Who makes AI-driven cooling control for AI data centers

This is software that sits above the building management system and the controllers, reads the live sensor data, and works out better setpoints than a fixed sequence would. The gains come from the fact that conventional control is tuned conservatively: a plant runs colder than it needs to because the margin protects against conditions that are rarely present, and a model that sees actual conditions can hold a tighter setpoint safely. The honest description of what these systems do is optimisation with a safety envelope, not autonomy. They write setpoints into the existing controls and the existing interlocks stay in force underneath, which is both a limit on what they can achieve and the reason operators will run them at all. The savings are real but they are a percentage of cooling energy, not a change in how the plant works, and they depend heavily on how well the site is instrumented to begin with. AI halls are an attractive target because their loads are large and unusually steady, which is the easiest case to model. What to look for is whether the vendor's claims come from measured before-and-after on comparable sites, how the system behaves when a sensor fails, and what the fallback is — a plant that has come to depend on an optimisation layer with no tested manual mode has added a dependency rather than removed one. The suppliers here are small specialists, not the controls majors.