On March 4, representatives from seven major technology companies, including Microsoft, Google, OpenAI, Amazon, Meta, xAI, and Oracle, visited the White House and signed a written commitment to self-supply or purchase the electricity required for their AI data centers. The rapid growth of data centers has driven up U.S. electricity prices, posing a risk for the Republican party in the midterm elections. These tech giants have also signed commitments with President Trump at the White House to build or procure their own power supply for data centers.
U.S. data center energy policy faces three major constraints: technical feasibility, economic rationality, and political acceptability. In terms of economic rationality, hyperscale data centers encounter two primary challenges when addressing power supply issues: rising financing costs and increased risk of stranded assets. Regarding political acceptability, energy investments for data centers generate negative externalities. The associated project risks are not only borne by the data companies but also impact external stakeholders such as utility companies, other consumers, local communities, and overall grid reliability. From a technical feasibility perspective, the most significant practical constraint on U.S. electricity growth is not power generation technology itself, but a speed mismatch. New power generation can take up to eight years to come online, while capacity auctions secure future supply on a two-to-three year forward basis, creating a structural gap.
Grid bottlenecks are accelerating behind-the-meter generation, with natural gas as the primary energy source due to its lower cost. According to estimates from Bloom Energy, by 2030, 38% of data centers will use on-site generation as their primary power source, with 27% being fully powered by on-site systems. The main energy source for behind-the-meter generation is natural gas. Since 2024, orders for large U.S. gas turbines have increased significantly, reaching 14 GW, the highest level in nearly two decades. This trend accelerated in 2025, with orders reaching 18 GW in the first half of the year. Currently, the cost of generating electricity with natural gas is three times lower than with small modular reactors (SMRs), though SMRs hold greater long-term potential.
Looking ahead, the ability of the power system to respond to data center demand will largely depend on the policy environment and the energy procurement targets of the data centers. According to calculations by the Electric Power Research Institute (EPRI), under current state and federal policies (the Reference policy scenario), natural gas is likely to dominate new supply in the short term. If all data centers adopt and achieve 24/7 carbon-free energy (CFE) targets, the energy mix will shift towards low-emission generation and energy storage. Under the Reference policy scenario, annual新增 installed capacity for natural gas is projected to be between 6.6 GW and 13.7 GW from 2025 to 2030, higher than the average of 5.7 GW per year over the past five years. Under a 24/7 clean energy target, the generation and capacity response to increased DC load would primarily come from wind, solar, nuclear, and energy storage.
Insufficient U.S. power supply poses potential risks. The tightening U.S. electricity market, driven by surging demand from the data center boom, coupled with coal plant retirements outpacing the deployment of renewables and natural gas generation, risks price increases, blackouts, and bottlenecks hindering AI development. Against this backdrop, the power supply chain is also being incorporated into AI infrastructure investment, presenting opportunities in an infrastructure "reliability cycle." Within the power theme, investment in grid equipment and grid-related engineering is recommended. Generation-side assets, fuel and electricity price-related assets, energy storage, and power electronics are also themes that can be allocated tactically.
During the week of March 2-6, the Trump administration issued three executive actions focusing on personnel appointments and the economic and financial sectors. Geopolitical risks exceed expectations.
Comments