One year after its launch, America’s AI Action Plan has prompted widespread federal administrative activity, with a majority of its 103 recommendations seeing progress, though tangible delivery in infrastructure, energy, procurement, and international exports remains ongoing.
America’s AI Action Plan, launched by the White House in July 2025, was not an exercise in vague future modelling. It set a series of near-term policy goals that the federal government could conceivably execute. The paper presented the judgment that “America is in a race to achieve global dominance in artificial intelligence,” and declared, “This Action Plan is America’s roadmap to win the race.” Altogether, the plan put forward more than one hundred recommended actions across three pillars: Innovation, Infrastructure, and Diplomacy & Security.
Since then, agencies have responded with a litany of requests for information, model evaluations, program updates, new standards, grant opportunities, procurement vehicles, and executive actions. As of August 27, 2026, the 103 recommendations of America’s AI Action Plan are in varying states of progress.
As of August 27, 2026
The administration has made significant progress toward implementing the plan’s recommendations. By even a strict measurement that requires public verification, a majority of the plan’s recommendations have made progress.
The plan rightly considers the AI ecosystem as a whole: not just AI models, but everything from energy to infrastructure to applications. The plan’s strategic premise is simple: the country with the largest AI ecosystem will set global standards and capture economic and security benefits for decades to come. President Trump described technological dominance as a national security imperative. The authors identified three immediate roles for government in that mission: accelerate innovation, build infrastructure, and make American technology the foundation of global AI deployment.
The plan is a full-stack industrial and diplomatic strategy. In remarks at the 2025 APEC Digital and AI Ministerial Meeting, White House Office of Science and Technology Policy (OSTP) Director Michael Kratsios called the United States the prospective “global partner of choice,” and said he had come to discuss technologies ready for application, not merely their future potential. He argued that the real AI revolution would come from what people do with frontier systems, and connected open-source and open-weight AI to sovereignty, privacy, and customization. In January 2026, Kratsios testified to the House Science Committee that the White House was coordinating implementation across the federal government and distilled the plan down to three priorities: removing barriers to innovation, securing energy dominance, and exporting American technology.
The plan relates to the other major administration initiatives and communications on AI, including the Executive Order on Promoting the Export of the American AI Technology Stack, Pax Silica, the OSTP report “Science: A New Golden Age,” and the National Security Science and Technology Strategy. In March 2026, the administration put forward the National Policy Framework for Artificial Intelligence, which asked Congress to protect children, address data-center effects on local communities, prevent censorship, expand sandboxes and AI-ready data, strengthen the workforce, and establish a national AI standard that would pre-empt some state AI laws while preserving specified state powers over matters such as child protection, zoning, procurement, and public services.
Pillar I assumes that American advantage will come not only from building the strongest models but from unleashing startups, researchers, workers, manufacturers, scientists, and federal agencies.
The plan called for a regulatory review of existing policy that would hinder AI development, deployment, and adoption. OSTP issued a request for information in September 2025, initiating the promised review. In December 2025, Executive Order 14365 created a process for identifying state AI policy that the administration may challenge and connected that review to federal grant policy. In January 2026, the Department of Justice created an AI Litigation Task Force whose sole responsibility is to challenge state AI laws that the administration considers inconsistent with that policy. These actions show an administration using regulatory review, grant policy, and litigation to pursue rapid growth of the domestic AI ecosystem.
The federal government has not created the spot or forward compute market proposed in the plan. Private firms have begun to build financing mechanisms for AI infrastructure. In August 2026, NVIDIA and six financial institutions announced memoranda of understanding to create independent compute-financing platforms that would treat NVIDIA compute and full-stack AI infrastructure as an investable asset class. The partnerships remain subject to final agreements.
The Department of Labor issued an AI Literacy Framework, launched an AI-focused Registered Apprenticeship portal, and announced approximately $50 million in Rapid Reskill grants. The National Science Foundation announced a $400 million national network of programmable cloud laboratories, including $380 million in NSF investment and upwards of $20 million in philanthropic support, and NSF funded AI-ready testbeds across agriculture, transportation, disaster response, wireless systems, and other domains. These efforts have moved beyond policy concepts to published guidance, operating portals, and funding mechanisms.
