Google Used AI to Make Its Data Centers 40% More Efficient
Meta later did the same thing, in a loop where AI improves the buildings that produce AI.
Lightspeed Venture Partners · The 7 Layers of the AI Data Center Stack ft. Guru Chahal | LightworkGoogle once used artificial intelligence to make its own data centers 40% more efficient.
It happened about a decade ago. According to Guru Chahal, an investor who backs data center startups, Google turned its own learning systems on its own data centers, the giant buildings that produce AI, and asked them to make those buildings more efficient. The AI studied how power and cooling were being used across the buildings and found ways to optimize them. By 2016, those data centers were 40% more efficient.
Meta did the same thing in 2021 and 2022. Chahal expects more of it to come, a circular improvement in which AI makes the buildings that make AI more efficient.
Questions people ask
What uses all that energy in a data center?
The computers need power to run, and they produce heat that must be removed with cooling. Google's project studied the utilization of power and cooling in its buildings, according to Chahal, and then optimized it.
Does that mean AI will solve its own power problem?
Chahal does not go that far. He says only that he expects data center operations to become a lot more efficient thanks to AI. The Google and Meta examples show improvement, not a complete fix.