Modal closed a $355 million Series C round, marking a dramatic scale-up from its $17 million Series A just two years ago.

The San Francisco-based serverless compute platform is positioning itself as the "agent cloud" — infrastructure designed specifically for AI workloads rather than traditional web applications. CTO Akshat Bubna argues that agents need fundamentally different primitives than human developers.

"The old infra stack was designed for a human who could read docs, reason through YAML, and understand dashboards," Bubna said in a recent podcast. "Agents don't have that luxury. Everything has to be tighter."

Why agents need different infrastructure

Modal has built specialized features for AI workloads that traditional cloud providers struggle with. The platform offers elastic inference that can scale custom models across audio, video, robotics, and computational biology applications.

The company's sandbox technology can spin up 100,000 isolated environments for reinforcement learning rollouts — something Bubna says would be impossible on Kubernetes. Modal also developed GPU snapshotting to reduce cold start times and DeFlash for speculative decoding.

Modal operates across 17 cloud providers in what it calls a "supercloud" strategy, pooling capacity to handle the bursty nature of AI workloads. The platform supports multi-node training for post-training and research workloads, with features like networked containers and private IPv6.

Bubna highlighted how agents themselves can now operate Modal's infrastructure programmatically, launching GPU experiments and managing compute resources without human intervention.

The funding comes as Modal competes with established players like AWS and Google Cloud, as well as AI-native infrastructure companies. Modal's decorator-based approach to serverless functions has attracted customers running everything from background agents to production inference workloads.

Modal plans to expand its agent-focused features, including specialized sandboxes with hard guardrails for production deployments and enhanced observability tools for AI applications.