The next phase of the AI race may well be about power supply more than models and chips

In the AI ecosystem, the US has been leveraging its dominance in chip design by strategically limiting exports to China, while Beijing has exerted pressure on the US through its control of rare-earth materials needed to make chips, magnets and other components of advanced technology.

But as the scale of the AI industry and its reliance on computing power grows, the bottleneck will move from chips to electricity because all the data centres in the world will not help if they lack a continuous supply of affordable energy. The International Energy Agency estimates that roughly 20% of planned global data-centre capacity will be at risk by 2030, owing to grid bottlenecks and interconnection queues. And as energy supplies are constrained, costs will rise, eventually trickling down to households and firms.

Which country will dominate this next leg of the race? China has certainly made a statement with its massive build-out of energy supply and distribution infrastructure, much of which focuses on renewables. According to the Financial Times, Chinese investments in clean energy cover everything from solar and hydropower to the hardware needed to move cheaper inland power to coastal demand centres, thereby lowering costs and improving reliability.

China has also invested massively in manufacturing, driving down the price of a solar panel by a factor of 20. All told, it is now capable of adding between 500 gigawatts and one terawatt of capacity per year.

Moreover, China has matched its carefully planned industrial policy with equally strong local execution. To offset the higher cost of using domestic chips, for example, local governments offer electricity subsidies for data centres under their jurisdiction. If these facilities are powered by domestic chips, they can cut their power bills by as much as 50%.

The US effort is rather unimpressive by comparison. According to the Financial Times, “China installed 429GW of new power generation capacity in 2024, more than six times the net capacity added in the US during that time.” Even as America’s locally planned grids face vast around-the-clock demand from data centres, US industrial policy is failing to give electricity the attention it deserves.

For example, OpenAI and its partners are planning to build data centres that will require 10GW of capacity, which is on the order of New York City’s summer peak load electricity usage. But, while data centres can be built in a couple of years, researchers at the Deloitte Research Center for Energy & Industrials point out that completing the transmission infrastructure needed to move energy takes almost a decade.

Such time frames are woefully misaligned with the blistering pace of private capital expenditure. A Bloomberg analysis of tens of thousands of pricing nodes already shows signs of distress in US energy markets. Wholesale electricity in some US regions near large data-centre hubs is up to 267% more expensive than it was five years ago. To keep its data-centre build-out on track, the US will need a structural overhaul of its energy policy, its grids and their interconnections.

Nor is the race down to just the US and China. The bottleneck that could set America back might represent an opportunity for Europe, which is running out of time to claim a seat at the AI table.

Europe’s energy infra is a potential strategic asset in the sense that Hirschman identified. Among the EU’s core strengths are clean energy hardware and knowhow. Over one-fifth of all clean and sustainable technologies globally are developed in the EU.

The bloc also has deep engineering capacity in grid equipment and storage, and its electricity system is among the world’s most interconnected. Europe’s energy policy cites grids as a strategic asset for autonomy and security, and its factories making net-zero technologies are to meet “at least 40% of [the EU’s] annual deployment needs by 2030.”

That said, Europe’s decarbonization has run into obstacles. With high energy costs threatening to hinder growth, energy generation and grid capacity must be ramped up.

The European Network of Transmission Systems Operators for Electricity has advanced a useful proposal. But while planning is done at the European level, execution remains locally bound. As a result, the average grid project takes more than 10 years, half of which is spent on permitting. The European Parliamentary Research Service says current projected grid investments are a mere 10-15% of what’s required, and more than 500GW of offshore projects in the EU is stuck in the queue.

The next leg of the AI race will be about energy. By fostering domestic chip production and consumption while improving electricity infra, China’s strategy addresses every dimension that matters.

Meanwhile, the US is failing to look ahead, complacent in its current position as the designer of first-in-class chips and AI foundation models. And though Europe is uniquely positioned to make AI’s energy demand cleaner and more secure, it may lack the institutions needed to accelerate progress. For Europe to be a challenger, it too must direct its energies toward energy for AI. ©2025/Project Syndicate

The author is a teaching assistant at Bocconi University and the founder and editor of MOPS, a newsletter on macro-finance, technology, and policy.