- The largest grid operator in the US just failed its own capacity auction for the first time in history.
PJM is 6,623 MW short. Texas has 233 GW in its interconnection queue but added only 23 GW of new generation.
The grid can't grow fast enough for AI. 🧵 - PJM's 2027/2028 auction cleared 134,479 MW of capacity.
That's 5.2% below the reliability target. Reserve margins dropped to 14.8% vs the 20% needed.
Capacity prices hit the FERC-approved cap of $333.44/MW-day across the entire footprint. - The root cause: a thermal retirement cliff.
PJM retired 54.2 GW of thermal capacity since 2011. Another 24-58 GW is at risk by 2030.
New entries can't keep pace. The auction cleared just 774 MW of new generation.
Here's the math that breaks most planning models: - Replacing 1 MW of retiring thermal requires ~5.2 MW of solar or ~14 MW of onshore wind to maintain equivalent reliability.
The grid isn't just losing capacity. It's losing dispatchability.
Meanwhile in Texas, ERCOT's large-load queue hit 233 GW — a 269% jump from 2024. - The entire ERCOT grid peaks at ~85 GW. The queue is nearly 3x total peak load.
77% of those requests? Data centers.
Much of that queue is "phantom" load — speculative requests from hyperscalers hedging across multiple sites. - It clogs engineering studies for projects that may never reach financial close. ERCOT hired McKinsey to overhaul the process.
The Texas Energy Fund set aside $9B for new dispatchable gas plants. - ~35% of proposed projects have already withdrawn. Global turbine shortages and permitting delays are real.
Financial incentives alone can't solve a physics problem.
We believe the answer is Deep AI — not LLM wrappers. - Physics-informed neural networks solve transient stability 87x faster than conventional methods. Graph neural networks predict substation failure with 0.89 F1 scores.
The grid is a graph. Treat it like one.
One immediate win: Dynamic Line Rating. - Traditional static ratings leave 20-40% of transmission capacity unused. AI-driven DLR unlocked 61% more capacity on 345 kV lines in Indiana/Ohio — at 76% lower cost than reconductoring.
No new lines needed. - The cost of inaction? One analysis projects $163B in cumulative capacity costs across PJM states through 2033.
In Northern Illinois alone: $21.4B — roughly $70/month more per household.
This isn't abstract. It hits ratepayers directly. - The grid can't be rebuilt fast enough for the AI revolution. But it can be made intelligent enough.
What do you think actually closes the gap first — new generation buildout, demand-side flexibility, or AI-optimized existing infrastructure?
#EnergyAI #GridReliability - We broke down the full technical architecture in our latest whitepaper →
https://veriprajna.com/whitepapers/sentinel-grid-pjm-shortfall-ercot-crisis-deep-ai