As Xi Jinping’s state visit to Washington draws to a close, the race for AI leadership is one issue that will outlast the pageantry. The Trump administration’s 2025 executive order and subsequent AI Action Plan put innovation, infrastructure, and national security at the center of that effort. The objective is clear: move fast enough to preserve American leadership. But speed alone will not decide this race.
The debate has shifted in recent weeks. Leaders at Anthropic, OpenAI, and xAI have publicly backed some form of pacing the frontier — slowing the rate of capability gains enough for safety practices, evaluations, and oversight to keep up. That is not the same as stopping AI development. It is an argument for ensuring that progress remains durable. Epoch AI estimates that Chinese frontier models have trailed the American frontier by an average of roughly seven months since 2023. That lead is meaningful, but it is not permanent.
A serious AI failure could narrow it quickly. An incident affecting critical infrastructure, financial systems, or public safety could trigger a regulatory backlash, shake investor confidence, and make allies more cautious about adopting American systems.
Competitiveness therefore depends on reliability as well as raw capability. If the United States wants to lead the world in AI, it also has to lead in demonstrating that the technology can be deployed securely and responsibly.
The strongest near-term source of strategic leverage remains advanced compute. China continues to face constraints in access to the most capable chips and semiconductor manufacturing tools, even as Chinese firms invest aggressively in domestic alternatives and more efficient models.
In a purported transcript of a private May investor meeting that circulated publicly this summer, DeepSeek founder Liang Wenfeng described compute resources as the principal gap separating Chinese labs from their American competitors. U.S. officials have separately alleged that DeepSeek obtained restricted American hardware through third countries, underscoring both the value and the limits of export controls. The lesson is not that controls have solved the problem. It is that enforcement matters.
Washington should therefore focus on the points of greatest leverage: reducing chip diversion and smuggling, tightening controls on critical semiconductor-manufacturing equipment, protecting model weights and other proprietary capabilities, and improving security partnerships between government and frontier labs. These measures can buy time. And time is strategically valuable if the United States uses it to improve safety, deployment, and adoption rather than simply race for the next benchmark.
None of this works if the United States acts alone. The semiconductor supply chain runs through allied and partner economies, including the Netherlands, Japan, South Korea, and Taiwan. Controls are only as strong as their weakest link. A durable U.S. advantage therefore requires sustained coordination with allies on export controls, technology protection, and enforcement.
The same logic applies to sovereign AI. From Europe to the Gulf to the Indo-Pacific, governments increasingly want AI capacity they can control, on infrastructure they trust, under legal frameworks they understand. That is not necessarily a threat to American leadership; it can be an opportunity. If the United States offers partners secure compute, competitive models, and meaningful control over deployment, it will strengthen the case for partners to build on an American technology stack. If it does not, Chinese providers will have more room to shape the standards and systems other countries adopt.
That is why the relationship between government and industry matters. Export controls, counterintelligence, model security, and incident response all depend on information the federal government cannot obtain on its own.
At the same time, AI companies should not be treated as extensions of the state. The better model is partnership with accountability: government sets clear national-security priorities and guardrails, while companies retain the freedom to innovate and accept meaningful external evaluation where the risks warrant it.
China operates under a very different political-economic model, one in which the state has far more direct authority over companies, data, and information. Its AI regulatory system includes state rules governing algorithms, training data, labeling, and security assessments. The United States should not imitate that system. But it should recognize the competitive asymmetry. A fragmented relationship between Washington and American AI companies would create vulnerabilities that Beijing does not face in the same way.
There is also an opposite risk: responding to legitimate safety, energy, labor, and social concerns with measures so broad that they constrain American capacity without reducing global risk. Sen. Bernie Sanders (I., Vt.) and Rep. Alexandria Ocasio-Cortez (D., N.Y.) have proposed a federal moratorium on new AI data centers, and Sanders has separately backed a temporary pause in advanced AI development.
Those proposals reflect real concerns that deserve debate. But a broad U.S.-only freeze would risk slowing domestic infrastructure and capability while leaving foreign competitors free to continue building. The better target is high-risk uses, infrastructure impacts, and specific safety failures — not AI development as a category.
President Donald Trump framed the stakes with a simple line: “Whoever wins AI, wins.” The administration’s AI Action Plan already points toward several tools that fit a strategic-pacing approach, including stronger export-control enforcement, location verification for advanced compute, coordination with allies, and national-security evaluations of frontier models.
The choice is not between acceleration and safety. The harder task is to build a system in which each reinforces the other. Strategic pacing means moving fast enough to preserve U.S. leadership, but deliberately enough to avoid the kind of failure that could halt progress, fracture public trust, or push allies toward competing systems. The objective is not less AI. It is durable American leadership in AI.
Dr. Jake Sotiriadis is executive director of Global Foresight and Strategy at Phaedrus Engineering. A geopolitical strategist and a retired U.S. Air Force intelligence officer, he is the author of The Revenge of Ideology: The Hidden Forces Reshaping Global Power and serves as a nonresident senior fellow at the Atlantic Council and a senior associate (nonresident) at the Center for Strategic and International Studies. The views expressed are his own.
