The Real AI Race

Discussions in Washington about artificial intelligence increasingly turn to how the United States can win the AI race with China. One of President Donald Trump’s first acts upon returning to office was to sign an executive order declaring the need to “sustain and enhance America’s global AI dominance.” At the Paris AI Action Summit in February, Vice President JD Vance emphasized the administration’s commitment to ensuring that “American AI technology continues to be the gold standard worldwide.” And in May, White House AI and Crypto Czar David Sacks cited the need “to win the AI race” to justify exporting advanced AI chips to the United Arab Emirates and Saudi Arabia.

Given the prospect that AI could transform the power and prosperity of nations in the decades to come, it is better to win the race than lose it. But determining who is ahead depends on what it means to win. A common definition is being the first to cross the threshold of artificial general intelligence, which in basic terms is an AI model that is as smart or smarter than the top human experts across a wide range of cognitive tasks. AGI could unlock extraordinary breakthroughs in science, technology, and economic productivity—and the first country to develop it could reap disproportionate benefits.

But the race to AGI is not the only critical race in the AI contest. Militaries and intelligence agencies must harness AI’s transformative potential and mitigate its disruptive effects. Similarly, countries stand to gain a competitive edge if they can adopt AI at scale across the economy and society. Governments are also battling to create and own the standards, supply chains, and infrastructure that will undergird the global technological ecosystem. And all must avoid a race to the bottom in AI safety by working—sometimes together—to manage security risks from misused or rogue AI.

Taking these additional AI races into account makes the United States’ position look precarious. Although U.S. companies maintain a meaningful—if narrowing—lead at the frontier of AI research and development, Washington could lose other AI races. China has considerable advantages, and neither superpower seems eager to cooperate to avoid catastrophe. And given AI’s world-changing potential, the stakes are profound: losing risks relegating the United States to economic dependency, military vulnerability, and diminished global leadership. That bleak future can still be avoided. But the United States will need to muster a coherent AI strategy, one that balances innovation, integration, and risk mitigation to translate the country’s immense technological dynamism into enduring strategic advantage.

THE RACE TO INNOVATE

The race to AGI is the most visible and immediate of the AI contests. Private companies such as OpenAI, Anthropic, and Google DeepMind in the United States and DeepSeek in China are rushing to innovate, supported by their respective governments. No one knows exactly how the technology will evolve. Large language models could be the first signs of emerging AGI, or true AGI could manifest suddenly when AI models pass a certain threshold. Either way, AGI has enough potential to transform the sources of national power and competitiveness that the world’s two AI leaders have a substantial interest in securing a first-mover advantage.

U.S. AI labs currently have a discernible edge, helped along by semiconductor export controls designed to maintain the United States’ computational advantage over China. But this lead is fragile. China’s indigenous innovation, circumvention of export controls, and intellectual property theft have kept the country in a close second place. Its leading AI companies, such as DeepSeek, are developing technologies that trail their U.S. counterparts by mere months. And Beijing’s centralized approach might help it nurture, consolidate, and harness private-sector innovations more quickly than Washington can.

The race to artificial general intelligence is not the only race in the AI contest.

The United States’ open system, meanwhile, fosters innovation but is inherently vulnerable to espionage and the rapid diffusion of algorithmic advances. Breakthroughs in algorithmic design or alternative paradigms for AI development could diminish the importance of U.S. semiconductor dominance, and their spread could enable China’s AI labs to leapfrog American competitors. The Trump administration, driven by other political priorities, is also pulling back investment in basic AI research and development and discouraging talented foreign nationals from working in the United States, potentially setting back U.S. AI efforts over the next several years.

The U.S. private sector, moreover, responds to commercial imperatives that are not always consistent with national priorities. AI firms, for example, will be tempted to build AI computing power wherever there is available energy infrastructure—whether or not it is based in the United States. Favorable regulatory environments and abundant resources in the Middle East are already proving attractive. The world’s first five-gigawatt AI data center cluster will be built in the UAE, not the United States—a development enabled by the Trump administration’s recent decision to export hundreds of thousands of leading-edge AI chips to Abu Dhabi. Washington does benefit from this arrangement, and U.S. firms including OpenAI and Microsoft are slated to operate most of the data centers’ capacity. But moving crucial infrastructure offshore, where security may be lax, could also provide a backdoor for China and other competitors to acquire advanced computing resources and AI models.

