The agent needs somewhere to run

When Modal raised its $355 million Series C in May, the headline number was the valuation — $4.65 billion, more than four times the company's worth eight months earlier. The more revealing number was buried in the revenue mix. A single product line, code-execution sandboxes, now brings in more than a third of Modal's $300-million-plus annual revenue.

A single product line, code-execution sandboxes, went from launch to more than a third of Modal's revenue — over a billion boxes — in under a year.

A sandbox is a walled-off container an AI agent spins up to run code it just wrote — try the script, catch the error, throw the box away. It is unglamorous plumbing. But agents produce a workload shape that looks nothing like a web server humming along at steady load: thousands of short, bursty jobs that appear and vanish in seconds. Modal has now launched over a billion of these boxes. "Somewhere for the agent to run code" turns out to be a standalone infrastructure market, not a feature — and it grew into a plurality of a leading AI-cloud's revenue in under a year.

The fundraise itself is the Nth AI-infrastructure megaround in a crowded 2026 field — Baseten sits at a $13 billion valuation, Together is reportedly raising near $7.5 billion. The signal isn't the money; it's what the money is chasing. The agent coding boom has quietly moved a real fraction of infra spend from serving models to running the code those models emit, and Modal's ledger is the clearest evidence yet that the shift has a price tag.