Modal

Modal is a serverless AI compute platform headquartered in New York City that allows engineers and data scientists to run Python code on GPU infrastructure without managing servers, containers, or orchestration. Founded by Erik Bernhardsson and a team of infrastructure engineers, Modal reached $50 million in annualised revenue in February 2026.

The platform's core innovation is its ability to import Python functions as GPU-accelerated cloud tasks using a simple decorator pattern—a developer writes standard Python, decorates it with @modal.function(), and Modal handles GPU provisioning, autoscaling, container building, and billing automatically. Modal supports NVIDIA H100, A100, A10G, and T4 GPUs, with cold-start times under two seconds and billing in 100-millisecond increments.

Key use cases include LLM fine-tuning pipelines, batch inference jobs, image and video generation workloads, data processing tasks, and scheduled ML jobs. The platform is particularly popular among ML engineers at AI startups who want to iterate quickly without DevOps overhead—model fine-tuning jobs that previously required managing cloud VMs or Kubernetes clusters can be launched with a single Python script.

Modal's developer experience is considered best-in-class in the GPU cloud market: it supports custom container images, persistent volumes, secrets management, and webhooks out of the box. Pricing is purely usage-based with no minimum spend, making it accessible to individual researchers and startups as well as growing AI companies with production inference workloads.