Even if the United States can hold on to its lead in the innovation race, that may not be enough. Current market trends suggest that frontier AI models are becoming so widely accessible and undifferentiated that they give no one a clear technological edge. If that trend holds when AGI emerges, victory will depend on effective AI adoption—and there is no guarantee the United States will come out on top.

SECURING THE EDGE

In the realm of national security, effective AI adoption requires both understanding the capabilities enabled (and threats posed) by frontier AI and integrating AI into existing structures in ways that secure a decisive military edge. Integration of AI promises to enhance intelligence processing, accelerate data-driven decision-making, optimize logistics and resource allocation, enable sophisticated autonomous systems, and possibly even lead to the development of a “wonder weapon”—such as a cyberweapon that could cripple an adversary’s critical infrastructure and command and control or, used defensively, make a country invulnerable to cyberattacks.

The U.S. government and private industry need to work together to achieve AI integration, but their current cooperation remains troublingly limited. National security agencies lack early access to the latest AI models, which would speed up the process of incorporating the latest technologies into their workflows. Partnerships between leading AI labs and the Pentagon and other national security agencies are only in nascent stages. And lengthy procurement cycles, operational cultures that are resistant to change, a lack of infrastructure and data, and misunderstandings about what AI can achieve—both underestimating and overestimating its abilities—hamper the government’s ability to take full advantage of innovations coming out of Silicon Valley. Although these problems are widely acknowledged, fixing them has proven difficult.

China’s authoritarian system, meanwhile, eases civil-military integration, providing it with a structural edge in AI adoption. State mandates ensure that technological advances are rapidly translated into military and intelligence capabilities. The People’s Liberation Army has embraced AI and is actively seeking contributions from the commercial and academic sectors. Making use of AI competitions and public purchasing platforms designed to translate civilian AI research to military applications, it plans to field “algorithmic warfare” and “network-centric warfare” capabilities by 2030. More than merely using algorithms in weapons systems, this entails a transition to a new type of warfare in which military superiority depends on the speed, sophistication, and reliability of those algorithms.

Remaining on the cutting edge of innovation is necessary but not sufficient to win the race in the national security space. The United States could produce scientific and technological breakthrough after breakthrough but still fail to recognize the point at which AI opens a new technological pathway to a revolutionary military or intelligence capability. Washington’s bureaucratic structures, designed for incremental improvements to existing systems, often make it difficult to imagine left-field possibilities for emerging technologies. Beijing’s centralized decision-making system, in comparison, could identify and exploit a disruptive pathway much faster, potentially leaving the United States technologically superior but strategically outmaneuvered.

THE INTEGRATION IMPERATIVE

The winner of the AI race will also need to integrate AI into the national economy—to ensure that AI is widely accessible and is diffused across the education, energy, finance, health, logistics, and manufacturing sectors. The U.S. technology companies that drive consumer and business AI applications, a vibrant venture capital ecosystem that funds innovation, relatively high digital literacy, and extensive digital infrastructure all provide the United States notable advantages. Yet success is not guaranteed. If corporate and governmental actors at all levels fail to create the right incentives for integration and build sufficient public trust in AI, the private sector could struggle to adopt AI quickly enough to capitalize on the benefits of enhanced productivity and new value creation. There is also a danger that AI will not simply augment human labor but replace it. Anthropic CEO Dario Amodei has recently warned that AI could lead to as much as 20 percent unemployment in the United States within five years—an outcome that could deeply destabilize both the economy and American society.

Meanwhile, China may be poised to perform surprisingly well in the race for economic adoption. Chinese business leaders are already more focused on AI applications than on developing AI models. DeepSeek’s open-source models, for example, are driving down costs for all Chinese AI models, which enables more businesses to experiment with the technology. This could give China an edge in creating game-changing products. The Chinese government is also less exposed to the political consequences of AI replacing labor. Whereas automating jobs raises alarm in Washington, Beijing could even welcome AI adoption as a solution to China’s labor shortages brought by a rapidly aging and shrinking population.

Even if the United States can keep its lead in innovation, that may not be enough.

