On 23 July 2025 the White House released Winning the AI Race: America’s AI Action Plan. The document lists more than 90 federal policy recommendations, and several AI-related executive orders were signed the same day. It marks a clear shift in the centre of gravity of US federal AI policy.
From risk constraint to competitiveness first
Under the previous administration, federal AI governance used executive orders to stress safety evaluation and risk disclosure. This action plan changes the main axis to “winning the race”: reducing friction, speeding deployment and expanding exports.
The specific directions include simplifying federal AI procurement and deployment; promoting outward exports of the US AI technology stack (chips, models, cloud, applications); and accelerating permitting and construction of data centres and compute infrastructure.
What the same-day executive orders reveal
The executive orders signed the same day point to three levels:
Procurement. Federal agencies are required to procure AI models that remain truthful and ideologically neutral, writing value orientation into procurement standards.
Export. Promotion of US AI technology-stack exports to allied and partner markets, tying technical standards to market access.
Infrastructure. Accelerated permitting for data-centre land, energy and grid connection, easing the bottleneck on compute expansion.
Together these form a closed loop: compute is the base, models are the product, and export is the market. The policy tools concentrate on reducing domestic friction and opening overseas markets.
Tension with state-level legislation
The federal push for “one rule, less fragmentation” is in tension with the AI legislation that keeps appearing at state level. See our notes on the California frontier AI transparency act and the US executive order on a national AI policy framework.
This federal-versus-state contest directly affects the compliance cost of companies operating in the US: the same product may need to meet different transparency and disclosure requirements in different states.
What it actually means for Chinese companies
First, compliance assessment for going to the US must separate the federal and state tracks and cannot rely on federal policy direction alone. See our note on the Korea AI framework act and decree for how another jurisdiction structures its regime.
Second, watch the access conditions inside technology-stack export: export promotion is often accompanied by security review and origin restrictions. See our note on the US executive order on AI model review.
Third, allow for policy uncertainty: executive orders can be adjusted by later ones, so companies should keep their compliance plans flexible.
The US route reminds us that there is no single template for AI governance. A governance path that puts competitiveness first, and one that puts risk first, will impose very different compliance demands on companies. See our note on the State Council’s AI development and deployment.
This article is based on public information on the White House’s Winning the AI Race: America’s AI Action Plan and related executive orders and is for reference only; it does not constitute legal advice.