The AI boom is straining the power grid, and data center operators are getting nervous

Business Insider

04-08-2025

Data center operators are increasingly worried about power amid rising electricity demand from AI.
An Uptime survey found that 36% of data center operators are "very concerned" about power constraints.
Big Tech plans to spend at least $350 billion on data centers in 2025.
Concerns over forecasting data center capacity, or planning for power needs, have grown by 9% since 2023, Uptime Institute found in its annual survey of data center owners and operators. The industry advisory said it considers the jump "significant."
The survey found that rising costs are the chief concern for data center operators this year, as they were in 2024, but concerns about power supply are quickly catching up.
Thirty-eight percent of survey respondents said they are "very concerned" about rising costs, 36% said they are "very concerned" about power constraints, and the same number of respondents said they are "very concerned" about demand forecasting.
According to the International Energy Agency, electricity demand from data centers worldwide is on track to double by 2030.
In the US, Big Tech companies are spending hundreds of billions on infrastructure— including data centers — needed to deploy and scale AI. Amazon, Microsoft, Meta, Google, and Apple are projected to spend upwards of $350 billion collectively on data centers in 2025.
Last week, as four of the five companies reported quarterly earnings, each told investors they would spend more this year than initial projections accounted for.
While rising costs — driven by inflation, labor shortages, and higher energy prices — can be solved for, data center operators don't control the electric grid.
Power is the biggest constraint facing Amazon as it grows its AI business, CEO Andy Jassy said.
Attempts at energy efficiency are not proving helpful in easing data center power constraints, the Uptime survey found.
As AI demands more power, the industry's power usage effectiveness—a metric used to track a facility's level of energy efficiency — has remained stagnant for the last six years.
While novel cooling technologies for data centers continue to emerge, the survey found that older facilities and "region-specific barriers" have prevented notable progress in energy efficiency.
Data center operators have yet to fully embrace AI when it comes to using the technology to help manage their own facilities.
"Trust in AI for data center operations depends on the use case: most operators would allow its use for analyzing sensor data and predictive maintenance tasks, but not for configuration changes, controlling equipment, or managing staff," the survey said.