Anthropic projects the US AI sector requires at least 50 gigawatts of electric capacity by 2028 to maintain global competitiveness, according to a new policy report "Build AI in America" released by the company.
The AI company estimates its own advanced model development will demand 2GW and 5GW data centres in 2027 and 2028 respectively. Combined with other US frontier AI companies, total training demands could reach 20-25GW by 2028 - roughly twice New York City's peak electricity demand. An equivalent amount will likely be needed for AI model inference operations.
Anthropic warns that China added over 400 gigawatts of power capacity last year compared to "just several dozen gigawatts" added in the US, creating concerning infrastructure gaps for AI development. The company states that frontier AI model training requires "continuous access to firm, reliable power sources" and recommends an "all of the above" energy approach including next-generation geothermal and advanced nuclear technologies.
The report proposes two strategic infrastructure pillars. The first focuses on large-scale AI training facilities through federal land availability for construction, accelerated National Environmental Policy Act reviews, private sector partnerships for power line buildouts, and utility interconnection reforms. The second pillar targets nationwide AI deployment infrastructure via accelerated permitting for geothermal, natural gas and nuclear projects, National Interest Electric Transmission Corridors, domestic production strengthening for critical grid components, and expanded technical training programmes.
Anthropic notes that while the Trump Administration has set nuclear power targets and accelerated NEPA reviews, further regulatory action is needed to compete with China's rapid energy infrastructure development.
The company's AI models are currently used by American businesses, scientists, educators and government entities for productivity improvements, economic growth and national security applications. The proposed infrastructure investments would support broad AI deployment across multiple business sectors.
The substantial power capacity requirements highlight the scale of infrastructure investment needed for US AI competitiveness. The report emphasises regulatory reform and federal coordination as critical success factors, with existing federal authorities proposed as implementation mechanisms while addressing energy development barriers.