Editor’s Note: The AI competition is evolving from a battle over model capabilities to a more complex, systemic rivalry.

This article discusses Anthropic’s latest assessment of the U.S.-China AI competition. The author argues that the next two to three years will be a critical window for shaping the future AI landscape: on one hand, the U.S. and its allies still maintain advantages in advanced chips, model capabilities, capital investment, and the global technology stack; on the other hand, Chinese AI labs are steadily closing the gap through their talent, data, engineering efficiency, and rapid advancement.

Based on this, Anthropic believes its current core mission is to maintain its lead in computing power and model capabilities. This includes continuing to strengthen export controls on advanced chips and restricting technology spillover pathways such as overseas data center deployments, chip transfers, and model distillation. Otherwise, Chinese AI companies may narrow the gap with U.S. frontier models by 2028 through access to computing power and replication of model capabilities.

This article also presents a broader industry insight: as AI enters a phase of accelerated capability, the focus of competition is no longer just on “who has the strongest model,” but on who can transform model capabilities into infrastructure, industrial efficiency, global markets, and governance frameworks. The closer AI technology comes to achieving general capabilities, the more critical its underlying chip supply chains, capital investments, policy tools, and global distribution networks become as decisive factors shaping the future landscape.

The following is the original text:

We have published a new paper outlining our perspective on the U.S.-China AI competition.

The United States and its allies must maintain a lead over major competitors such as China in the field of AI. As AI capabilities advance rapidly, this technology will soon profoundly impact social governance, national security, and the global balance of power. Meanwhile, the pace of AI development is accelerating, leaving little time for all parties to establish competitive rules, manage technological risks, and shape global governance frameworks. It is against this backdrop that we propose the measures necessary to ensure the United States maintains its leadership.

One of the most critical factors in developing AI is access to computing chips—known as "computing power"—used to train models. Since the most advanced chips are primarily developed by companies within the United States and its allied systems, the U.S. government currently restricts China’s access to these chips through export controls. Recent experience shows that these controls have had a clear impact. In fact, China’s AI labs have been able to develop models approaching U.S. standards largely by leveraging their talent advantage, exploiting loopholes in export controls, and engaging in large-scale model distillation—extracting outputs and capabilities from U.S. models to rapidly replicate certain technological achievements.

In this article, we present two scenarios for how the world might look in 2028. We anticipate that transformative AI systems will already have emerged by then.

In the first scenario, the United States successfully maintains its computational advantage. Policymakers further tighten export controls, reducing China’s ability to acquire U.S. frontier capabilities through methods such as model distillation, while accelerating the adoption of AI by the United States and its allies. In this world, the U.S.-led technology ecosystem is better positioned to shape the rules, standards, and governance frameworks for AI. It is also under this scenario that the United States is more likely to engage in effective communication with China on AI safety—we support such efforts where feasible.

In the second scenario, the United States did not take sufficient action. Policymakers failed to close the channels through which China gains access to advanced computing power, allowing Chinese AI companies to quickly fill these gaps, catch up to the forefront of AI, and even surpass it in some areas. In this world, the rules and standards of AI will be contested by a broader range of nations, and the most advanced models may be deployed for large-scale social governance, cyber operations, and security capabilities. Even if this scenario is built upon the foundation of U.S. computing power and technology spillover, it does not serve the long-term interests of the United States and its allies.

The United States and its allies entered the AI competition with a significant advantage. The key tools required for AI dominance have been established by highly innovative corporate ecosystems within the U.S. and its allied systems. Past success means that the most critical task today is largely to avoid squandering this existing advantage: do not make it easier for China to catch up.

Two scenarios for U.S.-China AI competition in 2028

The development and deployment of AI will determine the future direction of global technological rules, industry standards, and governance frameworks. Those who maintain leadership in AI are more likely to shape how these systems operate.

Currently, the United States and its allies hold a significant lead in computing power, one of the most critical factors in developing cutting-edge AI models. This advantage stems from technological innovation by the U.S. and its allies, as well as bipartisan U.S. export control policies. However, Chinese AI labs are no longer far behind in terms of model intelligence. Our focus on China’s AI development is not meant to deny the capabilities or contributions of the Chinese people and China’s AI community, but rather because China is the only country besides the U.S. with sufficient resources, top-tier talent, and a systematic effort to catch up in frontier AI.

