The US AI boom is facing an electric shock
The race for artificial intelligence dominance by Big Tech may hit a snag soon as U.S. power grids struggle to keep up with hyperscalers who spend a lot of money. Microsoft, Amazon Alphabet, and Meta are among the technology giants in America that have announced in recent months plans to spend over $600 billion in AI by 2026. Investors are already unsure about the viability of this strategy due to the investment wave. The ambitious U.S. AI plans will be hampered by power-infrastructure bottlenecks such as turbine shortages, slow 'grid expansion, and regulatory red tape.
The processing and cooling of data centers that are used to train and deploy AI models requires enormous amounts energy. The largest U.S. site consumes over a gigawatt of power, which is enough to power up to 850,000 households.
These electricity-hungry installations, which are often located in remote areas, will require independent energy plants that use gas, renewable technologies, or nuclear technology.
Cleanview, an energy consultancy, has identified 46 data centres that are planning to build their power plants themselves. These centers will primarily use gas-fired generators. The consultancy stated that their combined 56 GW capacity represents about 30% of the planned U.S. capacity for data centers.
Soon, developing independent power systems will not only be an option but also a requirement.
In this article,
State of the Union Address
On Tuesday, Donald Trump, the U.S. President who has been a champion of AI growth in the U.S., said that tech companies have "the obligation to provide their own energy needs."
Trump said, "They can build their power plants as a part of their factory so that nobody's prices will increase." The pressure is building. The annual U.S. electricity consumption reached a record level of 4,195 terawatt hours in 2025. According to data from the government, electricity prices in the United States have risen an average of 7.5% over the past year. According to the International Energy Agency, electricity demand will continue to increase, with an average annual growth rate of 2% between 2025 and 3025 - twice as fast as the previous decade. This is primarily due to data center expansion.
BOTTLENECKS AROUND
Soon, you may feel the squeeze.
PJM interconnection, the U.S.'s largest grid operator, which controls around 180 GW in 13 states, has warned earlier this month that data center demand will increase rapidly, resulting in a potential shortfall of 60?GW. The company warned that by 2027 the U.S. Grid could be lacking in reserves and capacity, increasing the risk of blackouts. Grid operator PJM announced plans last month to force large power users – primarily data centres – to develop their own energy supply or to agree to connect to a framework that allows PJM the ability to reduce power output.
The Electric Reliability Council of Texas, (ERCOT), is also at risk of being overwhelmed by the surge in data-center demand.
ERCOT reported in December that 226 GW, primarily data centres, were seeking to connect to the grid. This is roughly equivalent to three times the current capacity of data centers in the United States. Most of the requests concern projects larger than 1 GW. In addition to all of this, data centers may struggle to find the gas turbines needed to power them. Gas turbine manufacturers, such as GE Vernova Siemens Energy, and Mitsubishi Power have warned they can't meet the global demand for gas turbines. This is especially true of power generation. Siemens Energy and GE Vernova executives have stated that they have been sold out for many years.
The aging grid in the United States is attracting large investments. ERCOT, for instance, plans to spend $585 million annually in 2027, up from $414 in 2025. However, it's questionable if that amount will be sufficient to meet demand.
Not only the U.S. market is playing catch up. The global investment in power grids is also behind the deployment of new generation.
In a recent report, the IEA?said that more than 2,500 GW projects - such as renewables, batteries and large-load development, like data centers – remain stuck in grid connection queues. This puts around a quarter of global data center build-outs at risk of delay. The IEA stated that to meet global electricity demand by 2030, grid investments would need to increase from $400 billion today, along with a significant scaling up of grid-related supplies chains.
The rush to build more data centers in order to support the global AI arms races is likely to be a major economic factor of this decade, and possibly even this century.
The AI future could be held back due to the limitations of the physical world today.
The opinions expressed in this article are those of Ron Bousso a columnist at. Open Interest (ROI) is your new essential source of global financial commentary. Follow ROI on LinkedIn and X. Listen to the Morning Bid podcast daily on Apple, Spotify or the app. Subscribe to the Morning Bid podcast and hear journalists discussing the latest news in finance and markets seven days a weeks.
(source: Reuters)
