Build AI in America: Infrastructure Policy Guide

Why AI Infrastructure Is a National Security Imperative

The battle for AI leadership isn't being fought in boardrooms or research labs—it's being waged in the foundational infrastructure that powers computational supremacy. Anthropic's comprehensive policy blueprint, "Build AI in America," makes a stark case: without massive infrastructure buildout by 2028, the United States risks ceding AI leadership to China and compromising national security.

The report identifies AI infrastructure as a dual-pillar challenge requiring both large-scale training facilities and broad-based inference deployment. Unlike previous technology transitions, AI demands energy-intensive infrastructure at unprecedented scale, making traditional development timelines incompatible with strategic imperatives.

Current US electricity demand has grown less than 1% annually for two decades. AI is changing that trajectory dramatically, requiring growth several times faster. This isn't merely an economic opportunity—it's an existential question of technological sovereignty in an era where AI capabilities increasingly determine national power.

The Staggering Energy Demands of Frontier AI

The numbers are staggering: a single frontier AI training run will require a 5 GW data center by 2028—equivalent to America's largest nuclear plants. Training alone demands 20-25 GW across multiple locations, while inference deployment requires roughly equal or greater capacity nationwide.

Multiple forecasting models converge on similar projections. Semianalysis projects 80 GW globally for AI critical IT by 2028, with 56 GW in the United States. RAND estimates 117 GW globally, while DOE's Lawrence Berkeley National Lab projects US data centers broadly at 74-132 GW in 2028.

These energy demands occur against a backdrop of constrained grid capacity. The US added roughly 55 miles of high-voltage transmission lines in 2023, compared to a 1,700-mile annual average in 2010-2014. Generation interconnection approvals typically require 4-6 years, while transmission projects average over 10 years to complete since 2005.

Training vs. Inference: Two Distinct Infrastructure Challenges

Understanding AI infrastructure requires distinguishing between training and inference workloads. Training develops AI models through intensive computational processes requiring massive, concentrated energy at specific locations. Inference deploys trained models for real-world applications, requiring distributed capacity nationwide.

Training infrastructure can be geographically concentrated at a few optimal sites with abundant energy and favorable regulatory environments. These facilities need reliable baseload power, sophisticated cooling systems, and proximity to transmission infrastructure capable of handling multi-gigawatt loads.

Inference infrastructure must be distributed to serve end-users with low latency. This requires broad-based energy and data center capacity across metropolitan areas, creating different regulatory and infrastructure challenges than concentrated training facilities.

The Regulatory Gauntlet: Permitting and Transmission Delays

America's regulatory framework, designed for incremental infrastructure development, creates bottlenecks incompatible with AI infrastructure timelines. Environmental reviews, interconnection studies, and transmission approvals often require years-long processes that would push AI infrastructure completion well past 2028 deadlines.

The National Environmental Policy Act (NEPA) requires comprehensive environmental impact assessments for federal actions. While necessary for environmental protection, current NEPA processes can extend project timelines by years. Transmission interconnection queues, managed by regional grid operators, face similar delays as utilities struggle with unprecedented demand volumes.

State and local permitting adds additional complexity. Zoning approvals, construction permits, and environmental compliance at state levels create multiple veto points where projects can stall. These distributed approval processes, reasonable for traditional development, become strategic vulnerabilities when speed is essential for national competitiveness.

How China's Infrastructure Speed Creates Competitive Threats

China's infrastructure development speed provides sobering context for American challenges. China brought over 400 GW of new generation capacity online in 2024—roughly ten times US additions excluding storage. Chinese construction permitting requires 3-6 months versus years in America.

China's Eastern Data Western Computing Plan has invested $6.1 billion in data center hubs over two years, demonstrating coordinated state capability to rapidly deploy digital infrastructure. This centralized approach enables China to make strategic infrastructure decisions and implement them at unprecedented speed.

The competitive implications extend beyond mere capacity metrics. Rapid infrastructure deployment enables China to attract AI companies, researchers, and investment that might otherwise flow to the United States. As AI development becomes increasingly capital and energy-intensive, infrastructure constraints risk offshoring American AI leadership.

Leveraging Federal Lands to Bypass State and Local Bottlenecks

Federal lands offer a strategic pathway to accelerate AI infrastructure development by bypassing lengthy state and local approval processes. The Bureau of Land Management (BLM) manages over 30 million acres in western states where extensive environmental analyses have already been completed for energy development.

The strategy involves leasing Department of Defense and Department of Energy lands for data center construction near BLM lands available for power procurement. This approach leverages existing environmental clearances while maintaining federal control over approval timelines.

Western federal lands offer particular advantages: proximity to renewable energy resources, existing transmission infrastructure within 15 miles of many sites, and approximately 40 GW of untapped hydrothermal geothermal potential. Federal control eliminates state and local zoning constraints that can delay projects for years.