China has applied AI technologies in areas such as information censorship, social governance, cybersecurity, and military capability development. Chinese AI labs possess world-class talent. What truly limits their continued advancement is computational constraints. Chinese labs have been able to remain close to the cutting edge in part by exploiting loopholes in U.S. export control policies and using large-model distillation to acquire partial capabilities from U.S. models, thereby accelerating their own model training and capability progression.

As computing power supply expands rapidly, AI is increasingly being used to enhance the training of new models, and we are entering a period of accelerated AI capabilities. The so-called "nation of geniuses inside data centers"—our understanding of transformative AI intelligence—may now be within reach. This acceleration makes policy action more urgent.

To date, China’s AI ecosystem has continued to closely follow the cutting edge due to ongoing issues such as export control evasion and model distillation. However, if the United States and its allies now take action to simultaneously address both access to computing power and the spillover of model capabilities, it is still possible to secure a 12- to 24-month lead in frontier capabilities. By 2028, such an advantage would hold significant strategic importance. This lead would also enhance the United States’ ability to engage with Chinese AI experts on AI safety and governance—an engagement we support. But the window of opportunity to secure this lead will not remain open indefinitely.

Here, we present two possible scenarios for the state of U.S.-China AI competition in 2028. In the first scenario, the United States and its allies establish a significant lead in model intelligence, application adoption, and global distribution. This scenario could be realized if policymakers act now to tighten restrictions on advanced computing power access for Chinese labs, reduce their ability to catch up by distilling the best U.S. AI models, and accelerate AI adoption by the United States and its allies.

The second scenario is that China remains competitive at the cutting edge. This would occur if policymakers fail to build on their existing advantages or relax restrictions on Chinese companies’ access to advanced computing power.

Many in the U.S. Congress and the Trump administration already support export controls, curbing model distillation attacks, and promoting the overseas expansion of the U.S. AI technology stack. As these policies advance, we hope the United States and its allies can secure a significant lead by 2028, avoiding a highly competitive race with China just two years from now.

The necessity of staying ahead

We anticipate that cutting-edge AI will have a profound impact on the economy and society in the coming years, as described in "Machines of Loving Grace" and "The Adolescence of Technology." Our mission is to ensure that humanity navigates the transition to transformative AI safely and beneficially. We believe that a successful transition will lead to major breakthroughs in medicine, invention, and economic growth.

Whether this transition proceeds smoothly depends in part on which technological ecosystems first build the most powerful systems. The industrial structure, regulatory environment, and governance frameworks surrounding the most advanced AI will shape the rules for developing and deploying this technology. In turn, these rules will influence whether the technology is secure, whose security it protects, and whose interests it ultimately serves.

If the frontier of AI is primarily shaped by systems that use it for military advantage, cyber operations, social governance, and information control, this technological transformation will face greater uncertainty and security risks.

Historically, large-scale governance and surveillance capabilities have been limited by the human costs of enforcement. Powerful AI systems may reduce these costs, enabling automated governance, identification, and decision-making at a much larger scale. Therefore, China’s leadership in AI could have significant implications for global AI governance and security dynamics.

China possesses vast economic, military, and national governance resources. It is also the only country, apart from the United States, with a well-resourced, highly talented AI laboratory that is rapidly catching up to the cutting edge. Moreover, China places great emphasis on positioning itself as a global leader in AI. Beijing has already invested tens of billions of dollars into China’s AI and semiconductor industries.

China has already deployed AI systems in areas such as information censorship, social governance, cyber operations, and security capability building. The implementation of related technologies in certain regions—including facial recognition, biometric data collection, and communication surveillance—demonstrates AI’s potential for large-scale governance. Advanced AI systems will enable these capabilities to be maintained at lower costs, with broader coverage and higher levels of automation. As these technologies spread overseas, AI may also be adopted by more countries to strengthen governance and surveillance capabilities. The AI frontier led by China could significantly transform global patterns of technology use and governance models.

Frontier AI will shape the future balance of military power. China has recognized AI as a critical variable in future battlefields and is advancing the intelligent transformation of its military systems. Chinese military strategists view the "intelligence" of military capabilities as a vital pathway to catch up with and ultimately enhance their own military strength. The Chinese military has already begun procuring commercially developed AI systems from Chinese companies for military use, including deploying DeepSeek models to coordinate unmanned vehicle swarms and enhancing cyber operations capabilities.

