PJM seeks 15 gigawatts of new generation capacity to meet AI data-center demand

PJM Interconnection is preparing an emergency expansion of energy supply: the operator wants to attract up to 15 gigawatts of new generation capacity to keep the grid in pace with demand from AI data centers. For the industry, this is yet another signal: the main deficit in the AI race is becoming not only GPUs, but also megawatts.

Why the urgency is needed

This is not a distant scenario, but a quite practical problem of the coming years. PJM manages the largest power grid in the United States, and in its region electricity demand is already growing sharply due to the construction of new data centers and expansion of existing facilities for AI workloads. If before the power grid operated in a mode of relatively predictable consumption, now large clients are arriving simultaneously, needing tens and hundreds of megawatts with virtually no tolerance for outages. For a grid operator, this is already a reliability issue, not just long-term planning.

PJM has previously warned of the risk of serious capacity deficits in the next decade. Against this backdrop, ordinary market mechanisms and standard connection queues seem insufficient: new projects are being built slowly, some old power stations are closing, and the load from computational infrastructure is growing faster than expected just a couple of years ago. That is why the operator is pushing an emergency option—not as a new norm, but as an attempt to close the immediate gap between AI growth and available electricity.

How they want to close the deficit

Based on published parameters, the operator wants to attract nearly 15 gigawatts of additional capacity and simultaneously is gathering information from the market about the conditions under which such capacity can realistically be brought into the system. Two basic scenarios are being discussed: direct bilateral contracts between future large consumers and energy suppliers, as well as centralized procurement through a special backstop procurement mechanism. The idea is to link new data centers not with an abstract promise of future energy, but with specific new or reactivated sources that can enter the market on the necessary timeline.

This approach has both political and tariff logic. Regulators and market participants have long debated who should pay for the sharp increase in consumption from data centers: all grid consumers or the technology companies themselves that are creating this additional demand. In PJM discussions, 15-year long-term contracts have already been mentioned, as well as separate requirements for those who want to quickly connect large AI facilities.

The logic is simple: if new load comes into the system, it should bring with it a new source of capacity, not shift the deficit risk to households and regular businesses.

Where the pressure is strongest

The PJM story shows that the bottleneck in AI is now far from just chips and servers. The constraint is becoming physical infrastructure: generation, transmission lines, approval timelines, connection queues, and the region's ability to quickly deploy new facilities. This is precisely why even strong demand from hyperscale companies does not guarantee instant data center launch—first you need to ensure actual power delivery to the grid.

• Load from new AI and cloud facilities is growing fastest.
• Connection queues for generation capacity are not keeping pace with demand.
• New power stations and upgrades to old ones require years, not months.
• The risk of capacity deficits already affects prices and connection conditions.

For the energy market, this is also a signal that winners will not only be suppliers of computational hardware, but also companies that can quickly build, restart, or contract reliable generation capacity. At the same time, the mechanism itself remains transitional: PJM emphasizes that this is a one-time emergency solution while the market and connection rules adapt to the new scale of demand. But the very fact of such a measure shows how rapidly AI load has transformed from a technology topic into a matter of basic infrastructure.

What this means

The AI boom is increasingly running up against energy constraints. If before, competitive advantage came from access to GPUs and capital for data center construction, now guaranteed electricity supply for years ahead is becoming the decisive factor. For the market, this means a new phase: winners will not only be model developers and cloud providers, but also regions where megawatts, networks, and permissions can be quickly delivered to the facility.