NSF has made substantial progress on one part of the compute-access agenda. It reports that the National Artificial Intelligence Research Resource (NAIRR) has supported more than 600 research teams and 6,000 students in all 50 states, Washington, D.C., and Puerto Rico. Thirteen federal agencies and 28 private-sector contributors have supplied resources to the program. Those figures show meaningful reach. But the plan also proposed improving the financial market for compute, and the audit found no federal spot, forward, or similar market. NAIRR has expanded access without completing the broader market-making recommendation.
Model evaluation has moved further. CAISI, or the Center for AI Standards and Innovation, published evaluations of DeepSeek V4 Pro, GLM-5.2, and a preliminary joint assessment with the United Kingdom’s AI Security Institute of Kimi K3’s cyber capabilities. Multiple publications show an operating evaluation function rather than a single report. The public record does not yet show how agencies use the findings in procurement, critical-infrastructure, or other decisions, but that gap can be explained by confusion within the AI industry itself about the relevance or applicability of benchmarking.
The Government Services Administration (GSA) has delivered one of Pillar I’s clearest administrative outputs. Buy AI, USAi, and OneGov agreements give agencies a procurement toolbox and negotiated access to commercial AI systems. The Government Accountability Office found, however, that four agencies, including GSA, were not systematically collecting lessons from their AI acquisitions. The procurement vehicle exists; evidence that it improves acquisition quality across government remains limited. Small startups face procedural and cultural barriers to entry, and do not have the luxury of patience with respect to long procurement cycles. Well-financed startups are better off, but still struggle with the reality that capital cycles can be half as long as a single government contracting cycle for a new entity.
Agencies have achieved the most visible progress related to Pillar II’s in permitting and federal-site selection. Executive Order 14318 directed agencies to facilitate qualifying projects through FAST-41, review environmental permitting, and make federal land available for data centers and related power infrastructure. The QTS Richmond expansion became the first data-center project to receive FAST-41 coverage, and the PORTS Technology Campus in Ohio followed. These projects show that agencies are using the process. They do not yet establish shorter permitting times or coverage of all eligible projects.
The Department of Energy (DOE) began this work before the Action Plan, identifying 16 potential federal sites in April 2025. After the plan’s release, DOE selected four sites for the first round, and the National Nuclear Safety Administration (NNSA) selected Amentum to negotiate a phased lease for a data center and on-site generation on the Savannah River in South Carolina. The selection is not a final lease or a completed project. Negotiations, permitting, safety and security reviews, and other approvals remain.
On the grid, agencies remain earlier in the process. The DOE’s March 2026 Office of Electricity Strategic Plan organized its work around stabilizing, optimizing, and growing the grid. DOE’s Speed to Power initiative announced an approximately $1.9 billion SPARK funding opportunity for reconductoring and other advanced transmission upgrades, and the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators to justify or reform the rules governing the connection of data centers and other large loads. These are substantial funding and regulatory processes. They do not yet show that utilities have delivered new transmission capacity or shortened interconnection times. Many industry leaders repeatedly point to energy as a current and growing bottleneck, such as SpaceX’s Elon Musk and Microsoft’s Satya Nadella.
The National Institute of Standards and Technology (NIST) and NNSA have also produced partial outputs. NIST’s amended CHIPS research solicitation includes AI-enabled design and manufacturing, automation, digital twins, and AI applications for microelectronics. Its initial draft SP 800-239 analyzes threats and security gaps in AI data-center computing environments and offers basic recommendations. NNSA, meanwhile, launched its first enterprise cloud authorized to process Secret/Restricted Data. The solicitation and draft standard remain intermediate outputs. NNSA says the environment now supports multiple workstreams and shared tools, although it publishes no usage or performance measures. To be clear, the velocity gap between these efforts and private sector equivalents is significant and unresolved.
Taken together, the record shows more progress in creating infrastructure pathways than in delivering infrastructure. It contains covered projects, selected sites, funding opportunities, regulatory proceedings, and a draft standard. It does not yet show completed data centers, delivered megawatts, or measurable reductions in permitting and interconnection time attributable to the plan. That lack of progress is more a reflection of how difficult it has become to build infrastructure in America in recent decades rather than a criticism of the administration.
The Department of Commerce has moved furthest on the export process. Executive Order 14320 established the American AI Exports Program, and Commerce reported that it had received 78 applications. As of August 27, the department had said that final designations would follow interagency review, but it had not publicly named priority packages. The program and an applicant pool exist but final selections, financing commitments, and partner-country agreements are not yet public.
