MODAL

Erik Bernhardsson was the CTO of Better.com (the mortgage company) after years at Spotify where he built the recommendation algorithms. He noticed a pattern: every AI and data science project spent enormous time on infrastructure — setting up GPU clusters, managing Docker containers, dealing with dependency hell.

The actual interesting work (the model, the algorithm) was a small fraction of the effort. Modal's vision: write a Python function, Modal makes it run anywhere at any scale, you pay only for what runs.

No infrastructure. No configuration.

Just Python decorated with one line.

Modal charges for compute time consumed. You write Python code, decorate functions with @modal.function, and Modal provisions the infrastructure (containers, GPUs, CPUs) on demand, runs your function, and you pay for the seconds of compute used.

There is no minimum spend, no idle server cost, and no infrastructure to manage.

Modal serverless Python functions (CPU and GPU), Modal volumes (persistent storage), Modal web endpoints, Modal cron jobs

Modal grew through developer communities — it is a tool that experienced developers genuinely love using, which creates organic word-of-mouth. Support for GPU workloads (NVIDIA A100s, H100s) positioned it as an AI inference and training platform.

Nat Friedman and Daniel Gross (who ran Y Combinator and are deeply connected to the AI developer community) were early investors and advocates.

Modal competes directly with AWS Lambda, Google Cloud Run, and a growing list of GPU cloud providers (Lambda Labs, CoreWeave, Together AI) for AI compute workloads. The differentiation is developer experience: Modal is dramatically simpler than managing your own cloud infrastructure, but whether that simplicity justifies the cost premium versus running your own spot instances is a real question for large workloads.

Redpoint Ventures, Sequoia Capital, Nat Friedman, Daniel Gross, Elad Gil