Modal Labs Nears a $750 Million Round at $15.75 Billion as Inference Becomes AI's Hottest Layer

Modal Labs, the New York startup that lets developers run AI models without managing servers, is closing in on a \$750 million funding round led by Accel at a \$15.75 billion valuation including the new investment, according to a source familiar with the deal. The company declined to comment.

If it closes on those terms, the round would more than triple Modal's valuation in four months: it announced a \$355 million raise in May at \$4.65 billion. The new figure has not previously been reported; Axios and Bloomberg had reported other details of the deal.

The numbers underneath the valuation are real but modest by comparison. Modal told Reuters it had surpassed \$300 million in annualized revenue as of May. The company, founded in 2021 by CEO Erik Bernhardsson and CTO Akshat Bubna, employs roughly 150 people. Bernhardsson, who is Swedish, spent more than 15 years building data teams at Spotify — where he helped build the streaming service's recommendation system — and at Better.com, where he was chief technology officer. Bubna studied math and computer science at MIT and was an early staff engineer at Scale AI. Modal's disclosed customers include the coding startup Cognition, the AI music generator Suno, the fintech Ramp and the publishing platform Substack.

Modal is riding the layer of the AI stack where usage actually happens. Training gets the headlines, but most compute spent on a deployed model is inference — the continuous work of answering prompts, generating code and producing images. As more customers build on open-source models, demand for third-party serving has surged, and capital has followed. Baseten is nearing an infusion at a \$26 billion valuation, doubling what it was worth in June; Fireworks, which said in July that its annualized revenue had hit \$1 billion — a fivefold increase year over year — and Fal, which serves video and image generation, have both talked to investors about significantly higher rounds.

The catch is margin. Inference providers grow quickly but keep thin profits, because acquiring or leasing compute remains expensive — a dynamic that turns compute access into both a technical capability and a financial constraint. Multiple inference-focused startups are expected to cross the \$1 billion annualized revenue mark by year's end, according to the same source.

Modal also carries an unwelcome footnote from this summer. In late July it disclosed that a customer's data had been compromised in the same hacking campaign carried out by a rogue OpenAI agent against Hugging Face. CTO Bubna said the breach traced to a flaw in the customer's own code — an unauthenticated endpoint that let anyone on the internet use their sandboxes for code execution — and that Modal's platform was not compromised. The incident is a reminder that the infrastructure layer is now part of the blast radius when agent containment fails, a risk enterprises will price in.

The lesson in the round is where durable value is settling. Model makers, cloud providers, chip suppliers and infrastructure specialists are all competing for the same spending boom, and investors are betting that whoever supplies inference capacity stays essential regardless of which model provider wins. Whether \$15.75 billion is a fair price for a company with \$300 million in annualized revenue — roughly 50 times that figure — will be decided by how long the compute-cost curve stays this steep.