Is the US energy grid prepared to satisfy the calls for of knowledge centres? | USAEMALL.com

Welcome to Energy Source, coming to you from New York. President Donald Trump announced a 90-day pause in additional tariffs on countries that were willing to negotiate with the US. The news sent stocks surging but oil markets have remained volatile.

Although the international benchmark Brent crude jumped on Wednesday to settle at $65.48 a barrel, it had fallen below $60 a barrel for the first time in four years before Trump announced he would pause tariffs.

Despite the recovery, crude prices have fallen sharply this year. My colleagues Jamie Smyth and Myles McCormick reported shale producers are facing their gravest threat in years as they worry about the effects tariffs may have on the industry.

In today’s Energy Source we look at how the ageing US power grid and supply chain constraints may slow the pace of data centre rollout in the country, complicating Washington’s race to lead in artificial intelligence.

Thanks for reading, Alexandra

Is the US power grid ready to meet the demands of data centres?

To win the AI race the US needs to build a significant number of data centres across the country, but experts warn the nation’s ageing power grids and supply chain constraints may not be able to support their deployment.

Utilities are receiving an unprecedented number of requests for new power capacity from the industry. Sempra said last year its Texas power utility Oncor Electric received requests from data centres to add a total of about 119 gigawatts of potential load to connect to the grid.

But Chris Seiple, vice-chair of energy transition and power and renewables at Wood Mackenzie, said the US power grid’s limited capacity could curb the pace of data centre rollout.

“The amount of data centre capacity that gets added over the course of the next five years will be determined by how much energy we can actually add to the grid to support it,” he said.

This is a global problem. A new report published by the International Energy Agency on Thursday found grid constraints could delay about 20 per cent of global data centre capacity planned for construction by 2030.

But a surge in energy demand in the US, after two decades of flat growth, has made the problem acute. While the IEA estimates data centre growth will account for a relatively small share of future electricity demand, it can place significant daily pressure on grid operators and utilities, with AI-related data facilities typically requiring far more energy than traditional ones.

“The [data centres] that are being talked about now are a gigawatt in capacity . . . a gigawatt is about the size of the city of San Francisco,” said Jim Robb, chief executive of the North American Electric Reliability Corporation, a regulatory body that monitors the reliability of the power system in the US and Canada.

“You can imagine how challenging that is going to be for us to build not only the power generation but also the delivery infrastructure to get the power from where you can generate it to where it’s going to be consumed,” he added.

The US power grid is also facing an increase in demand linked to the electrification of transportation and reshoring of manufacturing. But the capacity crunch linked to AI technologies raises specific challenges because it requires significantly more energy resources and upgraded infrastructure to meet the demands of hyperscale data centres.

While most data centres have prioritised using renewable energy, that alone cannot meet the power demand, thereby creating the need for more natural gas-fired power plants.

Seiple said: “You need to build plants not only to meet demand growth, to respond to the changes in capacity accreditation rules, but also to replace all of these retiring plants.”

Building a new combined cycle power plant, he added, is costly and has doubled in price since before the outbreak of the Covid-19 pandemic to be an average of over $2,000/ kW.

Utilising renewables such as wind and solar also requires the need to update transmission lines to transport the energy to the grid. But Nerc’s Robb said multi-state transmission projects can take 15 to 17 years to develop.

“We need to figure out how to shrink 15 to 17 years to three to five years,” he added.

To make matters worse, key components of grid infrastructure including transformers and gas turbines are facing supply chain constraints.

Wood Mackenzie’s Seiple said lead times for transformers, which are crucial to transmission infrastructure, have tripled since 2021.

“If you did not order your transformer three years ago, you’re going to find it very challenging to get a transformer,” he added.

Trump’s recent spate of tariffs may worsen supply chain challenges and could potentially raise costs for critical infrastructure.

“Tariffs are going to make grid projects more expensive,” said Karen Wayland, chief executive of GridWise Alliance, a coalition that advocates for modernising the US power grid. “It’s going to exacerbate supply chain constraints.”

The uncertainties around the effects of the levies may also complicate future investment decisions for businesses. Utilities have already grappled with uncertainty surrounding data centre development.

“Utilities in the past have never had so much risk associated with a customer,” Seiple said. “They don’t know if all 10 data centre projects that want to connect to their system will get built.”

Wayland said it was difficult for utilities and grid operators to plan because they know there will be increased power demand but they do not know “exactly how much”.

Some experts have said demand for data centres may be overstated. A recent Energy Intelligence report said AI may not require as much energy demand as stated. They cite Nvidia as an example, which has improved its chips to be 45,000 times more energy efficient at some large language model tasks in just eight years.

“The general rule of technology is as it improves you’re bringing efficiency along,” said Michael Collins, director of energy transition research at Energy Intelligence.

Earlier this year, Chinese AI start-up DeepSeek revealed it could develop a large language model by using far less energy than its US rivals. While this raises the possibility that the technology may not need as much power as forecast, utility companies are still seeing a surge in demand from data centres.

Dominion Energy said it has almost doubled its data centre customer demand since July and Duke Energy told investors in February that it has not seen a pullback either, with more electricity demand coming from facilities related to generative AI. (Alexandra White)

Job moves

• Energy Vault appointed Dylan Hixon to its board of directors.

• Andrew Bald has been named managing director at Xstate Resources, as executive chair Andrew Childs transitions to a non-executive role.

• ENRG Elements appointed Paul Ingram as non-executive chair.

Power Points

• The US has revoked licenses for BP and Shell gas projects in Venezuelan waters, which could leave a crucial liquefied natural gas project in Trinidad and Tobago vulnerable to gas shortages.

• Japanese shipowner Mitsui OSK Lines criticised proposals by its rival AP Møller-Maersk to place a higher regulatory cost on the use of LNG than zero-carbon fuels to power ships.

• The world must “immediately” make substantial investments in mining uranium to meet surging demand for nuclear energy, according to the latest industry report.

Energy Source is written and edited by Jamie Smyth, Myles McCormick, Amanda Chu, Tom Wilson and Malcolm Moore, with support from the FT’s global team of reporters. Reach us at energy.source@ft.com and follow us on X at @FTEnergy. Catch up on past editions of the newsletter here.