These capabilities do not spread slowly. When a new model reaches a new level of capability in areas such as autonomous targeting, vulnerability discovery, or cluster coordination, the party that masters it can deploy it in practice within weeks, not years.

Risks will further compound as frontier AI becomes an accelerator for other critical technologies. Advanced AI models will be able to compress R&D cycles in semiconductors, biotechnology, and advanced materials. Leadership in frontier AI will enable a nation to continuously expand its advantage across its entire national security technology stack.

If a Chinese AI lab develops a model reaching the level of Claude Mythos Preview before a U.S. lab, China will be the first to gain a system capable of autonomously discovering and linking software vulnerabilities, potentially using it to further enhance its cyber capabilities. As future models' abilities will increase exponentially, their impact on the security interests of the United States and other nations will grow significantly.

The competitive parity between AI labs in the U.S. and China may make it more difficult for industry- and government-led efforts on safety and governance. If Chinese labs closely follow or match the capabilities of U.S. models, both American and Chinese private AI companies may feel increased pressure to release new models and products more quickly, without completing adequate safety evaluations. Governments around the world may also be reluctant to implement policies that encourage responsible AI development and deployment, fearing they will fall behind.

Although an increasing number of researchers in China’s AI labs and policy circles are beginning to focus on AI safety risks, this trend has not yet translated into safety practices comparable to those in U.S. labs. As of last year, only three of China’s 13 top AI labs had published safety evaluation results, and none had disclosed assessments of chemical, biological, radiological, and nuclear (CBRN) risks. The Center for AI Standards and Innovation (CAISI) found that under a common jailbreak technique, DeepSeek’s R1-0528 model responded to 94% of clearly malicious requests, compared to just 8% for U.S. reference models. This pattern continues in recently released models. For example, an independent evaluation of Moonshot’s Kimi K2.5, released in April this year, found that the model failed to reject CBRN-related requests at a higher rate than leading U.S. models.

More seriously, Chinese laboratories frequently release models with dual-use capabilities in open-weight formats. Once a model’s weights are made open, existing safety safeguards may be removed, allowing any state or non-state actor to use the model for malicious purposes—including cyberattacks and CBRN abuse—precisely the threats these safeguards were designed to prevent.

Our policy objective: to create and maintain a leading advantage for the United States and its allies.

We support policies by the United States and other nations to establish and maintain a secure, near-term lead over China in intelligent capabilities, domestic adoption, and global distribution. This lead is critical to safeguarding the national security interests of the United States and its allies and preventing the misuse of AI technologies. It is also a fundamental prerequisite for ensuring that the United States and its allies can secure a favorable position in future global AI governance.

Anthropic deeply respects the Chinese people and the achievements of China's AI community. We hope for peaceful relations between China and the rest of the world. Our concerns specifically focus on the risks that any powerful national system might pose to global security and governance after acquiring advanced AI systems.

Where feasible, Anthropic supports international AI safety dialogues with Chinese AI experts. The world shares a common interest in safe AI, regardless of where it is developed or deployed. Frontier AI systems may pose a range of risks that require communication between the United States and China. Identifying shared challenges and advancing ideas to prepare for and mitigate these risks serves the mutual interest of both sides.

The prospects for constructive engagement are best when the United States maintains a significant capability advantage. Leading responsibly in the development and deployment of advanced AI will enhance the United States’ ability to influence AI safety practices in China and other regions.

Mythos Preview is a model we released to select partners in April this year as part of Project Glasswing. It indicates that a period of capability acceleration has arrived, making policy action even more urgent. After gaining access to this model, Firefox patched more security vulnerabilities last month than it did throughout the entire year of 2025—nearly twenty times its average monthly patch rate for 2025. In response to this model, a Chinese cybersecurity analyst wrote that China is “still sharpening its knife, while the other side has suddenly set up a fully automatic Gatling gun.”

Cutting-edge AI capabilities will rapidly approach the transformative AI vision of a "genius nation inside the data center." This acceleration will be driven by the logic of scaling laws: as compute and data inputs increase, model performance improves predictably; meanwhile, AI itself is increasingly used to accelerate the development of new models.

