EIA sees data centre servers taking up to a third of US commercial power by 2050

EIA sees data centre servers taking up to a third of US commercial power by 2050 The agency's Annual Energy Outlook puts server demand at 22% to 33% of commercial building electricity by 2050, up from 7% in 2025. The US Energy Information Administration projects that electricity used by data centre servers will climb to between 22% and 33% of commercial building power consumption by 2050, up from an estimated 7% in 2025.¹ That range is wide enough to reshape how the commercial sector plans its grid. Servers alone would draw between 446 billion and 818 billion kilowatt-hours a year by 2050, a load addition that rivals entire demand categories counted in the sector now.¹ The same outlook logs industrial natural gas consumption averaging a record 23.6 billion cubic feet per day in 2025, 1% above the prior record of 23.4 Bcf/d set in 2023.¹ Factories and server farms pull from different parts of the gas and power system, so the two records are not mechanically linked.¹ There is a measurement quirk worth flagging. The EIA counts data centre servers as a commercial-sector subcategory, so the 22% to 33% figure understates total facility load because it excludes cooling, lighting and other building systems.¹ In Texas the shift is already showing up in the generation stack. Solar output on the ERCOT grid is expected to reach 78 billion kilowatt-hours in 2026, overtaking coal at 60 BkWh.¹ Total US energy production hit a record 107 quadrillion British thermal units in 2025, up 3.4% from the 2024 record.¹ Record supply is meeting record industrial demand, which leaves the data centre build as the swing factor for the next two decades.¹ There is a cautionary precedent in Texas power. The roughly $45 billion leveraged buyout of TXU Corporation by KKR, TPG and Goldman Sachs Alternatives, a firm that by 2000 ranked as the world's fifth-largest energy company after its $10 billion purchase of Britain's Energy Group plc in 1998, collapsed into bankruptcy protection on 29 April 2014 as debt and falling power prices overwhelmed it.² The current bull case is the inverse, resting on load growth rather than financial engineering. But the TXU failure is a reminder that long-dated demand forecasts can attract capital that turns toxic if the load never materialises.² Australia is wrestling with the same load-versus-coal problem from the policy side. On 8 June (2026-06-08), Greenpeace urged climate and energy minister Chris Bowen to "lead with vision and ambition" as he travelled to chair interim Bonn climate talks.³ The unresolved risk is timing. The EIA's cases show server load rising steadily to mid-century, but interconnection delays, transformer shortages and permitting fights could push the build into a later surge or stretch it thin enough that power prices never break their current range.¹ Whether commercial customers begin locking in longer-dated power hedges as that demand firms up is the signal that will tell traders which path the forecast is taking.¹