Even if the United States adopts AI across its economy as aggressively as China does, it may lose the overall race if China is better positioned to capitalize on the manufacturing advances enabled by AI, particularly in the realm of robotics. China leads the world in industrial robot installations: Chinese manufacturers purchased half of the global market share in 2024, and the country’s robots per employee significantly exceeded the global average. Extreme automation is becoming more common in the manufacturing sector with the proliferation of “dark factories,” such as the electronics company Xiaomi’s smartphone facility that operates 24/7 without human workers. As AI makes strides in spatial reasoning and embodied intelligence—AI that enables robots to interact with and learn from their physical environment—factory robots may become able to perform a much wider array of complex physical tasks.

Although U.S. firms excel in software and services—areas that are also expected to make significant productivity gains as a result of AI adoption—the United States has ceded ground to China in recent decades in physical industries, including manufacturing, logistics, energy, and infrastructure. With its state-driven industrial policy and massive manufacturing base, China can deploy AI at scale within these sectors and might unleash dramatic productivity gains that lead it to finally surpass the U.S. economy.

STACKING THE DECK

The world’s tech powers are also racing to provide the digital infrastructure that will undergird the global development, deployment, and use of AI. Although this is primarily a competition between the United States and China, other established tech powers (such as France, Japan, the Netherlands, South Korea, Taiwan, and the United Kingdom) and ambitious emerging players (such as Brazil, India, Saudi Arabia, and the UAE) are joining the contest, too. Each participant aims to control the data, the chips and data centers, and the foundational models required for AI use, as well as exert influence over global AI norms and standards. As Sacks, the White House czar, recently put it, “If 80 percent of the world uses the American tech stack, that’s winning. If 80 percent uses Chinese tech, that’s losing.”

Export controls on semiconductor chips have given the United States a meaningful edge. U.S. companies have access to the chips they need for computational power, and much of the world wants U.S. chips because they are the best on offer. The Trump administration is seeking to capitalize on this advantage by “flooding the zone” with U.S. chips and data centers, starting in partner countries such as Saudia Arabia and the UAE. The idea is to lock in the use of U.S. technology in places where market forces encourage massive investments in digital infrastructure.

But in countries with lower incomes, fewer customers, and less basic infrastructure such as broadband connectivity and electricity, Washington’s strategy of following market forces will lead to underinvestment. This is the dynamic across a broad group of countries in Africa, Latin America, the Middle East, and South and Southeast Asia that are turning to AI to boost their economic growth. China is positioned to outcompete the United States in these places by providing less advanced and significantly cheaper AI models—as well as subsidizing the physical and digital infrastructure needed to run them. It may not currently be able to export any top-of-the-line chips, but for many developing countries where cost and accessibility are more important than cutting-edge performance, China’s “good enough” offerings could prove highly attractive.

Ceding these emerging markets risks a future in which the United States wins the technological race at the frontier but surrenders leadership of the global AI ecosystem to China. The consequences for Washington extend beyond lost geopolitical influence and commercial opportunities. Chinese AI models and infrastructure frequently embody digital authoritarian values, enabling Beijing to export mechanisms of state control and shape historical and political narratives beyond its borders. They facilitate surveillance by powering facial and voice recognition systems and analyzing vast amounts of data to monitor individuals and flag “suspicious” behavior. They automatically censor content critical of the Chinese Communist Party or related to sensitive topics such as Tiananmen Square, Taiwan, Hong Kong, the Uyghurs, or Chinese leader Xi Jinping. Their algorithms also curate and disseminate pro-Chinese propaganda. In contrast, U.S. AI models generally reflect stronger commitments to democratic norms, transparency, privacy safeguards, user choice, and data protection frameworks, helping reduce opportunities for government abuse. A critical step toward ensuring the technologies shaping modern life across the world align with democratic values is for Washington to proactively incentivize U.S. AI firms to invest and build infrastructure in developing countries.