We will likely look back and view 2026 as the window of opportunity for the United States to achieve a breakthrough lead in AI. U.S. labs possess the most advanced AI models, with a significant lead in both the quantity and quality of cutting-edge AI chips required to push the frontier, along with substantial capital advantages from revenue and funding sufficient to support related investments. Chinese labs do have real strengths: world-class innovation talent, abundant and low-cost energy, and vast amounts of data—all essential conditions for developing frontier intelligence. However, they lack sufficient domestic computing power to compete, and insufficient revenue and capital to finance this competition.

Four Fronts of Competition

The United States and China are competing for strategic advantage in cutting-edge technologies such as AI. Public statements from Beijing and Washington both reflect this assessment. Describing this competition as a “race” may create a misleading impression—as if there is a finish line that, once crossed, would secure an absolute victory for one side. In reality, this will be an ongoing struggle for advantage. The future—whether democratic or non-democratic nations will more successfully shape the values, rules, and norms of the AI era—depends on the outcome of this long-term competition.

This competition is unfolding on four fronts:

Smart capabilities: Which countries are capable of developing the most powerful AI models.
Domestic adoption: Which countries are most effective at integrating AI into the business and public sectors.
Global distribution: Which countries can deploy the AI technology stack that powers the global economy.
Resilience: Which countries can maintain political stability during economic transition.

Among these four fronts, intelligent capability is the most critical. We expect frontier model capabilities to have the most profound impact on geopolitical competition. Model capability is also the core driver of market adoption and global distribution.

But intelligence alone is not enough. If China can more quickly and effectively integrate near-state-of-the-art AI systems into its economic and security frameworks, and promote the global adoption of low-cost, subsidized AI, it could gain advantages sufficient to offset any gaps in model intelligence. Beijing’s “AI+” initiative and its emphasis on “embodied intelligence” reflect a strategic policy focus on embedding cutting-edge intelligence into its economic and national systems. Similarly, the Trump administration’s AI action plan, with its emphasis on “promoting the export of U.S. AI technology stacks,” underscores the strategic advantage of driving global adoption.

Although this article will not focus extensively on the front of resilience, we believe it will become a critical aspect of the AI competition. Maintaining stability, cohesion, and strong policy-making capacity during this period will be a key advantage; conversely, for countries unable to achieve this, it will become a vulnerability.

Compute—the advanced semiconductors required to train and deploy cutting-edge AI—is a critical input across all of the aforementioned competitive fronts. The global race for AI leadership is, to a large extent, a race for compute. Over the past decade, model capabilities have consistently improved alongside increases in compute scale, with the majority of historical gains in AI performance driven primarily by scaling up compute resources.

In addition, computing power is not only used to train new models but also to support users in utilizing AI, known as "inference." Whether training the most advanced models or deploying them in commercial and national security applications, computing power is essential. While top talent, vast amounts of data, and key algorithmic breakthroughs are also crucial to the intelligence race, without sufficient computing power, these investments cannot truly deliver their potential.

Currently, democratic nations are winning the race for computing power leadership. Some worry that export controls may accelerate China’s efforts to develop a domestic advanced chip supply chain, but there is little evidence that China’s self-reliance initiatives can challenge the leadership of the United States and its allies in advanced computing technologies. Even before the implementation of export controls, Beijing had already invested heavily in China’s semiconductor industry and launched major industrial policies such as "Made in China 2025" and the National Integrated Circuit Industry Investment Fund. Despite these state-backed investments, Chinese AI labs and chip manufacturers remain constrained by export controls on advanced chips and semiconductor manufacturing equipment imposed by the United States and its allies.

As a result, the gap in computing power appears to be widening. An analysis of Huawei’s and NVIDIA’s product roadmaps found that, in terms of total processing performance, Huawei will only be able to produce products equivalent to 4% of NVIDIA’s total computing power in 2026, dropping to 2% by 2027. More importantly, NVIDIA is only one part of the computing ecosystem of the United States and its allies. Google and Amazon are also accelerating the production of their own chips—TPUs and Trainium—to meet the demands of U.S. cutting-edge AI labs and their customers.

Compounding China’s computing power shortage is its limited progress in several of the most technologically complex segments of the semiconductor supply chain. Without access to extreme ultraviolet (EUV) lithography—and especially as policymakers further close loopholes surrounding deep ultraviolet (DUV) lithography and its maintenance services—Chinese chip manufacturers will struggle to produce sufficient quantities and quality of chips to challenge U.S. leadership in computing power. China’s inability to mass-produce high-bandwidth memory further widens this gap. One study estimates that if the U.S. tightens restrictions on China’s access to American computing capabilities, the U.S. could possess roughly 11 times the computing power available to China’s AI industry.

