Here's where I landed... My supply-side screen points to 17-29 GW of new U.S. data-center facility load you could theoretically 'absorb' in 2026 and 12-30 GW in a separate 2027 snapshot (mid: 20.5GW 2026 / 16.0 GW 2027). These figures include cooling and other site overhead. Range is huge but there's a lot of moving parts. End of the day it'll come down to how much you believe BTM and FTM projects coming online on time. If storage or DR helps in a meaningful way or you credit some of these smaller markets I know less about, there's plenty of 'capacity' you can tap into. Reminder, NERC is inherently conservative, too so headroom could be bigger. Methodology: I start with NERC’s planned supply after forecast demand and reliability reserves, fully count 'core' markets, partially count less transparent Southeast and Western regions, and give California zero credit b/c CAISO's 'headroom' figure seemed too high and given the politics I doubt CA will want to absorb much. I then add usable onsite power tracked by Aterio; sellside and SA are much more bullish on BTM but I am more cautious until we get evidence it'll work. This is more art than science and you could totally swing around assumptions. EIA’s +8.9 average GW for 2026 and +13.8 GW for 2027 are cross-checks because the planned supply largely overlaps with NERC. Batteries and demand response shift or reduce peaks, so they stay outside the headline adds. 2027 optically falls because demand uses headroom faster than new supply replaces it. This is a national ceiling before local grid, interconnection and construction constraints.