Why American data centers can't plug in (worksinprogress.co)

A significant bottleneck is emerging in the expansion of AI-driven data centers across the U.S.—the ability to connect these facilities to the electrical grid. A landmark project called Stargate, spearheaded by OpenAI and SoftBank in Abilene, Texas, is expected to cost over $40 billion and consume 1.2 gigawatts of electricity at peak load, equivalent to the needs of approximately 313,000 average American homes. Industry leaders agree that the escalating energy requirements of burgeoning AI models and data centers underscore a critical infrastructure challenge: the time and process needed to connect to the grid. Interconnection delays have climbed from an average of 20 months in 2005 to 55 months in 2023 due to a "first-come, first-served" model that prioritizes less critical projects, stifling potential growth. As demand for electricity rises, power grids are struggling to deliver reliable energy where it's needed most, leading to significant economic inefficiencies. The strain on grids has resulted in increased reliance on local, often expensive, energy sources, causing costs to spike. Major operators, like ERCOT and PJM, predict that they won't meet future demands unless capacity increases significantly. Addressing these interconnection and grid efficiency issues is crucial for facilitating the expansion of AI infrastructures and ensuring the energy needs of future technologies are met, marking a pivotal moment for the AI/ML community and the broader economy.