(This is a repeat of an item issued on Wednesday. The opinions expressed here are those of the author https://www.reuters.com/authors/ron-bousso/, a columnist for Reuters)
By Ron Bousso
LONDON, Feb 25 (Reuters) – Big Tech’s race to dominate artificial intelligence may soon hit a nasty road bump as U.S. electricity grids struggle to keep pace with the big-spending hyperscalers. America’s technology giants, including Microsoft, Amazon, Alphabet and Meta, have in recent months announced plans to spend over $600 billion on AI in 2026 alone. The investment wave has already fuelled unease among some investors about the profitability of this strategy. They have reason to be worried, as the ambitious U.S. AI expansion plans are likely to be hobbled by severe power-infrastructure bottlenecks, including turbine shortages, slow grid expansion and regulatory red tape.
Data centers used for AI model training and deployment require enormous amounts of energy for processing and cooling. The largest U.S. sites consume over a gigawatt (GW) of continuous load, enough power to supply up to 850,000 homes.
The planned rapid build-out of these electricity-hungry facilities, often in remote locations, will frequently require the construction of independent energy plants powered by gas, renewables or nuclear technologies.
Energy consultancy Cleanview has already identified 46 data centers that plan to build their own power plants, primarily with gas-fired generation. Their combined 56 GW of capacity represents around 30% of all planned U.S. data-center capacity, the consultancy said.
And soon developing independent power systems may not be a choice but a requirement.
In his
State of the Union address
on Tuesday night, President Donald Trump, who has championed U.S. AI growth, said tech companies “have the obligation to provide for their own power needs.”
“They can build their own power plants as part of their factory, so that no one’s prices will go up,” Trump said. But the pressure is already building. Annual U.S. power consumption hit a second consecutive record high in 2025, reaching 4,195 terawatt-hours. Electricity prices nationwide have also risen by an average of 7% in the year to January, according to government data. Electricity demand is expected to rise further, by close to 2% per year on average between 2025 and 2030 – more than twice the pace of the past decade – driven largely by data-center expansion, according to the International Energy Agency, putting additional strain on grids.
BOTTLENECKS ALL AROUND
That squeeze may be felt soon.
PJM Interconnection, the largest power grid operator in the U.S. controlling around 180 GW of power flows across 13 states, warned earlier this month of potential power-supply shortfalls of up to 60 GW over the coming decades due to accelerated demand growth from data centers. The company also warned that the U.S. grid could lack sufficient capacity and reserves by 2027, raising the risk of blackouts. The grid operator last month unveiled plans to require large power users – primarily data centers – to either develop their own power supply or agree to connect under a framework allowing PJM to reduce power output.
One of the country’s other major power networks, the Electric Reliability Council of Texas (ERCOT), also risks being overwhelmed by surging data-center demand.
In December, ERCOT said 226 GW of large-load projects – primarily data centers – were seeking connections to the grid, equivalent to roughly three times current total U.S. data-center capacity. Many of the requests are for projects exceeding 1 GW. On top of all this, the data centers will likely struggle to source the gas turbines required to power many of them. Manufacturers of gas turbines, including GE Vernova, Siemens Energy and Mitsubishi Power, have warned that they cannot meet surging global demand, particularly from power generation. Executives at Siemens Energy and GE Vernova have said they are sold out for years, with delivery slots for large turbines stretching well into the late 2020s.
THE GRIDS GRIND The surge in U.S. power demand is attracting large investment in the country’s ageing grid. In Texas, for example, ERCOT plans to increase its annual spending to $585 million in 2027 from $414 million in 2025, though it is questionable whether that will be enough to meet the need.
The U.S. is not the only market playing catch-up. Global investment in power grids has also lagged the deployment of new generation capacity.
The IEA said in a recent report that globally more than 2,500 GW of projects – including renewables, battery storage and large-load developments such as data centers – remain stuck in grid-connection queues, putting around a fifth of global data-center build-out at risk of delays. Meeting global electricity demand through 2030 would require annual grid investment to rise by 50% from today’s $400 billion, alongside a significant scaling-up of grid-related supply chains, the IEA said.
The rush to build data centers to support the global AI arms race is set to become a defining economic feature of this decade, if not this century.
Meeting the required power demand is likely to prove tricky, however, meaning the AI future might be held back by the very real limitations of today’s physical world.
(The opinions expressed here are those of Ron Bousso, a columnist for Reuters) Enjoying this column? Check out Reuters Open Interest (ROI), your essential new source for global financial commentary. Follow ROI on LinkedIn, and X. And listen to the Morning Bid daily podcast on Apple, Spotify, or the Reuters app. Subscribe to hear Reuters journalists discuss the biggest news in markets and finance seven days a week.
(By Ron Bousso; Editing by Marguerita Choy)
