The queue is not a plan
Escherbridge
The grid story of the last two years is usually told as a generation story. Build more. The awkward part is the line to plug in, and the other line, of loads that show up bigger than the neighborhood was designed for.
The demand is no longer a rounding error
The IEA’s Energy and AI report puts electricity use from data centers at about 415 terawatt-hours in 2024, roughly 1.5% of global electricity, after growing about 12% a year for five years. In the same report’s base case, that reaches around 945 TWh by 2030, just under 3% of global electricity, growing around 15% a year — more than four times the growth rate of electricity use everywhere else. IEA, Energy demand from AI.
The IEA’s accompanying news note translates the 2030 figure: slightly more than the electricity consumption of Japan today, with AI-focused data centers the main driver. In the United States, data-center power consumption is on course to account for almost half of electricity-demand growth between now and 2030. In advanced economies more broadly, data centers are projected to drive more than 20% of demand growth over that stretch. IEA news release.
Queues contain wishes
The IEA’s Powering Artificial Intelligence summary describes connection queues, in places such as Texas, Virginia, Finland, and the United Kingdom, whose cumulative capacity is larger than today’s peak demand. It also says the quiet part planners already know: queues contain duplicative and speculative projects. Overestimate, and you build wires for ghosts. Underestimate, and the real projects wait. The same note says non-firm connections — access you can curtail when the system is tight — could enable an additional 750 to 900 gigawatts of grid hosting capacity globally, on the IEA’s analysis, because system stress is a handful of hours, not a lifestyle. Executive summary.
Renewables are in a queue of their own. The figure on our energy page is from the IEA’s 2023 grids report: at least 3,000 GW of renewable projects were waiting on a connection, 1,500 GW of them in advanced stages, equal to five times the solar and wind added in 2022. Queue data then covered countries with about half of global wind and solar capacity, so the true line is probably longer. A project with a 2030 date is not a power plant. It is a calendar entry. IEA, Electricity Grids and Secure Energy Transitions (2023).
Large loads do not sit still
NERC’s Level 2 alert of 9 September 2025 is about behavior, not branding. The alert describes bulk-power-system disturbances in which large loads dropped more than 1,000 MW without being asked, most of the events in 2024 or 2025. NERC’s definition of a large load includes data centers, crypto mines, hydrogen electrolyzers, manufacturing, and arc furnaces — anything whose size or habits can move the system. The recommendations are procedural and physical: interconnection requirements, modeling data, high-speed disturbance recorders. A slide that says “flexible demand” does not satisfy a transmission planner. A record of what the load did, when the voltage moved, does. NERC alert (PDF).
Local resilience is the unfashionable conclusion. If the interconnect is conditional, the site still has to know what it is drawing, what it can shed, and who is allowed to say so. That is coordination between a feeder, a building, and a person. It is not a second power plant.
The habit, and a better one
The habit is treating a queue position as a plan. A date in 2030 does not cool a building, and a speculative application does not train a model. The constructive version is smaller and more useful: publish which projects are real, connect what the wires can actually hold, and give each site a record of its own load — what it draws, what it can pause, and who is allowed to say so. Generation still matters. It is just not the part that is stuck.