The San Francisco startup could more than double its valuation just four months after raising $1.5 billion, as investors race to back the infrastructure running AI models.
Baseten is nearing a financing of approximately $2 billion that could value the company at $30 billion, according to Bloomberg, making it one of the most valuable independent AI infrastructure startups.
The proposed investment would be divided into two tranches. The first would value Baseten at $26 billion, while the second would price the company at $30 billion. Existing investor Sands Capital is expected to lead the financing.
The discussions have not resulted in a publicly announced closing, and Baseten has not confirmed the terms.
The numbers are striking even by the standards of this year's AI fundraising boom.
In June, Baseten raised $1.5 billion at a $13 billion valuation. A new round at $30 billion would put its value roughly 2.3 times higher in four months.
In January, the company was worth just $5 billion.
The enthusiasm reflects a major shift in how investors are valuing the AI industry. Training large models remains enormously expensive, but running them for millions of users is becoming an increasingly important business.
That's the market Baseten was built to serve.
Founded in 2019 by Tuhin Srivastava, Amir Haghighat, Phil Howes and Pankaj Gupta, Baseten provides the infrastructure companies need to run AI models inside production applications.
When someone uses an AI-powered coding assistant or medical documentation tool, the software needs access to computing infrastructure capable of processing requests quickly and reliably.
Baseten handles the systems behind that process: deploying models, managing GPU capacity, routing requests, controlling latency and keeping applications running as demand changes.
Its customers include Cursor, Notion, OpenEvidence, Abridge and several other AI companies.
Rather than building its own massive data centers, Baseten operates across computing infrastructure supplied by multiple cloud providers.
That approach allows customers to run specialized models without assembling their own large infrastructure teams.
It also creates a complicated commercial dependency. Baseten must secure enough computing capacity to serve its customers while keeping the cost of that capacity below what it charges.
The company reported that revenue increased twentyfold in the year leading up to its June financing, while inference volume grew fortyfold.
Those growth rates are substantial, although Baseten has not published detailed profitability figures.
The economics of AI are changing as companies begin deploying more specialized models.
Instead of sending every request to a frontier model from OpenAI, Anthropic or Google, developers can use smaller models optimized for individual tasks.
An insurance company might use one model to classify documents, another to extract financial information and a larger model for complex reasoning.
That can reduce operating costs, but introduces new infrastructure requirements.
Each model needs appropriate hardware, scaling, monitoring and performance optimization.
Baseten is betting that managing these workloads will become a major software infrastructure category.
In September, it acquired Blaxel, bringing execution environments and persistent storage for AI agents into its platform.
On October 29, Baseten released joey-1, a two-billion-parameter model designed to make fast classification and routing decisions. The company prices it at $0.01 per million tokens and says it can process a typical request with a median latency of approximately 20 milliseconds.
These moves suggest Baseten wants to own more of the infrastructure surrounding AI applications, from model execution to the environments in which agents operate.
Competition is intensifying.
Fireworks AI, Modal and other infrastructure startups are targeting parts of the same market, while cloud providers such as Amazon, Microsoft and Google offer their own model-serving services.
That creates a difficult strategic question for independent companies.
As AI models become cheaper and more widely available, demand for inference could grow substantially. But the infrastructure needed to deliver it may also become increasingly competitive.
Baseten's advantage will depend on keeping its services fast, reliable and economical while maintaining enough computing capacity for customers whose usage can grow unpredictably.
The company's proposed valuation reflects investor expectations that inference will become one of the largest recurring expenditures in AI.
For now, the $30 billion figure is still a reported financing price, not a completed transaction.
But the negotiations illustrate how quickly capital is moving toward companies that provide the computing infrastructure behind AI applications.
Sources: Bloomberg via Investing.com · Baseten Series F · joey-1 launch · Blaxel acquisition