Modal Tripled Its Valuation in Five Months. A Third of the Revenue Comes From Running Code the Models Wrote

Modal has raised a $750 million Series D led by Accel at a $15.75 billion post-money valuation, and says its annualized run rate has passed $1 billion. Kleiner Perkins is new to the cap table. Menlo Ventures, General Catalyst, Redpoint, Lux Capital and Bain Capital Ventures all took part again.

The valuation is not the interesting number. The gap between this round and the last one is. Modal closed a $355 million Series C at $4.65 billion on 21 May 2026, with annualized revenue it then described as "surpassing $300 million." Five months later both figures have more than tripled. Axios, which spoke to chief executive Erik Bernhardsson, places the round in the top 1 per cent of US enterprise software Series D rounds on record — its own characterisation, with no dataset attached, but the shape of the claim is right. This is a very large cheque for a company that was a Python deployment tool for machine-learning teams two years ago.

A third of the revenue is sandboxes

Modal says Sandboxes — short-lived containers it provisions on demand — reached a $300 million run rate on their own. SiliconANGLE reports that is close to a third of company revenue. Modal says it has started nearly three billion of them, has run a million concurrently in an internal test, and can provision tens of thousands per second using a scheduler that spreads the work across servers.

A sandbox is where code a language model wrote gets executed. That is what the product is for: giving an agent somewhere isolated to run the thing it just generated, with a credential-injection feature so the agent can authenticate to an application without ever seeing the secret, and an option to attach a second container alongside it running auxiliary software such as a hallucination filter.

Which makes that $300 million the most concrete public figure available for how much machine-written code is actually being executed in production rather than demonstrated in a keynote. It did not exist as a revenue line at the Series B in September 2025. It is now larger than Modal's entire business was in May.

The layer above the metal, below the app

This blog covered Character.AI's Chapter 11 filing yesterday. That company had twenty million users and never found a price that covered what their conversations cost to serve. Modal sits on the other side of the same trade. It rents hardware from more than twenty cloud providers and sells access to it with its own file system, container runtime, scheduler and networking in between — including RDMA-connected GPU clusters a developer can start from a script.

Renting rather than building cuts two ways. Modal commits no capital to data centres and can keep running when one provider has problems, but it also never owns the asset, and its margin is whatever spread the software layer can hold above wholesale GPU pricing. The company has not disclosed that margin, and the entire $15.75 billion rests on it being wide enough.

The customer list is the strongest evidence it is holding. Cognition, Ramp, Linear and Hebbia run background agents on it; Runway generates video; Physical Intelligence, Generalist AI and 1X train robots; Chai Discovery and Achira run drug discovery; Suno generates music; DoorDash's AI platform team uses it to move internal apps into production. TypeSafe AI's Jev served a trillion tokens on Modal in its first three days. Baseten, the nearest comparable company, has raised $2.1 billion. Investors are paying up for this layer because it is the only one in the AI stack that bills by usage and has no churn problem: every agent run is a metered event.

What the price assumes

Three things, none of them settled.

"Annualized run rate" is not annual recurring revenue, and Modal's own post uses the former while the Axios writeup calls it the latter. For a usage-metered business the distinction is real: a single customer's training run can move a month, and a trillion tokens in three days illustrates the volatility as much as the upside.

Growth above 15x year on year has to continue for a while before $15.75 billion looks like a sensible multiple of $1 billion in run rate.

And the sandbox line is downstream of someone else's product decisions. If the frontier labs bring code execution in-house, or agentic coding demand flattens, the fastest-growing third of this business is the first part to feel it. Modal is spending some of the new money on reinforcement-learning and post-training services, which reads like an attempt to add a revenue line that does not depend on that bet.

Sources: Modal · Modal Series C · SiliconANGLE · Dealroom (citing Axios Pro)