On July 21, 2025, Anthropic published a report calling for substantial investments in computing capacity and electric infrastructure so that leading AI models can be trained and deployed inside the United States. The company argues that as AI technology advances, the necessary resources, infrastructure, and regulatory conditions must adapt accordingly.
Anthropic notes that progress driven by U.S. firms is already reshaping the economy and society: its models are used by American businesses, scientists, educators, and governments. The report says decisive action now can help ensure future AI development and benefits occur in the United States.
Energy requirements for maintaining global AI leadership
The report emphasizes that frontier AI model training requires continuous access to firm, reliable power, and meeting that need will demand a broad mix of energy technologies. State‑of‑the‑art model development data centers need stable, uninterrupted power, and Anthropic recommends an "all of the above" energy approach to meet near‑term demand, expand firm power sources, and enable AI to catalyze emerging energy technologies such as next‑generation geothermal and advanced nuclear.
Anthropic warns the U.S. is falling behind in bringing new generation online while China has rapidly expanded energy capacity for AI: last year China added over 400 gigawatts (GW) of capacity versus only a few dozen GW added in the U.S. The company projects that building a single advanced AI model will require roughly 2 GW of data center capacity in 2027 and 5 GW in 2028. With several other U.S. companies developing frontier models, total frontier AI training demand in the U.S. is expected to reach 20–25 GW by 2028.
For context, Anthropic notes this is about twice New York City’s peak electricity demand. The report also estimates at least as much capacity will be required for inference (serving everyday AI model usage). Put together, the U.S. AI sector will need at least 50 GW of electric capacity by 2028 to sustain global leadership.
Regulatory and permitting barriers to accelerate
The report acknowledges that the Trump Administration has taken steps to remove some barriers, including setting ambitious nuclear goals and accelerating National Environmental Policy Act (NEPA) reviews, but says further action is necessary to compete with China. Regulatory challenges and permitting delays can slow energy projects; Anthropic proposes a set of policies to address these bottlenecks.
Key measures include making federal lands available for AI infrastructure to avoid lengthy state and local zoning processes; accelerating NEPA reviews by conducting advance assessments of certain projects; partnering with the private sector to expedite critical transmission builds and upgrades; and working with utilities to speed grid interconnection processes, including deploying AI to improve efficiency. Where national security requires, the report recommends using existing federal authorities to ensure timely interconnections for AI training infrastructure.
Two strategic pillars for American AI infrastructure
Anthropic frames its recommendations around two complementary pillars designed to cover both large‑scale training facilities and broader infrastructure supporting nationwide AI deployment.
Pillar 1 — Build large‑scale AI training infrastructure:
- Make federal lands available for AI infrastructure construction to reduce years‑long local zoning delays;
- Accelerate NEPA processes for AI infrastructure by conducting advance reviews of certain projects;
- Partner with private sector to speed critical power line buildouts and upgrades;
- Support utilities in reforms to accelerate interconnection processes, including deploying AI to boost efficiency; and where critical for national security, leverage federal authorities to secure timely interconnections.
Pillar 2 — Build broad‑based infrastructure for nationwide AI innovation:
- Accelerate geothermal, natural gas, and nuclear permitting;
- Establish National Interest Electric Transmission Corridors to speed transmission permitting in areas of significant AI data center growth;
- Strengthen domestic production of critical grid components and turbines through loan and loan guarantee programs and by building strategic reserves;
- Support training, apprenticeship, and entrepreneurship programs for critical energy workers such as electricians and construction workers.
Why this matters
Anthropic argues the United States has the economic capacity, technical expertise, and innovative culture to meet AI infrastructure challenges. The recommendations are presented as a practical blueprint that uses existing federal authorities to cut through regulatory barriers that have slowed energy development.
The report urges coordination among government, industry, and communities to build the infrastructure foundation that could support decades of American leadership in AI.
The full report, "Build AI in America," provides more detail on the proposals and related policy measures.
