Baseten Raises $1.5B Series F at a $13B Valuation as Inference Becomes AI's Most Contested Layer

Baseten Raises $1.5B Series F at a $13B Valuation as Inference Becomes AI's Most Contested Layer

Six years ago, betting a company on AI inference sounded like a bet on plumbing. On Monday, that plumbing was valued at $13 billion. Baseten announced a $1.5 billion Series F, more than doubling a valuation that had already tripled since January, and in the process handed the market its clearest signal yet that the fight for AI's most lucrative territory has moved decisively downstream — from training frontier models to serving them.

The round was led by Altimeter Capital, Conviction, and Spark Capital, and co-led by Sands Capital and Wellington Management, with participation from IVP, Greylock, 01A, Blackbird, Durable Capital Partners, Verified Capital, Battery Ventures, and D.E. Shaw Ventures. According to Baseten's disclosure, the capital came in across two tranches priced at $13 billion and $11 billion — an unusual structure that reflects how quickly the price moved during the raise itself.

The numbers underneath are the story. This is Baseten's fourth financing in 18 months. It arrives just five months after a $300 million Series E valued the company at $5 billion in January 2026, which itself came only three months after a Series D. Revenue has grown roughly 20x year-over-year, reaching an estimated $600 million annualized run-rate by March 2026 — up from around $200 million in December 2025, a jump of roughly 1,900%. Inference volume has grown 40x. The platform now processes more than one billion inference calls every day, spread across 87 clusters and 18 cloud providers.

That multi-cloud sprawl is not incidental. Baseten describes its multi-cloud architecture as "itself a core part of what customers are buying" — capacity management has become a product, not just an operational detail, as GPU scarcity forces AI companies to source compute wherever they can find it.

The multi-model thesis

Baseten's pitch is that the AI market is fragmenting into "millions of specialized models" rather than consolidating around a handful of closed frontier APIs. The company points to a shift in how the fastest-growing AI startups spend: leading app-layer companies now direct 30 to 50 percent of model spend toward custom and post-trained models, blending frontier systems with cheaper open-weight models tuned to specific workflows.

"The future of AI will be built on millions of specialized models, and the companies building the best ones know that post-training has become existential," said Tuhin Srivastava, Baseten's CEO and co-founder. "It's how they build intelligence they own, on data that's theirs, optimized for the customers they serve. The companies closing this loop fastest are using Baseten to build a system that compounds in value over time."

The customer roster is meant to prove the point. Baseten works with Cursor, Notion, Lovable, Harvey, HubSpot, OpenEvidence, Abridge, Decagon, Parallel, Clay, and Mercor — the cohort of AI-native companies where, as the company puts it, intelligence is core to the user experience. Legal AI firm Harvey and healthcare platforms Abridge and OpenEvidence are exactly the kind of accounts where latency, reliability, and data ownership outweigh the convenience of a single closed API.

Investors are underwriting the trajectory as much as the current business. "The inference market is compounding faster than almost anything we've seen, and some of the most sophisticated AI companies in the world are standardizing on Baseten to power it," said Apoorv Agrawal, a partner at Altimeter Capital. He called the company a rare case: "a company growing at this speed, with this level of customer trust, building infrastructure the entire ecosystem depends on."

The capital is earmarked for talent, compute, and enterprise go-to-market. Baseten says it is tripling headcount this year and has now raised more than $2 billion in total from a backer list that includes NVIDIA, CapitalG, and BOND alongside this round's leads.

Why It Matters

For most of the generative-AI boom, the money and the mythology lived in training — the multibillion-dollar clusters, the frontier labs, the race to the next model. Inference was treated as a cost center. That framing is inverting. Every query a user sends is inference, and as AI moves from demos into products with real usage, the recurring economics of serving models begin to dwarf the one-time cost of training them.

That has turned inference into the industry's most contested layer, and the field is crowded. Baseten competes with Together AI, Fireworks, Modal, and CoreWeave, with the hyperscalers' own inference services, and increasingly with NVIDIA's software stack. Baseten's counter-positioning is the multi-cloud, multi-model bet: neutrality on where compute runs and which models a customer uses, plus embedded engineers who post-train and optimize models on a client's behalf. If open-weight models keep closing the gap with closed APIs — the premise the whole thesis rests on — that neutral serving layer becomes more valuable, not less. If frontier labs re-establish a durable quality lead, the calculus shifts back toward their APIs.

The valuation math is aggressive by any conventional measure — roughly 20x a $600 million run-rate for a business whose margins depend on volatile GPU supply and pricing. Investors are betting the run-rate is a rounding error against where inference demand is heading.

What to Watch

Three questions will decide whether the $13 billion mark looks prescient or frothy. First, can Baseten defend gross margins as it scales across 18 clouds while GPU pricing swings? Second, does the 30-to-50-percent shift toward custom and post-trained models broaden beyond AI-native startups into the slower-moving enterprise, where Baseten is now pouring go-to-market dollars? Third, how do the hyperscalers and NVIDIA respond to a neutral serving layer capturing spend that could otherwise flow to their own inference products. With four raises in 18 months already behind it, the more immediate tell may simply be how long this round lasts before Baseten is back in the market again.