Modal

Modal is a serverless cloud platform built for Python-first AI and ML workloads, founded in 2021 by Erik Bernhardsson and Akshat Bubna. Bernhardsson previously built Spotify's recommendation system and created the Annoy vector similarity library; Bubna joined from Scale AI. Headquartered in New York City with offices in San Francisco and Stockholm, Modal handles GPU provisioning, container builds, and autoscaling through Python decorators rather than YAML or Terraform.

Key Capabilities:

• Python function decorators for cloud deployment without YAML or Dockerfiles

• Sub-second cold starts on a Rust-based custom container runtime

• On-demand Nvidia GPU access including A100, H100, and B200

• Autoscaling from zero to thousands of CPUs or GPUs

• Container image builder defined directly in Python code

• Multi-cloud capacity pooling across major providers

• Web endpoint deployment with automatic HTTP routing and TLS

• Sandboxed execution for AI agent code generation

• Multi-node distributed training

• Collaborative Jupyter-style notebooks for team ML workflows

• Native S3, GCS, and Azure Blob storage mounting

• Per-second usage-based pricing with public GPU rate cards

• Real-time observability dashboard with logs and metrics

• Open-source Python SDK with a public examples repository

See Modal pricing details →