Modal and Baseten seek funding at $15B and $26B valuations

Why it matters

Modal and Baseten are seeking sharply higher private valuations for businesses that must buy or secure compute to serve AI workloads. Whether the proposed prices hold will depend on undisclosed financing terms and on whether customer demand can support attractive economics after hardware costs.

Modal Labs and Baseten are in separate funding talks that could value them at roughly $15 billion and $26 billion, respectively, according to Bloomberg. The discussions, reported on September 23rd, have not produced completed rounds, and the amount of capital either startup hopes to raise is unclear.

For Modal co-founder and CEO Erik Bernhardsson, the potential jump would put a much higher price on an infrastructure idea he began pursuing after running engineering and machine-learning teams at Spotify and mortgage startup Better.com. Baseten CEO and co-founder Tuhin Srivastava is seeking a similar repricing for a company built around a different but related problem: getting models to work reliably once customers put them into production.

The valuations are the headline. The underlying business is a wager that companies will keep paying specialists to operate AI models after the expensive training work is finished - and that the specialists can secure enough computing capacity to serve those workloads at a profit.

A steep repricing, with the terms still open

Bloomberg reports that Modal's proposed valuation is roughly triple its level in a financing round about four months earlier. That comparison implies a prior mark near $5 billion, though Bloomberg did not give an exact figure for that round. Baseten's talks could value it at $26 billion, twice the $13 billion valuation Bloomberg says it reached in June.

Those are prospective private-market valuations, not cash raised or finalized prices. Bloomberg's report does not identify the size, structure, or lead investors for either proposed financing. The gap matters: a valuation can rise sharply while the amount of new capital, dilution to existing shareholders, and terms attached to the investment remain unknown.

The companies have previously disclosed rounds at far lower valuations. Modal said in September 2025 that it raised $87 million in a Series B led by Lux Capital at a $1.1 billion post-money valuation. Baseten announced a $300 million Series E in February 2026 at a $5 billion valuation, led by IVP and CapitalG, with Nvidia and other investors participating. Its announcement described that as its third financing in the preceding twelve months. Bloomberg's reported June valuation of $13 billion would place Baseten's latest talks amid a rapid sequence of private-market marks, rather than a single step up from the February round. Modal's reported annual recurring revenue was about $50 million in early 2026, according to TechCrunch.

The prior rounds also show what the new money would support. Modal said its Series B would fund infrastructure for AI workloads, including a platform designed to allocate GPUs and CPUs and run code without customers managing the underlying capacity. Baseten has described its financing as supporting work on inference speed, uptime, and developer experience. Both companies are selling software, but their service depends on access to costly computing hardware.

Founders who have worked on the problem from different sides

Bernhardsson's background gives Modal a founder with experience building machine-learning systems and managing the engineering organization that operates them. At Spotify, he worked on music recommendations and led analytics and machine-learning teams. At Better.com, he ran a technology group of roughly 300 people, according to his personal account. He started Modal in 2021 to make it easier for developers to run code in the cloud without assembling and managing the infrastructure themselves.

That origin shows up in Modal's product emphasis: a code-first platform for compute workloads, including inference, that can use both Nvidia GPUs and conventional CPUs. Its pitch is broader than model hosting alone. Modal is trying to make cloud infrastructure feel like a programmable extension of a developer's code, while taking capacity management off the customer's hands.

Srivastava has framed Baseten's customer problem more narrowly. In a previous Bloomberg interview, he described the service as giving companies a way to run models when they lack the infrastructure teams available at frontier AI labs. Baseten's four co-founders built the company after encountering the difficulties of deploying machine-learning systems, according to Baseten's account of its founding. The company's February financing announcement named customers including Abridge, Clay, Cursor, OpenEvidence, Mercor, and Notion; those examples are company-reported, not a measure of revenue or customer concentration.

Baseten focuses on software and computing capacity for deploying and operating models. That positioning can appeal to organizations that need production reliability and model-specific performance work without building a serving stack themselves. Modal's broader compute approach and Baseten's emphasis on managed production serving target overlapping demand, but they are not identical product strategies.

The cost behind the inference bet

Inference is the repeated computation required every time a trained model answers a prompt or performs a task. As businesses shift from experimenting with models to embedding them in products and workflows, those requests create ongoing demand for compute. Bloomberg reports that some estimates put future spending on inference chips and computing above training costs, as agents and other computationally intensive tools spread.

That demand is not automatically a durable margin pool. Modal and Baseten need access to chips and data-center capacity to deliver their services, while competing with cloud providers that already operate large infrastructure businesses. They also overlap with inference vendors such as Fireworks AI, Together AI, Replicate, and RunPod. Their case to customers is that specialized software and operations can make model deployment faster or more efficient than building the systems internally. Their case to investors is that enough customers will prefer that convenience to owning the complexity themselves.

The reported valuations put a striking price on that second proposition. Modal's and Baseten's growth in valuation talks suggests investors see inference as a large infrastructure category, but the figures alone do not show how much usage is recurring, how compute costs scale with demand, or how much of each company's economics depends on scarce hardware. The financing terms and operating metrics would determine whether these are prices for demonstrated businesses or for the capacity investors expect them to build.

For now, both rounds remain discussions. The reported figures mark investor appetite for the companies and the inference market; they do not establish that either startup has closed financing at those valuations.