The leading computing power primarily stems from two reasons.

The first reason is the ongoing innovation by companies such as NVIDIA, AMD, Micron, TSMC, Samsung, and ASML in democratic economies including the United States, Japan, South Korea, Taiwan, and the Netherlands. These companies together have developed the unique technologies required for the world’s most advanced semiconductors. Without these engineering breakthroughs and decades of sustained R&D investment, today’s AI achievements would not be possible.

The second reason is that the past three U.S. administrations have taken forward-looking and decisive policy actions. Bipartisan policy measures have protected America’s and its allies’ innovation engine by restricting Chinese-controlled companies’ access to the U.S. AI technology stack. Our CEO has publicly emphasized the importance of export controls. Over the past several years, these controls have limited the sale of cutting-edge AI chips and semiconductor manufacturing equipment to China, constraining China’s advancement in frontier AI despite Beijing’s substantial state investments in the sector. Without these restrictions on China’s access to U.S. computing power, China might have had all the conditions necessary to develop AI systems equal to or even superior to those of the United States.

Some observers worry that restricting access to computing power will force Chinese AI labs to innovate in other directions, thereby eroding America’s lead. Chinese labs are indeed innovating, but so far, these innovations have not been sufficient to close their computing gap. Algorithmic improvements are both a function of computing power and a multiplier of it, not a substitute. Discovering these algorithmic advances is itself a process highly dependent on computing power: more computing power enables labs to run more experiments, leading to the discovery of more algorithmic improvements. As frontier models increasingly participate in AI research, this cycle will tighten further, with frontier models helping to build their own successors. In short, a computing advantage will further translate into an algorithmic advantage, and ultimately into lasting leadership in AI.

Currently, U.S. frontier systems are estimated to be at least several months ahead of China’s top models in terms of intelligence, though such estimates inevitably involve uncertainty. Although China’s open-weight models have garnered significant attention, they still lag behind proprietary frontier models in enterprise adoption, and public market investors have begun to focus on their commercialization challenges. Additionally, Chinese AI labs appear to be moving away from open-source approaches, opting instead to keep their best models proprietary.

Leaders in China’s AI sector have also acknowledged the impact of export controls and the critical need for U.S. chips. Executives from China’s top AI labs have expressed concerns that China risks falling further behind due to compute constraints. Leading Chinese labs have identified compute scarcity as the primary bottleneck to accelerating model capabilities and attribute this constraint to export controls. A senior executive at a major Chinese cloud provider stated that supplying U.S. chips subject to export controls to China would have a “huge, really huge” impact, adding that any supply gap would severely hinder China’s AI development; he also dismissed concerns that importing U.S. chips would slow China’s self-reliance efforts. The main voices within China arguing that “export controls are ineffective” appear to originate largely from official statements and state media, likely aimed at influencing U.S. policymakers.

Although export controls have been effective in shaping the current advantage, their impact remains insufficient. Despite China's inability to manufacture enough advanced chips domestically or legally purchase them overseas, Chinese AI labs continue to maintain near-cutting-edge capabilities in model intelligence through two workarounds.

The first approach involves circumventing computing power acquisition, such as smuggling AI chips directly into China or accessing overseas data centers. The second approach involves illegal model access, namely conducting distillation attacks on cutting-edge U.S. models and using these models as tools to accelerate domestic AI development.

It is an open secret that China is circumventing U.S. export controls. For example, U.S. federal prosecutors have accused a co-founder of Supermicro and two others of transferring servers containing advanced U.S. chips worth $2.5 billion to China. According to U.S. government sources and media reports, DeepSeek used advanced U.S. chips—prohibited from sale to China—to train its latest model. The Financial Times reported that Alibaba and ByteDance are now using export-controlled U.S. chips in data centers located in Southeast Asia to train their flagship models. Current controls fail to cover this pathway because U.S. export laws primarily regulate chip sales, not remote access to chips. The U.S. export control system is struggling to address how Chinese AI labs are gaining access to advanced U.S. computing power.

