According to several operators, sudden fluctuations in load caused by AI computations lead to premature wear and tear on batteries, generators and cooling systems in data centres.
The rapid fluctuations in electricity demand generated by AI workloads place a heavy strain on data centre equipment: batteries, generators and cooling systems are wearing out or failing much sooner than expected, leading to unforeseen costs and an increased risk of downtime — with every minute of disruption having a direct impact on operators’ turnover.
This phenomenon compounds the wider strains on electricity grids: several facilities will exceed the one-gigawatt power threshold as early as 2026, each operated by a different hyperscaler, and a single AI task can consume up to 1,000 times more electricity than a standard web search — enough to destabilise the electricity supply across entire regions.
Read the full article on the original source
Go to the source →