Modal CEO Erik Bernhardsson on AI Compute Economics and Infrastructure
Modal CEO Erik Bernhardsson appeared in an in-depth interview discussing elastic GPUs, sandbox growth, and the economic challenges of powering cloud-based AI agents. Understanding the economics of cloud infrastructure and GPU allocation is critical as developers build and scale increasingly autonomous AI agents that demand dynamic compute resources. The discussion explores technical nuances of serverless cloud environments, specifically examining why AI compute is structured around rectangular architectures and how execution sandboxes scale.
## BACKGROUND
Cloud infrastructure refers to the hardware and software components—such as servers, storage, networking, and GPUs—required to support cloud computing. Serverless computing allows developers to build and run applications without managing physical servers, automatically scaling resources up or down based on demand.