Accelerating Environmental Reviews and Permitting

Smart permitting reform can dramatically reduce approval timelines without compromising environmental protection. The report outlines several executive actions requiring no Congressional approval that could streamline NEPA processes for AI infrastructure.

Programmatic NEPA reviews allow agencies to conduct broad environmental analyses before specific projects are proposed. By completing programmatic reviews of AI data centers' environmental impacts immediately, agencies can pre-clear requirements and reduce site-specific review timelines.

Categorical exclusions represent another powerful tool. The Air Force has developed categorical exclusions for projects similar to those with established insignificant environmental impact. Expanding these exclusions to AI infrastructure projects with proven low impact could eliminate lengthy review processes entirely.

Tiering allows new projects to reference existing environmental documents rather than starting fresh analyses. The Naval Air Station Lemoore data center analysis and BLM's solar and geothermal programmatic Environmental Impact Statements provide precedents for AI infrastructure projects to tier off established environmental clearances.

Unlocking Transmission: DOE Authorities and Federal Siting

Transmission bottlenecks require federal intervention to meet AI infrastructure timelines. The Department of Energy possesses partnership authorities under Sections 1222 and 40106 that can approve transmission lines federally, bypassing lengthy state Public Utility Commission processes.

Federal transmission siting offers multiple advantages: faster approval timelines, ability to site on federal lands to avoid state and local review, and allocation of transmission costs to developers rather than ratepayers. DOE's $5.75 billion in combined credit lines provides financing capability to support rapid transmission buildout.

The report proposes launching public DOE solicitations for transmission line proposals serving AI training infrastructure. This market-based approach would attract private capital while maintaining federal oversight of strategic infrastructure development.

Reforming Grid Interconnection for 2028 Deadlines

Grid interconnection reform represents perhaps the most complex challenge in AI infrastructure development. Current interconnection processes, designed for traditional generation resources, cannot accommodate the scale and urgency of AI infrastructure demands.

The report proposes several interconnection reforms: conditional interconnection agreements that limit peak-day consumption, AI-powered resilience testing to optimize grid integration, queue auctions to prioritize projects with firm development timelines, and national security priority lanes for critical AI infrastructure.

As a last resort, the Defense Production Act Title I could require timely interconnections for frontier AI training infrastructure critical to national defense. Section 202(c) of the Federal Power Act provides emergency authorities for temporary interconnection requirements, renewable every 90 days during national emergency conditions.

Enabling Broad-Based AI Infrastructure Nationwide

Beyond concentrated training facilities, AI leadership requires broad-based inference infrastructure deployed across metropolitan areas nationwide. This distributed deployment faces different but equally significant regulatory challenges.

Geothermal energy development on federal lands represents a significant opportunity for distributed AI infrastructure. Programmatic environmental reviews, categorical exclusions, and elimination of double NEPA requirements could accelerate geothermal permitting significantly.

Nuclear power provides baseload capacity essential for AI workloads. The ADVANCE Act and related executive orders have begun improving nuclear permitting, but additional reforms could enable the Trump administration's target of 300 GW of new nuclear capacity by 2050 with 18-month NRC application processing.

Implementation Strategy: Financing, Workforce, and Timeline

Infrastructure development requires coordinated attention to financing, supply chain resilience, and workforce development. Current lead times for domestic transformers and circuit breakers run approximately three years, creating potential bottlenecks for rapid infrastructure deployment.

The report proposes building strategic reserves of critical grid components through guaranteed price-floor purchases. This approach would support domestic manufacturing capacity while ensuring component availability for urgent infrastructure projects.

Workforce development requires expanded apprenticeship programs for electricians, mechanical engineers, and construction workers. Partnerships with community colleges and universities can create training programs specifically tailored to AI infrastructure construction requirements.

Cybersecurity requirements for imported energy products represent another critical consideration. FCC standards or International Emergency Economic Powers Act import controls could ensure supply chain security while supporting domestic manufacturing capacity.

Executive Roadmap: From Policy to Operations by 2028

The report's most striking feature is its comprehensive implementation timeline mapping specific milestones from July 2025 through early 2028. Preconstruction approvals must be cleared by 2026 to enable 2028 operations, given 18-24 month construction timelines for frontier data centers.

The timeline coordinates multiple parallel workstreams: data center construction, solar and natural gas development, geothermal deployment, and transmission infrastructure. This coordination is essential because delays in any component can prevent entire projects from meeting 2028 operational deadlines.

All proposed actions fall within existing executive branch authority, requiring no new Congressional legislation. However, the report acknowledges that some options are operationally or politically difficult, requiring sustained executive commitment and interagency coordination.

The ultimate message is clear: failure to act risks offshoring AI infrastructure and jobs, ceding technological leadership to China, and compromising national security. The tools exist within current executive authority—the question is whether America will use them decisively to maintain AI leadership in the coming decade.