TECH & AI
CREDIT: GOOGLE
Google’ s application of machine learning to data centre operations represents the industry’ s most visible and impactful deployment.
Beginning in 2016, DeepMind’ s neural networks were deployed across Google’ s global data centre fleet to optimise cooling systems, achieving a 40 % reduction in cooling energy consumption and 15 % improvement in overall power usage effectiveness( PUE).
The system processes thousands of sensor measurements every minute, including temperatures, power consumption, and equipment settings across vast facilities. Rather than relying on fixed rules, the ML models learn optimal control strategies through reinforcement learning, continuously adapting to changing conditions and workload patterns.
What distinguishes Google’ s approach is the scale of deployment and continuous refinement. The models have been trained on years of operational data, enabling predictions across diverse weather conditions, equipment configurations and workload profiles. The system now makes recommendations that human operators review and implement, with plans for increased automation.
Beyond cooling optimisation, Google has expanded ML applications to power infrastructure management, server utilisation forecasting, and network traffic engineering.
The company’ s open sharing of research findings has accelerated industry adoption, though replicating results requires significant data science expertise and infrastructure investment.
Google’ s work demonstrates ML’ s potential to transform data centre economics, proving that substantial efficiency gains remain achievable even in highly optimised facilities.
datacentremagazine. com 121