Three independent supply curves (power availability, chip production, advanced packaging) projected forward from current baselines; the binding constraint at each quarter is whichever curve is lowest. Adjustable assumptions: PJM approval rate, construction delay factor, chip yield improvement, CoWoS capacity growth, GPU density per MW. Confidence bounds widen from 30% to 60% over the 5-year horizon (no projection beyond the cited papers' horizon).
How much AI compute will be available in the coming years, and what physical limit binds at each horizon? Power supply, chip manufacturing, and advanced packaging each set an independent ceiling; the lowest determines total available compute, and the binding constraint shifts over time as different bottlenecks resolve. Useful input for AI capability projection work and for sizing where regulatory leverage will land: controls aimed at a non-binding constraint won’t reduce supply; controls at the binding one will.