According to a recent analysis by the Financial Times, as tech giants continue to invest heavily in data centers, the US power grid is facing unprecedented pressure. It is expected that AI data centers will add about 44 gigawatts (GW) of power demand by 2028. However, due to bottlenecks in the power grid infrastructure, only 25 gigawatts of the additional power can be met, resulting in a power gap of about 19 gigawatts, or 40% of the total demand. This situation has raised deep concerns among industry insiders, especially companies that rely on large computing power.
Major tech companies, such as Amazon, Google, Meta, and Microsoft, have already planned to invest over $400 billion specifically for building data centers. Notably, OpenAI has signed infrastructure contracts worth up to $1.4 trillion with partners to ensure access to about 28 gigawatts of power over the next eight years. OpenAI CEO Sam Altman even called this power shortage a "matter of life and death," because without sufficient computing power, the company would be unable to profit at the necessary scale or develop new models.
Additionally, aging power grid infrastructure has exacerbated the issue. Many power poles and transformers date back to the 1960s and 1970s, and lengthy regulatory approval processes have caused the average waiting time from applying for grid connection to actual commercial operation to exceed eight years. This situation has led many developers to submit multiple applications to different utility companies to improve competitiveness, further worsening the queue.
Facing power supply issues, more AI companies are exploring "behind-the-gate generation" solutions. For example, Elon Musk's xAI company used unpermitted natural gas turbine generators in its "Colossus" project in Tennessee. OpenAI also plans to use ten natural gas turbines in its "Stargate" project in Texas, while Microsoft aims to seek power support by restarting the Three Mile Island nuclear power plant.
