GPU compute

Runs your services, databases and GPU jobs on their cloud or inside yours.

The place an AI agent's code actually runs — microVM isolation, GPUs and your own VPC — which is where the agent-sandbox question stops being a model question and becomes an infrastructure one.

GPU clusters plus reinforcement-learning tools to train a model on your own work.

They are building the part of the stack most people cannot see — where a model is trained, given an environment to act in, and graded — and they publish the engineering behind it in the open, which makes them a primary source for us whether or not the relationship goes anywhere.

A marketplace for renting other people’s spare GPUs by the hour.

Vast is the GPU spot-price layer MenFem's /data product reads — the hourly floor for a rented GPU-hour, which is the denominator under the task-price series we publish.