AI cooling control at hyperscale
Google first used machine-learning recommendations to cut the energy used for cooling by up to 40%, then moved to direct AI control with layered safety constraints and operator override.
| 2016 | Up to 40% less cooling energy (recommendations) |
|---|---|
| 2018 | ~30% average savings (autonomous) |
| Ramp | 12% → ~30% over nine months |
| Safety | Two-layer verification, operator override |
In 2016, DeepMind's models recommended actions that operators implemented by hand, reducing cooling energy by up to 40% in a live Google data center.
By 2018 the system controlled cooling directly. Every five minutes it predicted the effect of candidate actions, discarded low-confidence ones, verified the rest against operator-defined constraints in the cloud and again against the local control system, and let operators exit AI control at any time.
Bounded autonomy works in production at the largest scale in the industry, and safety comes from layers outside the model: constraints, verification and human override. Conduit makes those layers the interface itself.
Bring us one site. We will show you its real capacity.
A six-week Site Assessment maps the power envelope, the cooling design and the compute plan for one facility, delivered by the engineers who build the platform.