Let AI run data-centre cooling, safely
DeepMind's AI directly controlled cooling in Google data centres, with safety checks; DeepMind says savings rose from 12% to about 30% over nine months.
Try it withClaudeChatGPT / OpenAI modelsGeminiGrok
Deployed, Google DeepMind; Google
Why it matters
Data centres, district cooling and buildings use huge amounts of energy, which matters especially in hot climates like the Gulf. Autonomous control with layered safety checks is a model for cutting that energy use.
How it works
Every five minutes, a cloud-based AI takes a snapshot from thousands of sensors and picks the actions that minimise energy use within safety constraints set by operators. The local control system verifies every instruction again before acting. DeepMind measures the savings in kW/ton. MIT Technology Review independently reported the hand-over of control; an expert it quoted noted that cooling is only around 10% of a data centre's energy use. An earlier recommendation-only system cut cooling energy by up to 40%.
Who it's for
Facility managers, utilities, district-cooling operators and data-centre teams.
The prompt
I manage [building, district-cooling plant or data hall]. Below is a CSV of sensor and meter data: timestamp, outdoor temperature and humidity, set-points, chiller and pump status, power use. 1. Describe the energy-use patterns and the biggest drivers of consumption. 2. Propose 5 operational changes (set-points, sequencing, schedules), each with a rough saving estimate, the method behind it, its uncertainty and the rows of data that support it. Say where the data is too limited in time or confounded (for example by weather) to draw a conclusion. 3. For each, list the safety and comfort limits it must respect and how to monitor them. 4. Design a 4-week A/B test plan with a baseline period, so savings can be measured properly. Recommendations only. A qualified engineer must approve any change to the controls. [Paste CSV]
Open inClaude (opens in a new tab)ChatGPT (opens in a new tab)Gemini (opens in a new tab)Grok (opens in a new tab)
Sources
The results are Google's own; savings vary by site.
1. Google DeepMind: Every five minutes; safety constraints; local verification; 12% to about 30% over nine months. Safety-first AI for autonomous data centre cooling and industrial control (opens in a new tab)
2. MIT Technology Review: Independent report on autonomous control. MIT Technology Review: Google gave control over data centre cooling to an AI (opens in a new tab)
3. Google DeepMind: Earlier recommendation system: up to 40% less cooling energy. DeepMind AI reduces Google data centre cooling bill by 40% (opens in a new tab)