Power demand keeps breaking records. The EIA's August Short-Term Energy Outlook (Aug 11) projects U.S. power consumption rising from a record 4,195 TWh in 2025 to 4,268 TWh in 2026 and 4,391 TWh in 2027 — a third consecutive record, driven by AI data centers and electrification. Power-sector natural gas burn is projected around 36.6 Bcf/d in 2026.
The demand pipeline is enormous — and split down the middle. East Daley Analytics tallies roughly 32 Bcf/d of announced incremental U.S. gas demand by 2031, with hyperscaler data-center campuses and LNG feedgas each accounting for about 16 Bcf/d. LNG dominates 2026 growth; data centers catch up toward the end of the decade. Regionally, ArkLaTex leads at a net 12.8 Bcf/d (Sabine Pass, Cameron ramps), with the Houston Ship Channel, PJM's data-center corridor and the Permian behind it. Williams responded by raising its long-term outlook, backed by Transco expansions and a $5.5 billion Haynesville push.
Texas hit pause. On August 3, Texas ordered a halt to new data-center interconnections pending project-by-project review — enough to make the EIA trim its 2027 Texas demand forecast. The signal: grid interconnection is now a rationed resource, which pushes even more projects behind the meter.
The behind-the-meter wave is no longer theoretical. Recent weeks brought NRG's 1.2 GW combined-cycle plant deal serving a Texas data center, NextEra securing $3.3 billion in state funding toward 10 GW of gas capacity across Texas and Pennsylvania, and Amazon advancing a 7.65 GW gas-fired program in Texas. Industry reporting now describes gas turbines and reciprocating engines as having shifted "from backup role to primary baseload solution" for on-site power — with supply chain, not technology, as the binding constraint. Some critical-component lead times are reported as long as 210 weeks.
Put the three threads together and the picture is consistent: demand is setting records, the equipment to serve it is rationed, and the grid connection itself is becoming a queue. Every actor in the chain — utilities, IPPs, hyperscalers, midstream — is being pushed toward the same conclusion: maximum value must be extracted from machines that are already installed, already connected, and already burning gas.
Note also what the behind-the-meter wave implies about where the new machines will run: fast-tracked sites, non-ideal climates, 24/7 duty for load that tolerates no derates, and fuel supplies ranging from rich shale gas to whatever the local system delivers. These are exactly the conditions where fleet-average factory calibrations give up the most.
When power demand sets records and turbine lead times stretch to 2031, one absolute point of combustion efficiency stops being a fuel-cost line and becomes found capacity — roughly 3% more output from the same fuel, on machines already connected. A fleet-wide calibration review delivers megawatts on a timeline no OEM can quote: weeks, not years. That is the entire premise of our turbine practice — and this week's numbers made the case louder.
| Signal | Number | Operator takeaway |
|---|---|---|
| EIA 2026/2027 power demand | 4,268 / 4,391 TWh | Installed fleets run harder; heat rate compounds faster |
| Announced gas demand growth to 2031 | ~32 Bcf/d | Fuel competition rising — efficiency is supply insurance |
| Power-sector gas burn 2026 | ~36.6 Bcf/d | Every % of heat rate is national-scale fuel |
| BTM programs announced (NRG, NextEra, Amazon) | ~19 GW combined | New machines, hard sites, generic calibrations arriving |
| Critical component lead times | up to 210 weeks | Recovered output is the only fast capacity |
Sources: U.S. EIA Short-Term Energy Outlook (Aug 11, 2026) via Reuters; East Daley Analytics; Natural Gas Intelligence; Utility Dive; industry press — week of Aug 18, 2026.
Send us a fuel analysis and a month of operating data — the first review is free, and the findings are yours either way.
Start with a Free Fuel-Data Review