Modal, a New York company that rents out the computing needed to run AI models once they are built, has closed a $750 million Series D round. Accel led the financing, which values the startup at $15.75 billion after the money is counted. The announcement came on 14 October 2026, and it brings the total Modal has raised to about $1.2 billion.
The company says its annualized run rate, meaning current revenue stretched over a year, has gone past $1 billion, and that usage has roughly tripled since its Series C only five months earlier. That $1 billion figure is Modal's own claim rather than an audited number. Kleiner Perkins joined as a new investor, while Menlo Ventures, General Catalyst, Redpoint, Lux Capital and Bain Capital Ventures also took part.
The jump in value is steep. Spanish-language outlet Ecosistema Startup, writing on 29 September when the round was still being finalised, recalled that Modal had been valued at $4.65 billion after a $355 million raise about four months before talks on this one began. That piece relied on TechCrunch, with Bloomberg and Axios repeating it, all citing people familiar with the discussions; the 14 October announcement confirms them. The $355 million size of the earlier round appears only in the Spanish outlet's coverage, so treat it with some care. The same outlet says Modal told Reuters around May that its annualized revenue had passed $300 million.
Modal was founded in 2021 by Erik Bernhardsson, a Swede who spent over 15 years building data teams at Spotify and Better.com, and Akshat Bubna, a former Scale AI engineer who studied at MIT. The company has about 150 staff. Its platform lets developers train models and run compute-heavy jobs without managing servers, and its website lists Cognition, Suno, Ramp and Substack among its customers. It owns no data centres, running a software layer over partners' capacity instead. That keeps debt lower, but it also means Modal has no guaranteed claim on the newest chips when rivals lock up supply.
The business is inference, the stage where a trained model answers real requests, rather than building the models. The Spanish report says money in AI infrastructure is moving from training in huge Nvidia clusters to serving models at scale, with demand lifted by AI agents, real-time voice and video generation, and by companies running open-source models.
Competitors are raising at a similar pace. Baseten is reportedly close to a $26 billion valuation, double its June level according to Bloomberg, Fireworks said its annual revenue reached $1 billion in July, and Fal is said to be in talks. Margins across the sector are narrow because compute is expensive to buy or rent, even as revenue grows quickly, and the report expects other startups in the field to pass $1 billion a year by the end of 2026. The report also mentions that Modal disclosed in July that a customer's data had been exposed in a wider hacking campaign; its CTO said the cause was an unauthenticated endpoint the customer had published, not a flaw in Modal's own systems.
Sources · 2