The explosive growth in demand for computing power, driven in large part by the spread of artificial intelligence, is running headlong into the physical limits of the electrical grid. Data centers that house the servers behind modern digital services consume electricity on an industrial scale, and the speed at which new facilities are being planned has begun to outpace the far slower process of building the power infrastructure to supply them. Energy, long treated as a background concern in technology, is emerging as a central constraint.

The arithmetic is straightforward but daunting. Training and running advanced computing systems requires enormous and sustained quantities of power, and the facilities that perform this work draw electricity continuously, day and night, in concentrations that can rival the consumption of small cities. As operators race to build capacity, the cumulative demand they place on regional grids has grown rapidly, and in some areas the requests for new connections have reached a scale that local systems were never designed to accommodate.

Electricity supply, by contrast, expands slowly. Building new generation, whether powered by gas, nuclear, or renewable sources, takes years from planning to operation, and the transmission lines needed to move power from where it is produced to where it is consumed can take even longer, often delayed by permitting, financing, and local opposition. The grid is a vast and tightly interconnected system that cannot be scaled up on the timeline that the technology sector is accustomed to, and the mismatch between the two clock speeds is becoming acute.

The collision has begun to produce visible effects. In some regions, utilities and grid operators have warned that the pace of new demand threatens to outstrip available supply, raising the prospect of higher prices, strained reliability, or delays in connecting new facilities. Communities that welcomed data center investment for its economic benefits have in some cases grown wary of the burden it places on local power systems and the upward pressure it can exert on electricity costs for households and other businesses sharing the same grid.

Operators are pursuing several responses. Some are seeking to secure dedicated sources of power, including arrangements to draw directly from specific generating facilities or to build their own. Interest in nuclear energy, including smaller reactor designs, has revived in part because it offers the constant, carbon-free output that large computing operations favor. Others are investing in efficiency, designing facilities and chips to accomplish more computation per unit of energy, though gains in efficiency have so far been outrun by growth in overall demand.

The situation carries implications for the broader energy transition. The surge in electricity demand from computing complicates efforts to reduce emissions, since meeting new load quickly can tempt operators and utilities toward whatever generation is readily available, which is not always the cleanest. At the same time, the deep pockets and steady demand of large technology firms have made them significant buyers of clean energy and, in some cases, catalysts for new generation that might not otherwise have been built. Whether the net effect accelerates or hinders the shift toward lower-carbon power is contested and likely to vary by region.

Policymakers and grid operators face difficult choices about how to allocate scarce capacity, how to share the costs of expanding the grid, and how to weigh the economic value of computing investment against its strain on shared infrastructure. The decisions involve trade-offs among competing users of electricity, among present and future needs, and among economic, environmental, and reliability goals that do not always align.

The deeper significance is that the digital economy, often imagined as weightless, rests on a very physical foundation, and that foundation is showing strain. The abundance of cheap, reliable power that the technology sector long took for granted can no longer be assumed, and the ability to build and operate the systems behind modern computing increasingly depends on solving an old and stubborn problem: generating and delivering electricity at the scale and speed that demand requires.