SAN FRANCISCO: Baseten, a San Francisco-based AI inference startup, is closing a roughly $1.5 billion funding round at a valuation of up to $13 billion, with some investors entering at an $11 billion mark and others at $13 billion. The round is co-led by Altimeter Capital, Conviction, Spark Capital, Sands Capital, and Wellington Management. The Wall Street Journal first reported the raise on June 18, 2026, followed by coverage from TechCrunch and PYMNTS. Baseten has not posted an official press release confirming the close as of June 19, 2026.
The pace is the headline. Baseten closed a $300 million Series E in January 2026 at a $5 billion valuation, led by IVP and CapitalG, with a $150 million check from Nvidia, per the company’s January announcement. Hitting $13 billion roughly five months later puts the company at about 2.6 times its prior valuation, after a Series E that itself doubled the company’s mark in a single round. Baseten’s annualized revenue run-rate climbed from about $200 million in late 2025 to roughly $600 million by the close of the first quarter of 2026, according to multiple reports on the deal.
Why inference became its own infrastructure war
Training a model and serving a model are two different problems. Training is a one-time capital expense measured in weeks of GPU time. Inference, the step of actually running a trained model to produce an output, runs every time a user clicks a button. As AI applications scale, inference becomes the recurring bill that determines whether a product has unit economics or just a demo.
Baseten sits in that layer. The company builds software and multi-cloud compute infrastructure that helps companies deploy and optimize open-source models in production, with a focus on latency, throughput, and cost per query. Customers include AI coding tool Cursor, talent marketplace Mercor, and clinical-evidence platform OpenEvidence, with users reporting cost savings of up to 30 percent compared with closed-source model APIs, according to MLQ News. That cost arbitrage between open-source models on optimized infrastructure and closed-model API calls is what is funding the run-rate jump and what investors are paying $13 billion for.
The split-priced structure of the round, with two valuation tiers in the same deal, signals the kind of demand that strains conventional pricing. Later entrants paid more for the right to get in. That pattern showed up earlier this year in other AI mega-rounds and is now spreading to the inference layer specifically.
What does Baseten’s $13 billion valuation mean for AI startups?
The Baseten round is a price tag on a specific bet: that inference infrastructure for open-source models is a large, defensible category, not a feature that will get absorbed by hyperscaler clouds or model labs. A 2.6x valuation jump in five months, paired with a 3x ARR jump in one quarter, tells founders that the spend on serving AI is now growing faster than the spend on training it, and that the picks-and-shovels layer for inference is being treated like a tier-one infrastructure category.
For founders building AI products, the practical implication is unit economics. If open-source models on optimized infrastructure can deliver 30 percent savings against the closed-model APIs that most applications run on today, the build-versus-buy math at the inference layer shifts. Baseten and competitors including TensorWave, Together AI, and Fireworks AI are pitching themselves as the way to capture that delta without standing up GPU clusters in-house.
For the venture side, the round also continues a 2026 pattern of compressed time between mega-rounds. Companies that hit product-market fit in AI infrastructure are raising at multi-billion-dollar increments every few months, not every 18 to 24 months. Capital is moving faster than corporate development calendars can keep up with, which is part of why split-tier valuations have started appearing.
What’s next for Baseten
Baseten has said it will use the capital to expand GPU clusters, deepen its open-source model optimization stack, and hire across engineering and go-to-market, per the reports on the round. An official close announcement from the company has not yet been posted, and the final lead investor lineup and exact valuation could shift before the deal is formally papered.
The other near-term watch item is the rest of the inference stack. Several competitors are reportedly in the market for new funding at similar valuation tiers, and at least one closed-source model provider is exploring whether to open its inference layer to third-party optimization. If both happen, the category Baseten just got valued at $13 billion will look very different by the end of the year.
Watch three things over the next 90 days. First, whether Baseten posts a formal close with a final lead investor and locks in the $13 billion top tier or settles closer to $11 billion. Second, whether enterprise customers beyond Cursor, Mercor, and OpenEvidence start naming Baseten as their inference layer in earnings calls and SEC filings, which is where the run-rate either holds or breaks. Third, the response from hyperscaler clouds, which have their own inference offerings and the most to lose if a third-party layer keeps taking share. Either way, expect more rounds, more tier-split pricing, and a sharper line between the companies that own the inference layer and the ones that rent it.
