AI load arrives in one of two Henry Hub regimes. The BTM pipeline sits between them.

AI load has made Henry Hub power burn a two-regime question through 2030. The hinge isn't gas supply — it's where the load clears, and on which side of the meter.

ERCOT's large-load queue moved from 63 GW at end-2024 to 226 GW a year later — roughly 75% data centers, with grid-realistic peak landing at 98–112 GW after the haircut. PJM's January 2026 Load Forecast applied data-center load adjustments to 14 of its zones; Dominion alone shows 4.1% annual summer peak growth over the next decade. AEO 2026 puts power-sector gas use rising from about 35 Bcf/d in 2025 to 38–50 Bcf/d by 2050 — the wide band itself signals the regime question.

In Regime 1, gas absorbs. Henry Hub stays roughly below $4. Coal retirements slip — industry tally puts 2025 at roughly 2.6 GW against an 8.5 GW plan, with about 3.2 GW held by federal emergency orders. FERC enables fast colocation; hyperscalers prioritize speed-to-power. Summer burn above 40 Bcf/d — NGSA already puts the 2026 outlook at 40.3 Bcf/d — becomes the baseline rather than the seasonal ceiling. Henry Hub trades with a higher structural floor: every incremental AI megawatt shows up as on-grid gas demand. Turbine availability back-loads the timeline; GE Vernova's order book stretches into 2029.

In Regime 2, the cap holds. Solar plus four-hour storage LCOE keeps falling. State-level cost-allocation rules and moratoria bite — the Ohio and Virginia debates lead. Hyperscaler PPA preferences shift back toward firm-clean; the 10-plus GW of recently-contracted nuclear deals signals the appetite. FERC capacity-market reform forces BTM to pay for grid services. Hub sees the peaks; the floor looks more like the current weather-and-storage trade because AI load is served by firm-clean or BTM options.

Sitting between the regimes is the behind-the-meter pipeline — gas-fired projects directly serving data centers off the grid. Dozens of gigawatts have been announced through 2030: Microsoft, Meta, the Chevron–Engine No. 1 joint venture, the Stargate consortium. Most of this load is scoped into utility IRPs as grid demand because the IRPs predate the BTM shift. FERC's December 2025 colocation order opened a multi-year transition that makes the optionality real. If BTM replaces on-grid gas, utility power-burn forecasts are over-stated. If it stacks on top, they're under-stated and the system is tighter than IRPs admit.

Which regime arrives depends on price and policy. Which corridors clear it depends on which gas-fired interconnects move and which BTM sites take the molecules off-grid.