Modal CEO Erik Bernhardsson on elastic GPUs, sandbox growth, and agent-cloud economics
Modal’s CEO discusses how serverless GPU infrastructure powers agent sandboxes, the shift from fixed GPU contracts to elastic demand, and the company’s $750M Series D at a $16B valuation.
1 source · cross-referenced
- Modal reports $1B in annualized run rate and nearly 3B sandboxes since its $5B Series C.
- Sandboxes now account for roughly a third of Modal’s revenue, driven by coding agents and background agents.
- The company raised a $750M Series D at a $16B valuation, citing demand for elastic GPU capacity.
- Erik Bernhardsson argues fixed GPU contracts are mismatched with real workloads that spike and fall throughout the day.
Modal, an AI infrastructure platform, reports crossing $1 billion in annualized run rate and spinning up nearly three billion sandboxes since its $5 billion Series C, according to an interview with CEO Erik Bernhardsson published by Latent Space.
Bernhardsson describes Modal’s approach as a form of duration transformation: pooling demand across thousands of customers to secure capacity and deliver elastic GPU usage in smaller slices. This contrasts with typical capacity contracts that provide fixed GPUs for fixed periods, which he argues do not match real workloads that grow, spike, and fall throughout the day.
Sandboxes, initially launched with modest adoption, have grown to roughly a third of Modal’s revenue, Bernhardsson says. The shift was driven by coding agents, background agents, and reinforcement learning workflows that repeatedly create, run, and inspect environments, making low-latency startup and restoration critical to user experience.
The company raised a $750 million Series D at a $16 billion valuation, citing demand for elastic GPU capacity and the operational scale required to support agent sandboxes and inference workloads.
Bernhardsson also discusses the financial risk of locking in capacity at today’s prices if GPU prices later fall, though he does not expect an imminent collapse. Modal’s infrastructure includes a buffer of available GPUs, a custom filesystem, process checkpointing, and GPU memory snapshots to reduce latency when bringing model replicas online.
- Nov 11, 2026 · Latent Space — swyx
OpenAI demonstrates voice steering, cloud handoffs, and personal agents for Codex at AI Engineer Code Summit
Trust79 - Nov 11, 2026 · The Verge — AI
OpenAI to launch verified AI Business Agents in ChatGPT for US e-commerce and service businesses
Trust79 - Nov 10, 2026 · Latent Space — swyx
NaiveAI releases Naive-N0.5-Flash coding model with ultrafast token throughput
Trust74