Baseten, the AI inference company behind the model serving for a long list of fast-growing apps, is reportedly about to more than double its valuation. On October 30, Bloomberg reported, citing people familiar with the matter, that Baseten is nearing a funding round of about $2 billion that could value the company at up to $30 billion. The round hasn't been announced and the terms could still change, but the report says a lot about where money in AI is flowing right now and what that means for founders building on top of models.
What Bloomberg reported
According to the Bloomberg report as summarized by Investing.com, the round would come in 2 tranches, the first at a $26 billion valuation and the second at $30 billion. Existing investor Sands Capital is expected to lead, and Sands declined to comment. Baseten did not respond to requests for comment, and ShiftDelete noted that the terms are not final and the company has not announced the round, so this is still a report about talks rather than a closed deal.
For context, Bloomberg says Baseten was last valued at $13 billion in June after raising $1.5 billion. Investing.com lists Altimeter Capital, IVP and Spark Capital among its existing investors. If the new round closes at the top of the range, the company's valuation would have more than doubled in roughly 5 months, which is the kind of pace that used to be reserved for frontier model labs.
What Baseten actually does
Baseten sells inference, which is the work of running AI models after they've been trained so that apps can get answers back in real time. On its about page, the company says it started in 2019 and describes itself as "the fastest, most reliable inference platform" for serving and scaling open-source and custom AI models. Its products include dedicated inference, model APIs, training and a program it calls Model Labs, and its careers page names customers such as Abridge, Cursor, Lovable, Notion and OpenEvidence.
The co-founders listed on the company's site are CEO Tuhin Srivastava, CTO Amir Haghighat, Chief Scientist Phil Howes and Pankaj Gupta. Press coverage describes Baseten as a San Francisco company. Its homepage tagline sums up the bet investors are making: "Inference will be the largest market ever created."
Why inference is where the money is going
For the last few years, most AI headlines were about training ever-larger models. The Baseten report is part of a shift in attention toward the cost of serving those models to millions of users. Every time someone asks a coding assistant to refactor a file or a healthcare app to draft a clinical note, that request runs on inference hardware, and the bill grows with usage. Peer company Fireworks AI raised $1.5 billion in July at a $17.5 billion valuation, according to Investing.com, which shows Baseten is not the only inference platform investors are piling into.
If you've followed the rise of open-weight models, this trend makes sense. The more teams that choose to run open or fine-tuned models instead of calling a single closed API, the more they need someone to host, scale and optimize those models. That's the gap Baseten, Fireworks and similar companies are filling, and it's one of the biggest AI trends of 2026 that doesn't get as much attention as new chatbot launches.
What this means for founders and AI engineers
For early-stage founders, the first takeaway is about cost planning. Inference is often the largest variable cost for an AI product, and choosing between a closed API, a managed inference platform and your own GPUs is one of the most important technical decisions you'll make in year 1. Big rounds for inference providers usually mean more capacity, more price competition and better tooling, which is good news for buyers, but it also means you should benchmark providers rather than defaulting to whatever your first prototype used.
The second takeaway is about careers. Engineers who understand model serving, GPU scheduling, quantization and latency optimization are some of the most sought-after people in the Bay Area right now, and the Baseten report suggests that demand won't cool soon. If you're weighing an offer from a fast-growing infra company, our explainer on startup tender offers for employees covers how secondary sales work at companies whose valuations move this quickly.
The third takeaway is about building agents. As more teams ship products like the ones in our AI agents guide, each user action can trigger many model calls behind the scenes. That multiplies inference demand, and it's a big reason why investors see serving infrastructure as a long-term business rather than a commodity.
The builders behind the AI stack
A lot of the people doing this work live and build on the Peninsula and in the South Bay, and they don't get many chances to compare notes in person. That's why our AI Infra Dinner: Builders Behind the Next AI Stack in Palo Alto on September 30, hosted by Prateek Joshi, focused on exactly these questions about serving, cost and capacity. If you want more of those rooms, South Bay Renaissance, the community Tribe organizes to rebuild the South Bay and Peninsula startup scene in person, runs AI deep dives and founders' dinners from Redwood City to San Jose.
For the online version of that conversation, the AI Trends Tribe is where our members swap notes on model releases, inference pricing and the tools that hold up in production. And to catch the next infra dinner or demo night, follow the Tribe Events calendar on Luma so new events land in your inbox as soon as they're posted.
