Fifteen times run rate, on Modal's own numbers

Modal, which sells cloud computing for AI workloads, said on Wednesday it has raised a $750m Series D at a $15.75bn post-money valuation, in a post by Erik Bernhardsson, its co-founder and chief executive. Accel led the round and Kleiner Perkins joined, the company said, and all of its existing investors took part, including Menlo Ventures, General Catalyst, Redpoint, Lux Capital, Amplify Partners and Bain Capital Ventures.

The growth figures come from Modal. "Since then, usage has tripled and we've grown more than 15x year-over-year. We recently crossed $1B in annualized run rate with no sign of slowing down," Bernhardsson wrote, referring to the company's Series C five months earlier. A run rate projects current revenue over a year; it is not revenue booked over twelve months, and Modal has not published margins or audited figures.

The round had been expected. TechCrunch reported on 28 September that Modal was in talks to raise $750m led by Accel at $15.75bn, after a $355m round at a $4.65bn valuation four months earlier. SiliconANGLE described the new valuation as more than triple the level after the May round. By our arithmetic, $15.75bn is 15.75 times the run rate Modal reports.

Modal points to one product as the engine. It says Sandboxes, isolated containers launched in 2023 that customers use to run AI agents, reinforcement-learning jobs and code that language models write, is now "a $300M run rate product just on its own", which would be 30 per cent of the total by our arithmetic. SiliconANGLE put it at nearly a third of revenue. Modal says close to 3bn sandboxes have been started and that they scale to a million running at once; SiliconANGLE reported that the million-container figure came from a recent internal test.

The company says it rebuilt most of the stack itself, including its own file system, container runtime, scheduler and networking, on a capacity pool that spans more than 20 clouds. SiliconANGLE reported that Modal rents the underlying hardware from those providers, and that workloads can keep running on the others if one cloud has problems. That model means Modal's margins depend partly on what it pays for GPU capacity, an inference from its structure rather than anything the company has disclosed.

Modal named its customers. It says DoorDash's AI platform team uses it, that Cognition, Ramp, Linear and Hebbia build agents on it, that Runway generates video on it, that Physical Intelligence, Generalist AI and 1X train robots and that Chai Discovery and Achira run drug discovery. It also says TypeSafe AI's Jev model served one trillion tokens on Modal in its first three days. These are the company's statements; the customers' own accounts were not part of the announcement.

On products, Modal says it recently launched multi-node training and is launching reinforcement-learning and post-training services, and that virtual machines are now generally available in Sandboxes. It has more than 200 staff across New York, San Francisco and Stockholm. SiliconANGLE reported that part of the money will go to product development; Modal did not give a breakdown.

The round adds to the capital going into the layer beneath the models. Komposite reported on Saturday on TypeSafe AI's $870m raise, and earlier on another infrastructure provider's backlog and on where infrastructure money is moving. Whether a multiple of 15.75 times run rate holds will depend on figures Modal has not yet published, starting with what it keeps from each dollar of compute it sells.