Evidence Passport
The price of a degree-hour in data-centre cooling: auditing deep RL against model predictive control
Top story 7 in the issue of 24 October 2026, in 39 AI Cloud & Data Centres.
Original source
The price of a degree-hour in data-centre cooling: auditing deep RL against model predictive control
Benchmarks used to show reinforcement-learning savings put very different implied prices on overheating; checked against two metered tropical sites, the best RL policies traded comfort for energy far more cheaply than real plants would.
Evidence
Evidence: Peer-reviewed (evidence step 5 of 6)
Published after review by independent experts.
The story in three lines
- Key finding
- Learned policies accepted overheating at 40–42.7 kWh per degree-hour, versus 0.73–1.22 measured at real plants.
- Significance
- Technical: some reported AI cooling savings come from benchmark weights rather than better control.
- Takeaway
- Operators should test AI cooling against tuned classical controllers with realistic penalties.
Later coverage in the journal
No later coverage in the journal yet.
Corrections and updates
No corrections or updates recorded for this item.
This page is generated from the issue and stays at the same address. It adds nothing that is not in the issue, the source link or the corrections log.