Baseten's $1.5bn Inference Bet Lands as Open Models Surge
Baseten is closing a $1.5bn round at a $13bn valuation, a 160% jump in five months. The cheaper-inference thesis it sells has sharp lessons for India's cost-squeezed AI builders.
The News
Baseten, a startup that runs artificial-intelligence models for other companies, is close to finalising a $1.5 billion funding round at a $13 billion valuation, according to a report in the Wall Street Journal. The deal is structured unusually, with some backers buying in at the $13 billion figure and others at a lower $11 billion mark.
The round is led by Spark Capital, Sands Capital, Altimeter Capital and Wellington Management. What makes the figure striking is its speed. Just five months ago Baseten closed a $300 million Series E at a $5 billion valuation, which itself arrived only nine months after a $150 million Series D. The new figure marks a 160% leap from that $5 billion valuation.
Founded in 2019, Baseten sells what the industry calls inference: the computing work that happens every time a user sends a prompt and waits for an answer. Its pitch is cost discipline. The company routes each request to whichever model handles the task best, leaning on capable but cheaper open-source options rather than defaulting to the priciest frontier systems.
Why It Matters
The raise is the clearest signal yet of what investors are calling the inference gold rush. For three years the money chased model builders. Now it flows to the plumbing underneath, the layer deciding how cheaply and quickly a trained model can serve millions of live queries. A 160% valuation jump in five months, with no public revenue figures attached, shows how aggressively capital is repricing that layer.
There is a familiar rhythm here. When cloud computing matured in the early 2010s, the richest returns drifted from the headline platforms to the optimisation tools that squeezed cost out of every workload. Inference is following the same arc. The expensive part of AI is no longer training a model once; it is answering questions with it forever, and Baseten is betting the winners will be the firms that make that forever-cost bearable.
The split-priced structure is its own quiet message. When some backers accept a $13 billion entry and others hold out for $11 billion, the market is still arguing about what an inference business is worth, a sign that valuation discipline has not entirely evaporated.
Indian Angle
For Indian AI builders, the cheaper-inference thesis is not a curiosity. It is the whole game. Most Indian startups earn in rupees and pay for GPU compute in dollars, a currency mismatch that turns inference bills into an existential line item. A platform that routes traffic to competent open-source models instead of premium frontier ones is selling exactly the cost relief that a Bengaluru or Pune team needs to keep gross margins alive.
This is also why India's own model effort matters. Sarvam, building open-weight Indic models, and Krutrim, which has launched its own AI cloud, both bet that locally tuned, cheaper-to-run models can undercut the imported frontier giants. Baseten's success validates the demand side of that bet: enterprises want a layer that picks the cheapest model that still does the job, and Indian inference and cloud players have a natural home market in that logic.
The policy backdrop fits too. MeitY's IndiaAI Mission is subsidising GPU access precisely because compute cost is the bottleneck for domestic builders. If routing and optimisation is where the margins sit, Indian engineers, who already staff much of the world's cloud-cost tooling, are well placed to build the equivalent for AI.
FAQ
What exactly does Baseten do?
It runs AI models for other companies, specialising in inference: the live computing that produces an answer each time someone queries a model. Its edge is routing each request to the most suitable and often cheapest model, frequently an open-source one.
How big is the valuation jump?
The reported $13 billion valuation is a 160% increase over the $5 billion valuation Baseten set at its $300 million Series E five months earlier. The new round totals $1.5 billion, with some investors entering at $13 billion and others at $11 billion.
Why should Indian startups care?
Indian firms pay for GPU compute in dollars while earning in rupees, so inference cost directly threatens their margins. A platform built around cheaper open-source models fits India's cost-sensitive market and the case for domestic efforts such as Sarvam and Krutrim.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.