Distillation attacks are another tactic used to catch up with U.S. counterparts and mitigate the impact of export controls. In this practice, Chinese laboratories create large numbers of fake accounts to bypass access controls on U.S. AI models and systematically collect their outputs to replicate cutting-edge capabilities. This enables these laboratories to free-ride on decades of foundational research, billions of dollars in investment, and the collective efforts of the world’s top engineers behind these advanced models. As a result, China can acquire near-frontier capabilities at an extremely low cost—effectively subsidized by the United States. From the perspective of long-term national security interests, this amounts to systematic industrial espionage targeting critical technologies. OpenAI, Google, Anthropic, and the Frontier Model Forum have all publicly condemned distillation attacks.

Chinese AI experts have also publicly acknowledged the scale and significance of model distillation attacks on China’s AI development. A recent article in state media described distillation attacks against U.S. models as a "backdoor" relied upon by Chinese AI labs, calling it a core component of their business model. A former ByteDance researcher stated that Chinese AI labs use distillation as a shortcut to train models, avoiding the investment required to build their own data pipelines.

U.S. policymakers have swiftly taken action to address this threat. The White House Office of Science and Technology Policy issued a memorandum on distillation attacks. Senior officials from the White House, the U.S. Department of Defense, and members of Congress have also expressed concern about this issue. Recently, related legislation introduced by the U.S. House Committee on Foreign Affairs to counter distillation attacks has been unanimously approved within the committee.

If policymakers in the United States and its allies can close off these two channels supporting the development of China’s AI models—circumventive access to computing power and illegal model access—we may have a rare opportunity to secure a lasting lead.

Two scenarios for 2028

Below, we describe two hypothetical future scenarios to illustrate how policy actions taken today will shape the competitive landscape in 2028.

The United States maintains a solid advantage in computing power. Although China has increased state support for its semiconductor industry, Chinese chip manufacturers still lag behind the United States and its allies by several years, partly due to their inability to access advanced semiconductor manufacturing equipment, related services, and maintenance. As the U.S. and its allies ramp up their chip manufacturing capabilities, and as leading chipmakers continue developing more efficient and higher-performance chips, the computing power gap between China and the United States is widening.

Meanwhile, U.S. policymakers are taking action to close loopholes in America’s economic security tools. With increased enforcement resources, efforts to smuggle chips into China or access controlled chips via overseas data centers are becoming increasingly difficult.

Therefore, U.S. AI models are 12 to 24 months ahead in intelligent capability, and this lead is continuing to grow. A small number of AI labs, all located in the United States, sit at the forefront with the most intelligent, powerful, and high-performing models. The "nation of geniuses in data centers" has already become a reality in critical industries such as cybersecurity, finance, healthcare, and life sciences.

When U.S. frontier labs release a new model in 2028 that achieves a step-change improvement in capability—similar in relative impact to the Mythos Preview in April 2026—China may not attain comparable AI capabilities until 2029 or 2030. This would provide democratic nations with a critical buffer period to establish rules and norms for frontier AI systems.

U.S. AI has become the infrastructure of the global economy, driving new economic and scientific vitality. The Trump administration’s efforts to promote domestic AI adoption and boost U.S. AI exports have succeeded, leading to widespread adoption of powerful AI systems both domestically and internationally, with the resulting benefits fueling unprecedented economic growth and technological advancement. Global adoption of U.S. AI has risen sharply. The democratic nations’ lead in capability and computing power means Chinese AI companies struggle to compete for global market share outside a few niche markets. The world’s leading frontier AI systems are shaped by democratic values, making it harder for certain countries to exploit AI systems to infringe upon rights and civil liberties.

Cybersecurity and other national security advantages continue to expand. Cybersecurity personnel in both the public and private sectors use advanced AI systems to reduce the attack surface of the United States and other democratic nations, and to undermine China’s ability to gain and maintain a network foothold in critical systems, thereby enhancing the security of national security assets, intellectual property, and communication networks. America’s overwhelming AI advantage has also become a crucial force in deterring external threats.

A self-reinforcing cycle will further solidify the leadership of democratic nations. Overwhelming AI advantages make the United States and its allies more attractive partners. This alliance expands the market for U.S. AI and broadens the coalition shaping global AI standards. In turn, this fosters the development and deployment of AI systems that are secure, reliable, and protective of civil liberties. The world’s top technological and scientific talent continues to flow toward centers of cutting-edge innovation. As a result, the United States gains significant leverage to promote cooperation with Beijing on critical issues such as AI governance, strategic competition, and trade.

This cycle will continuously reinforce itself: an initial advantage strengthens the alliance, and the alliance, in turn, strengthens the advantage; the international order led by democratic nations will also be anchored during the transition to transformative AI.

AI developed and deployed in China is approaching the state of the art in model intelligence. Despite weaker semiconductor production capabilities, Chinese AI labs have trained models that are only months behind their U.S. counterparts. Ongoing model distillation attacks, access to overseas computing power, weak enforcement of semiconductor manufacturing equipment export controls, and easing of U.S. semiconductor export restrictions have all aided China’s追赶. Continuous access to cutting-edge U.S. AI for AI research has enabled Chinese AI labs to narrow the gap and approach their U.S. peers.

Adoption at both commercial and national levels is advancing rapidly. Beijing is driving nationwide domestic adoption through its "AI+" policy. Even though Chinese AI models are slightly less capable than their U.S. counterparts, China’s efforts to promote adoption have already yielded results. As a result, China can more advantageously deploy near-cutting-edge AI capabilities in economic, military, and technological domains, shifting the balance of power in China’s favor.

China’s AI-enabled cyber capabilities have become a serious threat. China has integrated AI-enabled cyber capabilities into its already highly advanced cyber power structure, enabling its military to remain a formidable cyber competitor. Relevant cyber actors have gained increased access to critical infrastructure and dual-use infrastructure in the United States and most countries worldwide, allowing them to disrupt essential national security and societal functions. As AI becomes more deeply embedded in the most critical systems, even democratic nations that pioneered this technology cannot gain an advantage over China in AI security.

Beijing is gaining global adoption through cost advantages and flexibility in local deployment. Huawei and Alibaba’s data centers are widely present globally, especially in low-cost markets in the Global South, though not limited to these regions. These data centers scale using older chips, and China is able to export these chips because its domestic market meets demand through licensed purchases of U.S. chips, smuggled chips, or remote access to overseas data centers. These data centers host secondary models developed by Chinese labs—while not among the most advanced, they are significantly cheaper and still highly effective.

Similar to Huawei’s past strategy of “affordable and sufficient,” China’s near-cutting-edge models and hardware are supporting a significant and rapidly growing segment of the global economy. This infrastructure advantage will give China substantial influence in related markets.

To ensure the outcome aligns with the first scenario, we support the following policy action directions.

Close the loopholes: smuggling chips, accessing overseas data centers, and semiconductor manufacturing equipment.
Currently, Chinese laboratories are acquiring U.S. chips subject to export controls through smuggling and overseas data centers, while gaps in the control of semiconductor manufacturing equipment are accelerating their efforts toward self-reliance. Tightening restrictions and increasing enforcement budgets would help close these vulnerabilities that support China’s AI ecosystem. This would lower China’s computational capacity ceiling and consequently slow its AI progress, thereby maintaining and expanding the democratic nations’ lead in AI. It is important to note that a lower computational ceiling could also substantially weaken distillation attacks, as Chinese AI laboratories still need to reach a minimum computational threshold to effectively conduct illegal distillation.

Protecting our innovation: Restricting model access to deter model distillation attacks.
U.S. policymakers in Congress and the executive branch can continue supporting policy actions that penalize and deter distillation attacks originating from Chinese laboratories, while also implementing measures to help U.S. laboratories detect and prevent such attacks. These measures could include enacting legislation that explicitly classifies distillation attacks as illegal, and promoting threat intelligence and technology sharing among U.S. peer laboratories and between laboratories and the U.S. government. Curbing these activities can substantially extend the democratic nations’ lead over the coming months and years.

Promote U.S. AI exports.
As global public and private sectors increasingly adopt AI, the Trump administration should continue promoting the global adoption of trustworthy AI hardware and models developed and shaped by democratic principles. Securing trustworthy U.S. infrastructure now can prevent China’s AI ecosystem from gaining the global foothold it needs to compete on cost and adoption in the future.

Conclusion

The United States and its allies have developed the world’s most advanced AI models and control the globe’s most sophisticated AI inputs, providing a significant advantage. This advantage can continue to grow if we maintain priority access to these technologies—but it will be lost if these technologies are handed directly to competitors. The decisions made by policymakers this year will determine the future of transformative AI. We support those committed to ensuring the United States and its democratic allies remain ahead by 2028.