving capacity.

It also lands at an odd moment. Public markets spent the last week of October punishing AI-linked stocks over spending concerns, and OpenAI reportedly cut its own funding target from $30 billion to about $20 billion at a lower valuation. Private infrastructure deals are still getting done, but investors are negotiating harder on price and structure - hence Baseten's two-tranche design, which lets investors step up only as milestones are hit.

For now the signal is straightforward: even as the market questions AI capex at the model labs, the serving layer that turns tokens into revenue is still viewed as the safer bet. Whether a $30 billion valuation for a four-year-old infrastructure company survives contact with an eventual IPO is the question the second tranche is quietly hedging against.