Global Compute and National Security

1. Harden export controls through stronger enforcement, technology-forward solutions, and engagement with key allies and partners.
2. Support the development of and access to valuable downstream AI applications to reduce demand for sovereign frontier compute and to increase the appeal of the United States as the AI partner of choice.
3. Use strategic investment vehicles to promote U.S. AI projects and applications overseas.
4. Expand engagement with key strategic partners to design mutually beneficial AI partnerships, specifically:
   1. The United Arab Emirates and Saudi Arabia: Continue to partner on the development of secure data centers in the Gulf and globally, accompanied by security measures and safeguards that ensure that these nations do not overtake the United States’ lead in AI compute.
   2. Brazil and India: Through bespoke government-to-government partnerships, leverage their statuses as emerging technology economies with strong AI ambitions to bring them closer into the U.S. orbit.
   3. Australia and the United Kingdom: Leverage the Australia–United Kingdom–United States trilateral security partnership, AUKUS, to enlist the United States’ closest allies as trusted frontier AI partners, helping overcome U.S. domestic bottlenecks and ensuring that America can continue to benefit from returns to scale.
5. Champion technical privacy-preserving solutions to help address partners’ “sovereignty” concerns.
6. Engage in multilateral and standard-setting bodies to build legitimacy and to strengthen AI data center security and risk management.

The United States faces a choice: leverage its current lead to promote U.S. AI infrastructure and applications globally, while preserving its edge at the frontier; or continue to primarily focus on protection, while other countries gradually narrow the gap. As Michael Kratsios, President Donald Trump’s science and technology advisor, put it: “It is not enough to seek to protect America’s technological lead. We also have a duty to promote American technological leadership.”⁷ The protect and promote strategy outlined in this report offers a path to sustainable leadership that both safeguards critical capabilities and expands American influence in the global AI ecosystem.

Introduction

In March 2025, President Donald Trump wrote to Michael Kratsios, the director for the White House Office of Science and Technology Policy, to ask a momentous question: “How can the United States secure its position as the unrivaled world leader in critical and emerging technologies—such as artificial intelligence . . . ?”10 This question reflects a growing consensus in Washington and capitals around the world that the nation that leads the world in advanced artificial intelligence (AI) will pioneer new scientific breakthroughs, unlock powerful military and intelligence capabilities, and turbocharge economic competitiveness. Although there is ongoing debate about the best pathway to long-term AI progress, large-scale AI models show the most transformative promise. Historically, AI systems were specialized, with training data narrowly tailored to their intended function. Now, leading foundation models—which ingest vast collections spanning trillions of words from diverse sources—currently perform as well as purpose-built systems in specialized domains, as well as across general-purpose tasks.¹⁰ The United States leads the world in large-scale AI development, driven in part by its leading talent and innovation ecosystem, but also by its access to cutting-edge “compute”—the specialized chips, data centers, and infrastructure needed to train and deploy the most capable AI systems.

But U.S. AI leadership is not guaranteed. The release of DeepSeek-R1, the first highly capable reasoning model from a Chinese firm, demonstrated China’s capacity to apply breakthroughs in algorithmic efficiency to do more with fewer chips in the face of U.S. export controls. DeepSeek released R1 a little more than four months after OpenAI released its o1 model, with approximately the same performance. It is widely believed that DeepSeek-R1 leveraged OpenAI’s know-how, whether through training a new model from the o1 model’s outputs in a process known as “distillation,” or through informal conversations at Silicon Valley parties.¹⁰ This highlights the inherent difficulty of controlling the spread of technical knowledge and expertise.¹¹ In light of China’s advancements, the U.S. government needs a clear strategy for protecting and enhancing its AI edge, and compute is one of its best levers available. Unlike algorithms and know-how, compute is physical, with a narrow, specialized supply chain, making it significantly more controllable through policy.

