The AI power story is no longer theoretical. PJM is fast-tracking 11.8 GW of new resources. Texas is staring at 410 GW of load applications, about five times its current peak demand. And the American Capacity Project is trying to put a human number on what happens when states and localities say no to the infrastructure needed to serve new load.

That is the real story underneath the week’s grid news. Data centers are not just testing interconnection queues. They are exposing a basic governing problem: America has become very good at announcing demand and very bad at permitting the capacity needed to meet it.

PJM approved 51 projects for fast-track interconnection, totaling 11.8 GW of new capacity. The mix tells you almost everything about the moment: 69% gas, 19% battery storage, and 12% nuclear upgrades. The projects were selected from a much larger pool of applications and screened for readiness, site control, and the ability to actually get built rather than simply decorate the queue.

The driver is not subtle. Data centers have become the largest source of new load growth in PJM territory, with the RTO expecting roughly 30 GW of incremental demand from the sector by 2030. The normal interconnection queue is too slow for that world. PJM’s backlog still contains more than 2,000 projects, many of which will not be studied, financed, permitted, or built on a timeline that bears any relationship to the load now showing up.

Fast-tracking 11.8 GW does not fix the queue. It does, however, mark a useful change in posture. PJM is no longer pretending the old process can absorb the AI load boom with a few stakeholder meetings and some polite revisions to the spreadsheet.

GridTake - The gas-heavy mix will irritate the usual people for the usual reasons. But PJM is not writing a climate plan here. It is making a reliability call under ugly constraints. If the demand is real and the grid is tight, the choice is not between gas and a frictionless clean-energy buildout. It is between dispatchable capacity, delayed campuses, or a system that keeps making promises it cannot electrify.

The American Capacity Project’s new white paper, Capacity Delayed, Cures Deferred, argues that data center moratoria and electric-capacity delays are not just local land-use fights. They are constraints on the compute infrastructure now powering AI-enabled drug discovery, a pipeline that has doubled every 1.6 years for nine straight years and now includes more than 200 AI-originated drug candidates in clinical development. ACP estimates that moratoria, permitting delays, tax reversals, and interconnection barriers blocked or delayed more than $156 billion in planned compute infrastructure in 2025 alone.

The paper’s headline estimate: those delays destroyed or deferred about 18,480 megawatt-years of AI-capable compute and could defer 1.70 million healthy life-years over the next decade if the AI drug pipeline keeps growing at its recent pace. The argument is not that every data center cures cancer. It is that policymakers weighing a “temporary pause” on data centers are also, whether they know it or not, slowing the infrastructure behind faster drug discovery, clinical modeling, and medical innovation.

GridTake - This is the right argument because it makes “pause” defend itself as a policy choice, not hide behind the soft lighting of process. The model will get challenged, and it should, but ACP is putting the missing cost in the room: the cure that arrives late because America treated the infrastructure behind discovery as optional clutter.

ERCOT is now sitting on roughly 410 GW of load interconnection applications. Texas currently peaks around 85 GW. About 90% of the application volume is tied to data centers, much of it filed since early 2023. That does not mean Texas is about to add five grids’ worth of demand. It means the queue has become a speculative instrument.

A meaningful share of those applications are not mature projects. Some have not secured land. Some have not solved water. Some may be placeholders filed by developers who want optionality before they have certainty. Texas has limited penalties for queue withdrawal, so applications pile up because there is little reason not to file.

The real number is almost certainly much smaller, but still enormous. If only 10% of the queue materializes, ERCOT is dealing with a load shock larger than many states’ entire power systems. The Texas Energy Fund is helping bring new dispatchable capacity into the pipeline, and ERCOT has added new reliability tools, but neither solves the basic problem: the market is producing demand signals faster than the system can sort real projects from vaporware.

GridTake - The 410 GW number is less a forecast than a confession. It tells us the queue is no longer a planning instrument. It is a waiting room where serious projects, speculative filings, and PowerPoint empires all sit in the same chairs. Texas does not need to study every request as though it is real. It needs a way to make developers prove seriousness before they consume grid-planning oxygen.

PJM is also floating three broad approaches to capacity market reform, and none of them are painless. One option would patch the current Reliability Pricing Model with changes to demand curves and offer rules. Another would allow prices to better reflect scarcity, which is economist-speak for “stop pretending reliability is cheap when it is not.” A third would move away from the standalone capacity market and toward a model that relies more heavily on energy and ancillary services pricing.

This is not an academic debate. PJM’s capacity market clears billions of dollars across a 13-state footprint and helps determine which plants stay open, which projects get financed, and how much consumers pay for reliability. The market was designed for a more stable demand world. That world is gone. Coal retirements, electrification, industrial load growth, and data centers have turned capacity planning into a live-fire exercise.

GridTake - PJM will probably choose the least disruptive reform it can defend, then spend two years arguing over the details. That is how these things go. But the deeper problem remains: a capacity market that pays for theoretical reliability while the real grid gets tighter is not a market. It is a very expensive reassurance ritual.

Australian developer Stockland has filed applications for a 250 MW data center campus in Melbourne’s outer suburban corridor, with operations targeted in phases beginning in 2027. The project fits a pattern now visible across the Anglophone grid map. Primary data center markets are saturated or politically constrained, so developers push into secondary cities where land may be easier but transmission is not.

Melbourne is not Northern Virginia. But the infrastructure math rhymes. Large loads arrive faster than grid upgrades. Developers can announce campuses in months. Transmission planning takes years. Permitting can take longer. The gap between those timelines is where most of the political pain now lives.

GridTake - Every region thinks it has more time than it does. Then the applications arrive, the load forecasts jump, and the grid operator starts using phrases like “reinforcement needs.” Melbourne is just another reminder that the data center buildout is not a U.S. story. It is a global capacity test.

Upgrade to Grid Brief Premium to get extra deep dives into energy issues all over the world.

"The queue itself is the problem. You can fast-track 51 projects and still have 2,000 more sitting there. At some point you have to redesign the process, not just run it faster." — PJM stakeholder comment, queue reform docket, April 2025

We rely on word of mouth to grow. If you're enjoying this, don't forget to forward Grid Brief to your friends and ask them to subscribe!