The Export-Import Bank of the United States (EXIM) and the Development Finance Corporation (DFC) have begun organizing federal financing capacity around the export program. EXIM launched its ExportAI initiative, and DFC described its financing role in supporting data centers, power infrastructure, and trusted networks. These institutions can provide credit, insurance, and political-risk support. The audit found no publicly documented financing commitment tied to a selected export package by the cutoff.
The Department of State and its partners have expanded the diplomatic machinery through Pax Silica. The United States hosted the initiative’s second summit in June 2026, with 35 countries signing its AI Opportunity Statement. The summit covered AI infrastructure security, critical mineral and semiconductor supply chains, data and infrastructure cooperation, and common standards. The State Department also posted a Pax Silica AI Assistance Project funding opportunity.
CAISI’s evaluations and two presidential memoranda provide the strongest public evidence on national security implementation. NSPM-11 directed agencies to accelerate access to advanced models, develop a roadmap for high-security computing, and create private-sector security partnerships. NSPM-12 addressed cloud security and cyber governance for national-security systems.
The highest- profile national security issues have revolved around the deployment of frontier AI systems, open-source AI, and semiconductor export controls. In March 2026, Anthropic leaders asserted to the White House that its new AI system, Mythos/Fable, was an “unstoppable weapon,” according to people involved in the discussions. In June, the administration applied a temporary export control limiting its access for non-US persons after Anthropic partners flagged jailbreaking methods that undermined developer controls. Open- source AI has been a repeated source of tension, particularly in light of the divergent business strategies between leading US and Chinese labs. Anthropic, in particular, has emerged as an ardent opponent of open- source AI, with CEO Dario Amodei repeatedly professing his view that open models are inherently unsafe. Other American AI companies, led by Nvidia, have formed the Open Secure AI Alliance in response. The Trump Administration has yet to make major revisions to Biden-era export controls on advanced semiconductors and chipmaking equipment, which were intended to maintain and grow America’s AI advantage over China.
Biosecurity, a related concept, offers less public evidence. The 2024 nucleic-acid synthesis screening framework already applies to federal life-sciences funding and requires covered recipients to obtain services from providers and manufacturers that attest to following screening, reporting, recordkeeping, and cybersecurity practices. The audit found no post-plan government-wide replacement with the requested enforcement mechanisms. It also found no OSTP-convened mechanism through which synthesis providers could share information about potentially fraudulent or malicious customers. Some work may be non-public, but the specific new mechanisms remain unavailable for public assessment.
The public record shows substantial administrative activity during the first year following the release of America’s AI Action Plan. Of the plan’s 103 recommendations, the administration has completed a public administrative output for seven and partially implemented 54; work remains in progress on nine, while 33 remain not publicly verifiable. Most of the supporting evidence consists of guidance, requests for information, solicitations, task forces, program announcements, procurement vehicles, site selections, and regulatory proceedings. Much less evidence shows that agencies and their partners have deployed capital, brought infrastructure online, trained workers, improved procurement, or completed transactions.
The record is strongest where a department can satisfy a recommendation by issuing an administrative artifact. It is weaker where implementation depends on construction, appropriations, sustained coordination across agencies, or agreement with companies and foreign governments. Much of that work is underway, but will not be achievable through unilateral executive action alone.
By next year, an updated implementation review should move beyond the existence of programs and processes. At some point, workforce grants must reach workers, agencies must markedly improve the quality of AI procurement, permitting and grid initiatives must reduce delivery times, federal sites must produce operating infrastructure, and export applications must become financed agreements with partner countries. Those measures would demonstrate implementation rather than administrative preparation.
Congress can strengthen that record without taking over implementation. The administration’s National Policy Framework for Artificial Intelligence asks Congress to act on infrastructure, workforce policy, national rules, and federal capacity. Congress can provide durable authority and appropriations where executive action is insufficient, and it can require a regular public implementation report tied to the plan’s 103 recommendations.
The administration has created much of the machinery that America’s AI Action Plan calls for. The second year will show whether that machinery produces results: researchers gain access, workers receive training, agencies improve procurement, developers build infrastructure, utilities deliver power, financing institutions support exports, and agencies enforce security measures. That standard should guide the next assessment of implementation.