Compute has become the engine of AI progress, with advanced chips and massive data centers enabling breakthroughs in AI capabilities and large-scale deployment. Although algorithmic progress is allowing AI models to become far more efficient—the compute needed to train models of equivalent capability drops by half every eight months—the availability of compute still drives the sophistication and scale of their deployment.¹² Critically, advancing the AI frontier continues to require exponentially more compute resources. All of this underscores the strategic imperative of maintaining and sustaining America’s edge in compute.¹³ In an age of escalating technology competition, nations with robust compute ecosystems will likely command a significant advantage in both AI development and deployment.

The United States’ approach to preserving its AI compute advantage has historically relied heavily on export controls—an inherently restrictive measure. These controls have sought to solidify America’s advantage over China and other competitors by limiting their access to advanced AI chips. This has allowed the United States to continue to lead in AI data center build-outs. But this strategy faces mounting challenges: overreliance on restrictive measures risks alienating allies and encouraging the rise of alternative AI supply chains. The People’s Republic of China (PRC) is investing to indigenize AI semiconductor production, including close to $100 billion through its state-owned National Integrated Circuit Industry Investment Fund.¹⁴ Domestically, U.S. companies face constraints in energy availability, which hinders their ambitions to deploy large-scale computing infrastructure. Even traditionally close partners are growing skeptical of Washington’s expanding use of economic tools for national security objectives, with even close allies calling for non-U.S. AI compute.¹⁵

Faced with these headwinds, the current U.S. lead may be short-lived. It is therefore imperative for policymakers to think strategically about how to leverage this advantage now to shape the global AI ecosystem. Washington needs a more proactive “promote” approach to turn its short-term lead into an enduring advantage, without allowing competitors to leapfrog the United States in AI compute.

The Trump administration has commenced decisive action on this front, announcing major AI partnerships with the United Arab Emirates and the Kingdom of Saudi Arabia in May 2025. The United Arab Emirates agreement includes plans for a five-gigawatt (GW) AI campus in Abu Dhabi and a deal for the United States to enable the import of 500,000 advanced NVIDIA AI chips annually, while the Saudi Arabia deal includes over $80 billion in cross-border AI investments through multiple company-to-company agreements.¹⁶ While the details—and security measures—are still being developed, these initiatives represent a significant shift toward using AI partnerships as instruments of strategic competition, demonstrating a new willingness to deploy compute resources as tools of technological diplomacy rather than merely restricting access to competitors.

U.S. policy settings now need to catch up. Beyond individual deals, U.S. AI leadership requires moving from a policy of global restriction and control to responsible deployment and diffusion, accompanied by security measures and safeguards. Only then can the United States appropriately manage emerging national security threats from foreign and domestic AI progress and ensure democratic leadership of the global AI ecosystem.

This report outlines the ongoing importance of compute for the development and deployment of AI—even in the face of increasingly cost-efficient models. It articulates the national security rationale for active efforts to maintain and strengthen U.S. AI leadership. Assessing compute governance actions to date, it argues for a more proactive U.S. approach to global compute provision, shifting from purely protecting U.S. technology to a strategy of promotion and AI diplomacy, underpinned by partnerships with strategic countries. This report lays a path for near-term engagements, identifying six priority nations to engage with based on strategic importance, compute needs, and alignment potential: the United Arab Emirates, Saudi Arabia, Brazil, India, and AUKUS partners (Australia and the United Kingdom). The report sketches out next steps for advancing these critical partnerships, while strengthening controls on adversaries such as China. Finally, this report evaluates the role of multilateralism in sustaining and supporting U.S. AI leadership and makes a series of recommendations to inform American AI leadership.

Ultimately, AI and its impacts will transcend national borders. But the infrastructure that underpins it will remain firmly rooted in the physical world. The United States has a choice: proactively promote its AI globally to maintain technological leadership and shape the rules of AI development, or risk watching its advantage erode as competitors build alternative ecosystems that diminish America’s power to manage AI opportunities and risks and ensure U.S. security and prosperity.