Modal is a serverless cloud platform purpose-built for AI and data workloads. It allows developers to run inference, training, batch processing, and sandboxed code execution with elastic GPU scaling, sub-second cold starts, and a code-first developer experience that does away with YAML and traditional configuration files. Workloads are defined entirely in Python — functions are decorated to run on Modal's infrastructure, with GPU type, dependencies, and concurrency expressed as code.
The company was founded in 2021 by CEO Erik Bernhardsson, previously machine learning lead at Spotify and CTO at Better.com, alongside co-founder Akshat Bubna, formerly an engineer at Scale AI. Bernhardsson is also widely known in the open-source community as the creator of the Annoy approximate nearest neighbour library.
In September 2025, Modal raised an $87M Series B led by Lux Capital, with participation from Redpoint and Amplify Partners. The round valued the company at $1.1B, making it a unicorn and bringing total funding to roughly $111M. The company markets itself as AI-native cloud infrastructure built from the ground up for the realities of large-model serving, rather than as a wrapper around traditional virtual machines.
Modal's technical differentiation centers on three things: very fast cold starts, scale-to-zero serverless economics, and a developer experience optimized for ML engineers rather than DevOps teams. It supports LLM serving, multi-node training, sandboxed code execution for agentic workloads, and large-scale batch jobs such as embeddings generation. The platform has gained traction with AI startups that want to avoid the operational complexity of managing Kubernetes and GPU clusters.
Modal competes with hyperscaler offerings like AWS SageMaker and with newer AI-native platforms such as Replicate, Baseten, and Banana. Its core bet is that purpose-built infrastructure with strong Python ergonomics will continue to win developer mindshare in AI-first companies, even as cloud providers race to add comparable abstractions.