In July, the Trump Administration released a document entitled “Winning the Race: America’s AI Action Plan,” introducing a fresh perspective on export controls for artificial intelligence (AI), open-source AI, and U.S. leadership in the development of this technology. At its core, the plan recognizes that the United States cannot rely simply on restricting other countries’ access to U.S. AI technologies to maintain leadership. Instead, the administration argues that the United States must promote the global adoption of its AI stack – including its semiconductor chips, AI systems, and standards.

For years, strict AI chip export controls and skepticism toward open-source models defined U.S. policy. The assumption was if others lacked the best chips and models, they would never catch up. But the world had other plans.

Recent developments, such as the release of the Chinese open model DeepSeek, demonstrated that countries could leverage models developed elsewhere – optimizing and building on them to advance their own AI capabilities – which could move further developments abroad and, in turn, undermine the intended effect of export restrictions, risking U.S. competitiveness in the global AI market.

The Trump Administration’s AI Action Plan marks a shift from an overreliance on export restrictions to a more strategic play. It calls for exporting the full AI stack of hardware, models, software, applications, and standards by soliciting proposals from industry consortia and facilitating deals that meet U.S.-approved security requirements. The plan also backs the use of creative enforcement tools, such as location-verification features on advanced AI compute to keep chips out of countries of concern (among them China) and closing loopholes in their export controls. The goal is clear: Shape, and not restrict, international AI development and distribution through strategic exports and targeted enforcement to drive innovation and capture global markets.

Notably, while the AI action plan ultimately leaves decisions on whether to release AI models that are “open” or “closed” with developers, it does endorse the creation of an environment that encourages openness where appropriate. There’s a great deal of potential benefit to this approach: As open models are freely available for anyone to download, study, and modify, these models allow startups, researchers, and institutions to innovate without dependence on closed-model providers. In doing so, they can help maintain U.S. leadership – not by locking others out, but by ensuring the United States, and its AI technology, remains the most attractive and dynamic hub for AI development.

The plan does face a critical obstacle, however: Without congressional action, it remains largely aspirational. Previous restrictions have often been overly tight and potentially harmful to U.S. AI development, and therefore fostering innovation and safeguarding leadership will depend on Congress’ willingness to codify the administration’s plan – and on agencies’ ability to implement these rules effectively. After all, leadership isn’t won by hiding the ball. It’s won by making sure everyone wants to play on your field.