Modal

Modal is a cloud infrastructure company whose platform lets teams develop, train and serve AI applications at scale. Its pitch centres on instant GPU access, sub-second container starts and native storage, so that customers can serve low-latency inference, fine-tune models and run production-ready sandboxes. The company frames AI as a new class of workload that older infrastructure was not built for, comparable to earlier shifts such as mainframes, databases and the cloud. To deliver the developer experience it wanted, the company built its own infrastructure, including a custom file system, container runtime, scheduler and container image builder.

The product is organised around a few primitives: Functions for serverless compute, Sandboxes for running untrusted code such as agent workloads, and Servers for low-latency LLM inference, supported by a distributed filesystem for images, model weights and datasets. Training support ranges from single-GPU fine-tuning to multi-node runs, while capacity spans more than 20 clouds with routing across regions in North America, Europe, Asia Pacific and elsewhere. Pricing is usage-based and billed by the second, a model that suits the bursty demand typical of inference and experimentation, where reserving GPUs ahead of time can leave expensive hardware idle.

Customers use Modal for generative AI inference, LLM fine-tuning, computational biotech and media processing, and the client companies quoted include DoorDash, Cognition and Ramp. For enterprise buyers it offers workload isolation, granular access controls, audit logs and data residency controls, and it states SOC 2 Type II compliance and HIPAA readiness, which are common requirements for teams in regulated industries. Operating as Modal Labs, Inc., the company has offices in New York, Stockholm, San Francisco and London, and its team includes creators of open-source projects such as Seaborn and Luigi.