The mechanism by which large electricity consumers contract for power has been one of the quieter forces shaping the energy transition, and it is now being reshaped in turn by the appetite of artificial intelligence workloads for sustained, large blocks of electricity. Power purchase agreements between developers of generation and the firms that consume their output have long been a workhorse of project finance, providing the predictable revenue streams that lenders require to fund construction. The terms of those agreements, the duration of the commitments, and the kinds of generation they support are all being adjusted to fit a customer base whose load profile and growth trajectory differ markedly from the corporate buyers who came before.

Operators of large data centers, whose facilities now constitute a significant and rapidly growing share of industrial electricity demand, have become the most visible counterparties in the new wave of contracts. Their interest is not in occasional power for a campus of office buildings but in continuous, high-density loads that run twenty-four hours a day at scales measured in hundreds of megawatts per site. Securing that power requires not only enough generation to meet the load but also the transmission capacity to deliver it and the dispatch flexibility to handle moments when intermittent sources are unavailable. The contracts have grown longer, larger, and more complex to reflect these requirements.

The shift has changed the kinds of generation that get built. Where corporate buyers in earlier years used power purchase agreements primarily to procure renewable energy and meet sustainability commitments, the new wave of agreements is more agnostic about source, prioritizing reliability and round-the-clock availability over particular fuel mixes. Some buyers continue to seek renewable supply but pair it with firming arrangements — battery storage, gas-fired backup, or contracts that effectively guarantee equivalent generation when their preferred source is not available. Others have signaled willingness to underwrite nuclear capacity, including small-scale designs and the restart of facilities previously slated for retirement, as the most direct route to the kind of continuous, low-emission power they need.

The economic consequences of the demand surge are showing up across the electricity system. Wholesale prices in regions hosting large data center buildouts have firmed, particularly during periods when generation is tight or transmission is constrained. Capacity markets that pay generators to remain available have responded with higher clearing prices, signaling the value of generation that can dispatch reliably. Interconnection queues, already long before the surge, have stretched further as developers race to bring new projects online to meet demand that is growing faster than the system can comfortably accommodate.

The interaction with grid planning has become a central challenge for utilities and grid operators. A traditional electricity system was designed around relatively predictable demand growth, with new generation and transmission planned years in advance to meet incremental needs. The arrival of multi-gigawatt clusters of demand in particular locations within compressed timeframes strains that model, forcing accelerated planning processes and difficult choices about where to build, what to build, and how to allocate the costs. Decisions about transmission corridors and substation upgrades that once unfolded over decades are now being compressed into years.

The competition for sites and power has produced a geographic redistribution of where data center capacity locates. Regions with abundant land, available transmission, and supportive permitting have attracted disproportionate investment, while regions where the grid is constrained have seen projects deferred or relocated. The pattern has implications for local economies, for the labor markets that staff the facilities, and for the tax bases of jurisdictions that win or lose the investments. It also has implications for the operators of generation in those regions, whose contracts and pricing reflect the local balance of supply and demand.

The longer-term question is whether the surge in demand from AI workloads will sustain, plateau, or eventually moderate as efficiency improvements in models and hardware reduce the energy intensity of computation. The history of computing suggests that efficiency gains tend to be absorbed by expanded usage rather than translated into lower aggregate demand, a pattern that would point toward continued growth. Whether the supply of generation, transmission, and capable workforce can keep pace is one of the central questions facing both the energy and the technology industries, and the answer will shape the trajectory of both.

What the contracts being signed today reveal, in their length, scale, and structure, is a quiet but consequential repricing of electricity as an input. The era in which power was treated as a commodity available at modest cost is giving way to one in which securing reliable supply at scale is a strategic problem, and in which the firms that need that supply are willing to commit to long-dated, multi-billion-dollar arrangements to obtain it. The grid is responding, but slowly, and the gap between what the new customers want and what the system can provide will shape both for years to come.