Baseten company profile: valuation, funding & revenue

This Baseten company profile covers the company's valuation, revenue, funding rounds, investors and history. Baseten works in AI infrastructure, with a focus on inference platform for AI models. It was founded in 2019 and is headquartered in San Francisco, California, United States. Its latest valuation on record is US$13 billion (June 2026). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Baseten timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2019 — Baseten is founded in San Francisco by Tuhin Srivastava (CEO), Amir Haghighat, Philip Howes and Pankaj Gupta.
• Jan 2026 — Equity round: US$300M, valuing the company at US$5B.
• Jun 2026 — Series F: US$1.5B, valuing the company at US$13B. Investors: Sands Capital and Wellington Management (leads); Blackbird.
• Q1 2026 — Reported revenue run rate: Annualised revenue reported to have risen from about US$200 million to about US$600 million during the first quarter of 2026; not confirmed by the company.

Investors

Overview

Baseten provides the infrastructure that companies use to run AI models in production. Its platform takes a model, whether an open model like Llama or DeepSeek, a company's own fine-tuned version or a custom model built in-house, and serves it to users with low latency, high reliability and automatic scaling across many GPUs and cloud providers. Baseten's customers include fast-growing AI application companies that need their products to respond instantly to millions of users, and the company has grown into one of the most valuable AI infrastructure startups as demand for inference has surged.

Tuhin Srivastava, the CEO, Amir Haghighat, Philip Howes and Pankaj Gupta founded Baseten in San Francisco in 2019. Before Baseten, members of the founding team worked on machine learning and data at Gumroad and other technology companies. Baseten started as a tool to help data scientists turn models into web applications, then shifted its focus to production model serving as large language models took off, a pivot that put it at the centre of the AI boom.

Products

• Dedicated deployments: customers deploy any model on Baseten-managed GPUs, with autoscaling that adds or removes capacity as traffic changes.
• Model APIs: ready-to-use endpoints for popular open models, priced per token.
• Truss: an open-source tool for packaging models so they can be deployed consistently.
• Multi-cloud capacity management: Baseten spreads workloads across GPUs from several cloud providers and regions, which improves availability and lets customers run in their own cloud accounts if required.
• Training: tools for fine-tuning and training models that can then be served on the same platform.

Funding

Baseten's early backers include Greylock, Spark Capital, IVP, Conviction and CapitalG. In January 2026 it raised US$300 million at a US$5 billion valuation, Bloomberg reported. Just five months later, in June 2026, it announced a US$1.5 billion Series F at a US$13 billion valuation, led by Sands Capital and Wellington Management, according to a Reuters report summarised by Dealroom. Australian venture firm Blackbird said its investment was the largest in its history. Reports said some investors entered at an US$11 billion valuation and others at US$13 billion; we use the headline US$13 billion figure.

Revenue growth

Press reports said Baseten's annualised revenue climbed from roughly US$200 million to about US$600 million during the first quarter of 2026, driven by customers whose AI products were growing rapidly. Baseten hasn't confirmed those figures publicly, so we list them as reported and show revenue as not disclosed in the fact tiles. If accurate, that growth rate explains why its valuation more than doubled within months.

Competition

Baseten competes with other inference platforms including Fireworks AI, Together AI, Modal and Replicate; with GPU cloud providers such as CoreWeave, Lambda and Crusoe; and with the big cloud platforms' machine learning services. Its differentiation is its focus on running custom and fine-tuned models for demanding production workloads, its multi-cloud approach to capacity, and its engineering support for customers at scale.

Baseten at a glance

Baseten is a San Francisco AI inference company founded in 2019 by Tuhin Srivastava, Amir Haghighat, Philip Howes and Pankaj Gupta. It raised US$300 million at US$5 billion in January 2026 and a US$1.5 billion Series F at US$13 billion in June 2026, led by Sands Capital and Wellington Management. Reports said its annualised revenue roughly tripled in early 2026.

Why multi-cloud matters for AI

GPU capacity is scarce and unevenly distributed. A company relying on a single cloud region can run out of capacity during traffic spikes or outages. Baseten's approach is to pool GPUs across multiple providers and regions, routing workloads wherever capacity is available. This improves uptime and can reduce costs, because Baseten can buy capacity where it's cheapest. It also lets customers with strict data requirements run models inside their own cloud accounts while Baseten manages the software. As AI products become critical to businesses, reliability matters as much as raw speed.

The economics of inference companies

Inference platforms buy or rent GPU capacity and resell it as a managed service with software that makes models run faster and more reliably. Their gross margins depend on how efficiently they use that capacity: idle GPUs cost money, while well-utilised GPUs running optimised software can generate healthy profits. Rapid revenue growth often comes from a small number of very fast-growing customers, which can make revenue volatile if one of those customers slows down, switches providers or builds its own infrastructure. Investors in the sector watch customer concentration, utilisation and gross margin closely, alongside headline growth.

Custom models and the AI application layer

Many of the fastest-growing AI applications don't rely only on general models from big labs. They fine-tune open models for their specific tasks, distil large models into smaller, faster ones, or train specialised models for areas like speech, transcription or code. Each of these needs to be served somewhere, and the companies behind them often want more control than a simple API allows. Baseten built its business around this need, which is why it has done particularly well with application companies that have grown from startups into large businesses within a couple of years.

Questions for engineering teams

• What latency and uptime does your product need, and what are the provider's service levels?
• Do you need to run models in your own cloud account for compliance?
• How does pricing scale as your traffic grows?
• How portable is your deployment if you decide to move?

Risks

• Customer concentration: rapid growth may depend heavily on a few large customers.
• Competition: inference is a crowded market with well-funded rivals and hyperscalers.
• Valuation: a valuation that rose more than twofold in months leaves little room for slower growth.

Latest developments (October 2026)

After its June 2026 Series F of US$1.5 billion led by Altimeter Capital, Conviction and Spark Capital, which valued it at US$13 billion, Bloomberg reported in September 2026 that Baseten and peer Modal were in new funding talks; one tracker cited a valuation around US$26 billion for Baseten. No new round had been announced at the time of our update, so US$13 billion remains the latest confirmed figure. See our AI cloud providers guide.

What to watch

Watch for confirmed revenue figures, new large customers, expansion of training products, gross margin trends and whether Baseten begins preparing for an IPO.

Takeaways

AI product teams can serve custom models reliably without building infrastructure teams. Founders can learn from Baseten's well-timed pivot towards production inference. Investors should look past headline run-rates to customer concentration and margins.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Dealroom (citing Reuters)
2. Bloomberg