Factchecking with the companies before publishing my own article. This backgrounder is based on Zeke Hausfather’s article, The Climate Brink, Oct 26, 2026. Read it here He’s a respected scientist but I must doublecheck Dave
Reciprocating engines and simple-cycle turbines — the two technologies data centers are actually buying to skip the grid — emit 63 to 80% more climate pollution per megawatt-hour than the grid itself, by Hausfather’s own admittedly generous accounting.
The comparison covers four technologies planned for the roughly 27 GW of off-grid data-center gas that Cleanview’s tracker can currently attribute to specific equipment: simple-cycle turbines (46% of that capacity), reciprocating engines (34%), fuel cells (14%), and on-site combined-cycle plants (effectively none). Hausfather built the grid-average side from 2023 EPA eGRID data for ERCOT, PJM, and the full U.S. — which already bakes in the grid’s own upstream leaks and transmission losses — then layered on EPA’s own reporting factor for engine methane slip (2.8% of fuel passing through unburned, versus roughly 0.02% for modern turbines) and Alvarez et al. 2018’s 2.3% national estimate for upstream gas leakage, applied evenly to both sides of the comparison.
Reciprocating engines come out around 740 kg of CO2-equivalent per megawatt-hour and simple-cycle turbines around 705 — 63 to 80% above the three grid benchmarks. Methane alone adds 70 to 190 kg/MWh to that gap; strip it out entirely and engines and turbines are still 31 to 63% dirtier than the grid on CO2 alone, simply from running less efficiently. Fuel cells (mostly Bloom Energy’s) and on-site combined-cycle plants — the options almost nobody is actually building — land only 7 to 10% above the U.S. average.
Location moves the number, but not the conclusion
Hausfather reran the comparison for seven data-center hotspots using 2025 regional grid data and basin-specific leakage rates measured by aircraft and satellite surveys (MacKay et al. 2026, Varon et al. 2026). Off-grid engines and turbines beat the local grid in six of seven regions, by 36 to 95%. The exception is Cheyenne, Wyoming, where coal still supplies nearly 60% of the grid — there, off-grid gas is merely comparable to the grid average, though still 55 to 64% dirtier than a new grid-connected gas plant would be.
The risk isn’t just today’s number
The piece doubles as a gut-check on a bigger number making the rounds: a Latitude Media podcast estimate that 50 GW of new off-grid gas could add 380 million tons of CO2-equivalent in 18 months. Hausfather doesn’t dispute the direction but notes the figure assumes all 50 GW is already running, which it isn’t. His own flagged risk runs the other way: grid-connected gas can throttle down as cheap renewables come online; islanded generators mostly can’t, so today’s gap could widen as the rest of the grid decarbonizes around them.
His proposed fix is narrow rather than sweeping: connect data centers to the grid instead of building private gas plants, and where that’s not possible near-term, hold them to existing air-pollution rules and pair them with on-site solar and storage. He notes, in passing, that federal efforts to cut the methane leakage baked into his own numbers have recently been rolled back.
authored by Dave Burstein with close collaboration by Claude
A reader’s-note caveat: the 46/34/14% technology-mix breakdown comes from a commercial tracker (Cleanview) that only has equipment identified for about 30% of planned off-grid capacity — Hausfather flags this himself as directional, not exact.