AI Demand Spurs US Gas Plants at Twice China's Pace

Artificial intelligence is driving a sharp increase in proposed gas-fired power plants across the United States, reversing a long-standing pattern in which China led global additions. Global Energy Monitor data show US gas capacity in development rose from 252 GW in January to 378 GW by mid-2026, with roughly half tied directly to data-center projects. That total now exceeds China's pipeline by a factor of two and accounts for about one-third of all such projects worldwide.

The scale of the buildout reflects concrete choices by hyperscalers. Amazon, Microsoft, Google, Meta and Oracle plan more than $750 billion in capital spending this year, with roughly three-quarters directed at AI infrastructure. Many developers are turning to behind-the-meter gas plants to avoid multi-year grid interconnection delays and to limit cost shifts onto existing ratepayers. Global Energy Monitor analyst Jenny Martos noted that the surge in data-center proposals has flipped the US-China construction ranking within six months.

State-level responses have varied. New York enacted a one-year moratorium on new hyperscale data centers in July. Texas Governor Greg Abbott ordered an audit of existing facilities before further approvals. Pennsylvania and Michigan have also tightened permitting or considered pauses. A Heatmap poll found three-quarters of Americans oppose living near a data center.

Federal policy is moving in the opposite direction. The Trump EPA has proposed ending mandatory public disclosure for minor-source air permits that cover diesel generators and gas turbines at many data centers. The agency stated the change would reduce administrative burden and accelerate economic development. Former EPA official Joe Goffman and attorneys at the Environmental Law and Policy Center warned that communities would lose access to emissions data needed for oversight or litigation.

Emissions and infrastructure questions remain unresolved. If all tracked US projects reach completion, gas capacity would rise by roughly two-thirds at a cost exceeding $647 billion. Global Energy Monitor estimates this could increase power-sector emissions by as much as 20 percent. Two-thirds of more than 800 planned data centers sit in drought-prone areas, though aggregate national water figures show data-center consumption remains small relative to agriculture. Utilities in PJM have proposed curtailing data centers first during peak stress.

China's data-center expansion relies more heavily on renewables, particularly solar and hydropower sited in rural regions with surplus generation, according to Brookings fellow Kyle Chan. The US lead in gas therefore reflects near-term speed and turbine availability rather than a uniform global reversal in fossil-fuel trends. Several projects face local opposition, financing uncertainty and equipment constraints, so final build rates are not assured.

The central tension is whether the speed required by AI developers can be reconciled with existing permitting, grid capacity and community consent processes. Bipartisan legislation has been introduced in Congress to address electricity costs and transparency, while companies such as Microsoft have announced voluntary commitments to cover incremental power costs and limit nondisclosure agreements with host communities.