Poor policy choices could set the United States back in the race to build and manage the world’s AI infrastructure. Overly restrictive export controls could alienate allies or drive countries currently on the fence toward Chinese products. Conversely, applying too little control over advanced AI chips could inadvertently create opportunities for Chinese companies to acquire or remotely access them, potentially accelerating China’s technological innovation. And if Washington cannot articulate a compelling vision for technological governance—one that carefully balances national security with economic openness and democratic values—potential international partners may turn elsewhere. China is not just selling AI; it offers a comprehensive toolkit for rapid modernization on financially and politically appealing terms to a substantial portion of the world. For a developing country eager to harness AI for economic gains and improved governance, the Chinese offer is often the most practical and readily available path forward.

A RACE TO THE BOTTOM?

Even as the United States and China compete in the AI race, they cannot forget that AGI and other highly capable AI models create potential threats. It is imperative that neither competitor allows its rapid development and deployment of AI to make a disaster more likely. A nonstate actor or rogue state weaponizing AI models, unintentional military escalation caused by AI, or a loss of control of superintelligent systems could prove catastrophic. Preventing such outcomes requires that AI powers avoid cutting corners on safety in their haste to compete.

The most pressing near-term catastrophic risks are that nonstate actors with access to advanced AI systems could launch large-scale cyberattacks that devastate financial systems or design and release highly lethal and transmissible pathogens. These hazards are no longer confined to the realm of science fiction. Companies at the forefront of AI development predict that their internally defined thresholds for dangerous capabilities—measured as “uplift,” or the degree to which an AI model significantly enhances a malicious actor’s abilities—are likely to be crossed this year. Another rapidly approaching threat is the potential emergence of a superintelligence misaligned with human values or intentions, pursuing actions that endanger human well-being because of flawed design, ambiguous instructions, or unforeseen consequences. Growing evidence of frontier AI models exhibiting deceptive or scheming behaviors makes this risk increasingly credible.

Both Washington and Beijing have an interest in preventing the proliferation of dangerous AI capabilities and the emergence of rogue AGI. This common ground creates an opportunity for cooperation between the two AI superpowers—even amid their intense technological competition—to better understand risks of misuse and misalignment and to identify and develop effective mitigation measures.

For Washington, seeking ways to mitigate threats from AI—an imperative the Trump administration has recognized—is sound policy regardless of what other AI powers do. Even if rogue AI does not lead to global calamity, a major AI-related incident originating from the United States, whether accidental or through negligence, would undermine confidence in American technology. And if the United States is seen as unable to manage the immense power of AI, its global leadership and moral authority would be called into question, and China could exploit the resulting power vacuum by promising stability and control.

RECIPE FOR SUCCESS

The trajectories of the various AI races between China and the United States are tightly intertwined. Winning the race to AGI development, for example, will boost the leading country’s national security, economic vitality, and global technological influence. And the competitive pressures unleashed by the races to develop and adopt AI exacerbate the risk of a catastrophic outcome as haste and rivalry undercut safety—a danger no country can outrun. Success in the AI race therefore requires a strategy that advances progress across several fronts simultaneously while managing the security risks of unchecked AI.

As part of a holistic strategy, Washington must make every effort to avoid being caught off guard by AI advances in Silicon Valley, China, and emerging AI hubs around the world. Being surprised could mean failing to recognize the emergence of new threats or missing opportunities to capitalize on AI progress ahead of China. To prevent this, the U.S. government must maintain close communication with domestic industry leaders and closely monitor imminent technological breakthroughs abroad. The Commerce Department’s repurposing of its AI Safety Institute, now called the U.S. Center for AI Standards and Innovation, to focus on working with industry to research and test frontier AI is a step in the right direction. National security departments and agencies, too, should remain apprised of the latest frontier AI developments and explore potential use cases. Likewise, the intelligence community must expand its monitoring of foreign AI efforts, particularly focusing on China’s advancements and objectives, as well as those of emerging AI powers in the Middle East.

The Trump administration should also identify ways to facilitate frontier AI development. It must ensure that AI companies have access to the resources they need for AI model development and deployment, including vast computational power (in the form of semiconductors), high-quality data, world-class talent, and sufficient energy supplies. It will be important not to create new problems in the process; to meet AI’s growing energy demands without exacerbating climate change, for example, Washington should invest in cleaner energy sources such as nuclear power.

The private sector could struggle to adopt AI quickly enough to capitalize on its benefits.

