Two Companies You've Probably Never Heard Of Are Suddenly Worth $41 Billion Combined
Modal Labs tripled its valuation to $15 billion in four months and Baseten doubled to $26 billion in three, because investors just noticed that running AI models costs more than building them.
Modal Labs was worth $4.65 billion in May. It is now, per people familiar with the ongoing talks, in discussions to raise money at roughly $15 billion — a tripling in four months, with Accel reportedly lined up to lead the round. Baseten was worth $13 billion in June. It is now in talks that could value it at $26 billion, a doubling in three. Neither company has a product most people could name, let alone use. Both make their money running other people’s AI models after they’ve already been trained — the unglamorous, expensive, absolutely unavoidable part of the business that nobody puts in a demo video.
For a while, the entire AI conversation was about training: bigger clusters, longer runs, the next model that would finally justify the last one’s price tag. Inference — the part where a trained model actually answers your question, writes your code, or books your flight — was treated as a rounding error, the thing that happens after the interesting work is done. That ratio has flipped. Inference now accounts for roughly two-thirds of AI computing demand, up from about a third in 2023, and industry estimates put the inference market at $103.7 billion this year, growing to $312.6 billion by 2034. Modal’s own numbers illustrate the shift bluntly: annualized revenue went from $119 million at the end of 2025 to $300 million by this past April, a fivefold jump in six months, driven largely by AI coding tools that need to run inference continuously, not occasionally.
Neither company builds models. Neither company, strictly, builds chips. What Modal and Baseten sell is the plumbing in between — infrastructure that lets a company deploy a model into production without hand-building a serving stack, managing GPU capacity, or hiring a team whose entire job is keeping inference fast and cheap at scale. It’s the AI equivalent of a utility company, which is either the least exciting pitch in venture capital or the most reliable one, depending on how the last decade of cloud computing treated your portfolio. Given that AWS, and not any particular app built on it, ended up being the trade of the last twenty years, “least exciting” and “most reliable” may just be the same sentence twice.
The two aren’t alone in the category — Together AI sits at roughly $8.3 billion, Fireworks AI at around $4 billion — but Modal and Baseten are the two currently being handed valuations that treat “inference plumbing” as a category worth pricing like a frontier lab. Modal was founded in 2021 by Erik Bernhardsson, formerly of Spotify and Better.com, and Akshat Bubna; it has raised roughly $465 million total and runs engineering teams out of New York, San Francisco, and Stockholm with just over 120 employees, which means a company you could seat in a single conference room is being valued not far off what plenty of Fortune 500 companies are worth on the public markets. Baseten, for its part, closed a $1.5 billion Series F at $13 billion in June, led by Altimeter Capital, Conviction Partners, and Spark Capital — and is now, per Axios, being approached about doubling that figure before the ink has had much time to dry.
Nothing here is finalized. Both rounds are, as of this writing, talks — sourced to people familiar with private conversations, not press releases, and both companies declined to comment when asked. Valuations built on “in talks” have a way of drifting before they close, and a tripling in four months is the kind of number that makes for a great headline and a slightly awkward due-diligence memo. But the direction of the bet is not subtle: investors are no longer pricing the frontier labs as the only place where AI money gets made. They’re pricing the pipes.
That’s a meaningfully different story than the one the industry has been telling about itself for the past three years, which was mostly about which lab has the smartest model. Increasingly, the more durable business might be the one nobody outside the industry has heard of — the company that doesn’t need to win the model race, just needs everyone who does win it to still need somewhere to run the thing. Modal and Baseten didn’t build GPT-6 or Gemini 4 or Opus 5.5. They just get to bill for every time someone asks one of them a question.
Sources: