The most ironic part of this wave of AI enthusiasm is that, despite being the most virtual industry, it is actually forcing companies and capital back to the most tangible resource—electricity.

Massive data centers have become real electricity monsters. Training and inference of large models require thousands of GPUs running nonstop, with racks, liquid cooling, and supporting facilities all consuming power. The electricity demand of a top-tier AI data center is comparable to that of a city with tens of thousands of residents. Many such data centers already exist in the U.S. and continue to grow.

But here’s the problem. The traditional power grid and lengthy approval processes create two major hurdles. Critical equipment like transmission lines and transformers are in severe shortage. Building new transmission lines from blueprint to connection often takes years, and obtaining permits for new power plants is an even longer wait. For tech companies, the value of electricity has surpassed ordinary cost considerations—it is now a key factor determining whether projects can be implemented and delivered on time. Launching a cloud computing center months ahead of schedule could be worth billions in business value.

The direct consequence of the grid lagging behind is the emergence of issues around electricity prices and cost sharing. Residents in data center clusters face increasing risks of higher electricity bills. Research institutions and media have issued warnings, and policymakers are beginning to ask: who will pay the bill for AI electricity consumption?

Tech giants are forced to make gestures—investing in grid upgrades themselves and refusing to pass the pressure onto ordinary households. But saying it is easy; executing it is very complex. The benefits of grid upgrades are usually shared by society, but the construction costs fall on the companies, and this calculation often doesn’t seem very cost-effective.

For this reason, some giants are exploring other solutions. Some are actively developing nuclear power for self-sufficient energy supply, some are building dedicated power plants, and there are even rumors of relocating power infrastructure directly. These moves may seem aggressive, but they reflect a reality—relying solely on the public grid is no longer enough. Controlling their own energy sources is essential to ensure the continuous operation of AI projects. Power has shifted from a production factor to a strategic asset.