Washington must defend U.S. technological superiority, too. To ensure that AI advancements are not rapidly replicated by competitors, the U.S. government will need to enforce stringent controls over technologies such as advanced semiconductors and manufacturing equipment, strengthen security measures at research labs and data centers to prevent espionage and intellectual property theft, and require rigorous user verification on cloud computing platforms so that they do not become unwitting tools for an adversary’s technological advancement.

To keep the United States at the forefront of innovation while safeguarding national interests, the U.S. government must develop a scalable and adaptive public-private partnership model to cooperate with companies working on frontier AI. These initiatives should help the government ramp up its adoption of advanced AI and promote prudent security practices. U.S. AI companies can benefit from access to sensitive government intelligence about adversaries attempting to target them, and both the public and private sectors can benefit from co-developing AI applications that can enhance national security, such as advanced cybersecurity and biodefense tools.

As China expands its industrial capacity using AI and robotics, policymakers in Washington must also broaden AI adoption beyond the technology sector. The Trump administration should work with Congress to launch a dedicated “industrial AI” initiative to accelerate research, development, and investment in robotics and AI deployment across the manufacturing, logistics, energy, and infrastructure sectors. With tax credits, innovation grants, and public-private pilot projects, the government can incentivize factories, warehouses, and transportation hubs to integrate AI-driven systems, bridging the gap between the United States’ cutting-edge software capabilities and its lagging factory floors.

Policymakers must also act now to help the workers who will lose their jobs to AI in these industrial sectors. This means significantly increasing investments in STEM education, vocational training, and retraining and upskilling programs—services that would enable workers displaced by automation to swiftly transition into new roles, such as robot maintenance or AI system supervision. To provide clarity for businesses and protections for workers in an AI-centric economy, labor laws and regulations also require updates. As companies increasingly use algorithms to schedule work shifts and more employees work on short-term contracts, wage-and-hour laws should be modernized to ensure transparent, fair compensation and clearly defined working hours. Workplace safety guidelines will need to be revised to incorporate standards for safe human-robot interaction on factory floors. Strengthening unemployment benefits and other forms of direct income support for people disproportionally affected by automation will also be crucial to mitigating the potentially destabilizing consequences of significant labor displacement.

AI powers must avoid cutting corners on safety in their haste to compete.

If the United States is to lead the global AI technology ecosystem, it will also need to provide advanced AI and data centers to more than just wealthy countries. To compete with China’s “good enough” AI systems across the developing world, the Trump administration should explore public-private partnerships to offer generous access to U.S. cloud computing systems to researchers and entrepreneurs in these countries. It should also scale up government-backed financial tools—such as low-interest loans, loan guarantees, equity investments, political risk insurance, and tax incentives—through agencies such as the International Development Finance Corporation. These incentives should be geared toward building digital infrastructure in important emerging markets, such as in Brazil, Ghana, India, Indonesia, Kenya, Mexico, Nigeria, the Philippines, and Vietnam.

Finally, the Trump administration must take steps to reduce the risk of worst-case scenarios—and prepare for those contingencies should they arise. It should run tabletop exercises and crisis simulations of cases involving catastrophic misuse of AI or rogue superintelligence. Doing so would give senior leaders opportunities to rehearse crisis responses, identify gaps in readiness, and improve their decision-making under pressure.

The United States and China, moreover, have a profound obligation to themselves and the world to collaborate to reduce AI risks. Following the 1962 Cuban missile crisis, the United States and the Soviet Union built guardrails around their nuclear competition. They cooperated on negotiations for the 1970 Nuclear Non-Proliferation Treaty, recognizing that an uncontrolled race to the nuclear frontier could send humanity hurtling off a cliff. Washington and Beijing today need to chart a similarly narrow path between AI competition and collaboration. They could start with a bilateral agreement to share AI incident information and exchange best practices on AI safety, control, and alignment with human values. Further talks should focus on how to handle scenarios of misused or uncontrolled AGI.

The notion of a singular AI race between the United States and China fails to capture the true complexity of the rivalry unfolding today. The challenge is to win not one definitive contest but a multifront competition whose outcome will shape the international balance of power. Navigating these deeply intertwined domains of technological and strategic AI competition demands that Washington adopt a holistic strategy. Without it, success in one race could create vulnerabilities in another—and neglecting any of them risks irreparably weakening the United States’ global position.

Loading…