Modal raises at $15.75B valuation with Cognition, Ramp, Suno, Runway, and DoorDash as customers

Modal raises $750M at $15.75B valuation

Modal Labs has raised $750M at a $15.75B valuation, with Erik Bernhardsson pointing to surging demand from AI-native companies as the primary driver. Customers include Cognition, Ramp, Suno, Runway, Decagon, and DoorDash — a mix that spans pure-play AI startups and more established tech companies.

What Modal actually does

Modal aggregates compute from 20 to 25 clouds, abstracts away the differences between them, and lets customers deploy inference, training, sandboxes, and reinforcement learning workloads with a few lines of Python. Bernhardsson is blunt that the multi-cloud reality is painful, with each provider meaningfully different from the next. Modal's internal "Modal Max" rating system tracks the variation. The value proposition is that customers never have to touch that complexity directly.

Most of the $750M will go toward expanding capacity. Compute availability, not capital, has been the binding constraint.

“We raised a bunch of money... $750M, $15,750,000,000 valuation... We see so much demand from all these incredible AI applications out there — Cognition, Ramp, Suno, Runway, Decagon, DoorDash. We aggregate capacity from 20, maybe 25 clouds today and add a pretty rich software layer on top.”

Geographic and temporal arbitrage

Bernhardsson sees material optimization opportunity as Modal's share of global compute grows. Running batch jobs overnight, shifting workloads to Asian data centers during off-peak hours in the US, and arbitraging weekend capacity are all on the table. Because Modal presents all of its aggregated supply as a single pool, customers don't need to manage any of that routing themselves.

Where adoption actually stands

Enterprise diffusion is still early. Bernhardsson describes a wide gap between AI-native companies building from scratch and the rest of the market — companies with capable teams that simply aren't following the frontier closely. DoorDash is the example he cites as a slightly more mature company beginning to adopt Modal's infrastructure. Deutsche Bank, he says, is probably years away. The gap is "massive," and he repeats the word three times.

General purpose as competitive strategy

On a market Bernhardsson says turned out to be 10x, 100x, or 1,000x larger than he originally expected, he argues Modal's generalist positioning is the right bet. Vanilla inference on a single model is going to be brutally competitive. Building the fuller stack around it — sandboxes, agentic workflows, post-training, reinforcement learning — is where he thinks a generational company gets built. He's also looking at alternative chips, including Cerebras, for ultra-low-latency use cases like real-time voice and audio, though he describes that as early-stage exploration.

The central bet is that as the market fragments into specialized workloads, Modal is positioned to serve across the full latency spectrum rather than getting squeezed into one corner of it.