New Power Challenge for AI Data Centers: Frequent Damage to Critical Equipment from Power Fluctuations Tests Trillion-Dollar Investments

The massive electricity demand of AI is well known, but a more hidden and thorny problem is emerging: severe fluctuations in power consumption at data centers are damaging critical equipment, threatening facility reliability, and exporting instability risks to the entire power grid.

According to a Bloomberg report on August 6, core systems of AI computing facilities, including batteries, generators, and cooling units, are failing or being prematurely retired due to abnormal power shocks. At xAI’s Colossus supercomputing facility in Memphis, Tennessee, gas turbines have developed cracks; similar turbine cracking issues have appeared in several smaller data centers across the UK. Some installed voltage-stabilizing batteries need replacement within weeks or even months under high-intensity operation.

The market impact of this issue is becoming apparent. According to reports, an insider involved in data center financing revealed that the actual operational uptime of some facilities has dropped to around 80%, far below the design expectation of year-round uninterrupted operation. If the problem remains unsolved, it will directly impact investors in related projects within the next 12 to 24 months. Meanwhile, the North American Electric Reliability Corporation (NERC) issued a rare Level 3 alert earlier this year, requiring large data centers to immediately address grid stability risks.

The Root of Power Fluctuations: "Millisecond-Level" Shocks from GPU Clusters

The power consumption patterns of AI data centers differ fundamentally from those of traditional data centers. While traditional facilities have relatively stable power usage, AI computing—especially during model training phases—drives hundreds of thousands of GPUs to start and stop synchronously, causing power consumption to surge or plummet within milliseconds.

Shannon Miller, founder and president of Mainspring Energy Inc., vividly described this phenomenon: In a 1-gigawatt data center the size of Boston, half of its power load might flicker on and off every few seconds. Some AI parks planned in Texas and the U.S. Midwest exceed 5 gigawatts in scale, with average power consumption approaching that of New York City.

Drew Baglino, former Tesla executive and founder of Heron Power Electronics Co., pointed out that instantaneous power consumption in AI data centers can sometimes spike to more than 50% above design capacity—"A 1-gigawatt facility might instantly consume 1.5 gigawatts." His company is developing power fluctuation management equipment for Nvidia’s next-generation servers, expected to launch in 2027.

Amber Villegas-Williamson, chief consultant at Uptime Institute UK, compared this shock to repeatedly slamming on the accelerator while driving:

"It’s like constantly over-revving the engine, which causes much faster wear than steady driving."

Equipment Damage Is a Reality: From Cracks to Arc Flash

According to more than thirty power industry professionals in the U.S. and Europe interviewed by Bloomberg, the physical stress AI data centers place on equipment is evident.

The report stated that multiple insiders indicated that the crankshafts of small natural gas internal combustion engines used in data centers have fractured. At xAI’s Colossus facility, gas turbines developed cracks, prompting operators to install batteries to smooth power fluctuations and reduce turbine load. Andrew Cunningham, CEO of GeoPura Ltd., also confirmed that similar turbine cracking issues have occurred in several smaller data centers in the UK.

Jennifer Scanlon, CEO of UL Solutions Inc., pointed out that equipment cracking or wear can lead to arc flash—where electricity jumps between conductors—thereby damaging AI chips.

Uptime Institute and several industry insiders stated that batteries installed for voltage stabilization sometimes need replacement within weeks or even months. Although voltage-stabilizing equipment such as batteries, capacitors, transformers, and flywheels already exists, new data centers have significantly under-deployed these technologies in the rush to build AI computing capacity.

Notably, this problem is global, appearing in regions ranging from the Middle East and Africa to Europe and the United States.

Declining Reliability Hits Revenue: Losses Can Reach Hundreds of Thousands of Dollars Per Minute

The direct consequence of equipment damage is facility downtime, which comes at a steep cost.

Jason Hoffman, Chief Strategy Officer at Switch Data Centers, stated that the financial cost of premature equipment failure is "not primarily the cost of replacing a pump or circuit breaker, but the loss of revenue from expensive computing assets being offline." It is estimated that revenue losses from downtime range from thousands to hundreds of thousands of dollars per minute, depending on the facility type and workload.

According to reports, a specific case involves the 2.67-gigawatt AI park in West Texas, jointly developed by Joulent Inc. and Chevron. To ensure the 99.999% reliability required by Microsoft, additional engineering time had to be reserved in the schedule, pushing the power-on date from 2027 to 2028. Chris James, CEO of Joulent, stated that this adjustment was made specifically to address the engineering challenges posed by power fluctuations.

The report cited an insider involved in data center financing who revealed that while the industry generally assumes facilities will operate uninterruptedly year-round after going live, the actual operational rate of some facilities has approached only 80%. If the problem cannot be resolved, it will have a substantial impact on investors in related projects within the next 12 to 24 months.

Analysts point out that the hundreds of billions of dollars in capital expenditure by hyperscale cloud providers have already put investors and lenders on edge. The issue of accelerated depreciation due to equipment damage overlaps with previous market concerns about the depreciation speed of GPU racks, leading to deeper doubts about whether AI data centers can deliver on their profitability promises.

Even a few minutes of downtime directly erodes the revenue of data center developers. If actual operational rates remain below expectations for an extended period, the financial models of these projects will face fundamental revaluation. As the investment boom in AI infrastructure has not yet cooled, this structural risk is becoming a new variable that investors must confront.

Grid Stability Emergency: Regulators Issue Rare Alert

Notably, power fluctuations in AI data centers not only threaten the facilities themselves but also export instability to the broader power grid.

According to reports, Sreemant Roy, global product manager and power quality expert at Schneider Electric, warned that such highly dynamic loads "can cause grid instability, potentially leading to blackouts or power outages if not corrected." He specifically noted that data centers could trigger subsynchronous oscillations, damaging connected equipment at other nodes in the grid, "a situation that has caused deep concern among power companies globally."

On the regulatory front, NERC has issued multiple warnings over the past two years, listing data centers as one of the biggest risks to grid stability.

According to a NERC report last September, in an assessment of over 33 gigawatts of operating data centers in the U.S., about three-quarters of the load models were "insufficient to reflect the dynamic behavior of data centers." Earlier this year, NERC issued a rare Level 3 alert, requiring large data centers to submit response plans by August 3.

In the face of these challenges, various links in the industry chain are actively seeking solutions.

Nvidia stated that since the release of the Blackwell GPU in 2024, it has strengthened cooperation with power experts. Dion Harris, Senior Director of Hyperscale Infrastructure Solutions at Nvidia, said that the company is working towards smoother deployment across all stages of data center construction, design, engineering, and power delivery.

At the operational level, some data center users have adopted "sidecar computing" technology—running virtual computing tasks that are not part of the training process—to maintain stable GPU operation and smooth out power fluctuations. However, critics point out that this method causes additional power waste at a time when electricity demand is already soaring.

At the policy level, the U.S. Department of Energy commissioned the National Rocky Mountain Laboratory (NLR) near Denver, Colorado, last year to establish a test platform specifically studying how to safely integrate AI into the grid.

Martha Symko-Davies, project manager at NLR, stated that the platform is equipped with GPUs and power generation equipment, allowing developers to test whether their systems can handle the volatility of AI loads. It will also be used to test batteries, software, and other voltage-stabilizing equipment. "We now have the opportunity to get this right," she said.

Risk Warning and Disclaimer

The market carries risks; investment requires caution. This article does not constitute personal investment advice, nor does it consider the specific investment goals, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this content is at the user's own risk.