Modal takes a distinctive approach to GPU infrastructure: rather than exposing servers, containers, or YAML configs, it lets you decorate Python functions with @modal.function and runs them on cloud GPUs. The result is developer experience that feels closer to writing a script than configuring infrastructure.

Founded in 2021, Modal has built a strong reputation among ML engineers who resent the operational overhead of traditional cloud. Its serverless model is the headline product, but the platform also supports dedicated and single-tenant deployment for teams with steady workloads or compliance needs.

Cost-wise, Modal's serverless premium can make it expensive at high utilisation — once a GPU runs more than a few minutes per hour, dedicated alternatives become cheaper. But for bursty, irregular, or experimentation- heavy workloads, the developer-time savings often justify the cost.