// power grid

Source: Worksinprogress

The explosive demand for electricity from AI training and cloud computing is colliding with aging electrical grids designed for a different era of consumption, forcing utilities and policymakers to confront decades of underinvestment in transmission capacity. This isn't a problem that Moore's Law or software optimization can solve—it requires physical infrastructure upgrades that take years to permit and build, creating a hard constraint on where and how quickly hyperscalers can expand their operations. Power availability is already a limiting factor in real estate value and regional economic development, rivaling fiber connectivity or labor in site selection.

Source: The Parnas Perspective

The explosive growth of AI infrastructure—driven by companies like OpenAI, Google, and Meta building massive data centers—is colliding with grid capacity in regions like Virginia and Texas that lack the generation and transmission infrastructure to support these loads. Utilities are already reporting strain, and the energy demands of training and running large language models are doubling every few months. The infrastructure gap will worsen without significant capital investment in power generation and grid modernization that currently isn't happening at scale. Continued AI growth without addressing energy constraints is becoming untenable. This bottleneck could force either massive government intervention, a slowdown in model development, or both.