Accel in Talks to Triple Modal to $15 Billion and Mark Up Its Own May Investment

The mark tripled in four months

Modal Labs, the serverless compute platform founded in 2021 that runs inference on Nvidia GPUs and ordinary chips, is in talks to raise at roughly $15 billion, about triple its worth in a round four months ago, according to people familiar with the matter↗ ¹². The reported lead is Accel↗ ³, which joined Modal's $355 million Series C in May ⁴. Modal declined to comment ¹.

The multiple is the story

May's $4.65 billion post-money valuation ⁴ sat on more than $300 million in annualized revenue, a company-disclosed run-rate figure↗ ⁴: about 15 times. A $15 billion price on that same revenue is 50 times. The multiple only holds if the run rate has roughly tripled since May, to around $1 billion; it went from roughly $60 million to $300 million between September and May ⁴. The new mark assumes the pace never paused.

Accel repricing Accel

Leading at $15 billion would mark up Accel's own May position threefold, and every earlier holder with it ³⁴. The new money pays for that repricing, buying at roughly 50 times the last disclosed revenue; the old money, Accel included, rides a threefold paper gain without paying in again. Headline marks can also overstate the price actually paid: Baseten's June round closed at $13 billion only after talks involving different prices for different investors ⁴, and its current discussions, at $26 billion, would double that mark in a quarter ¹.

Sandboxes carry the bet

More than a third of Modal's revenue comes from sandboxes, isolated environments where agents run and test untrusted code; over one billion have launched ⁴, and Modal's engineers spent July writing up how to scale to a million of them running at once↗ ⁵. The wager is that agent workloads keep renting someone else's GPU plumbing rather than building on hyperscale clouds, the do-it-yourself route that is the specialists' real opponent ⁴. Modal rents its silicon rather than owning it, spread across thirteen cloud providers as compute got scarcer ⁴; whether an orchestration layer keeps durable margins at 50 times revenue, with the hyperscaler route as the alternative, is the question the mark assumes away.

Only a revenue print near $1 billion justifies it

Two numbers decide whether this is a repricing or a reach: whether the round signs near $15 billion on clean terms↗, and whether Modal's next disclosure puts annualized revenue near $1 billion. Inference compute spending is, by some estimates, set to pass spending on training ¹. Until the revenue prints, 50 times is a belief, not a booking.