The next phase of the US-China AI race may come down to chips, compute access, and how quickly Washington can close the gaps in its own rules. A new policy paper from Anthropic has thrust the US–China AI rivalry back into the spotlight as President Donald Trump visits China alongside several Silicon Valley heavyweights, including Elon Musk, Tim Cook, Jensen Huang, and Larry Fink, warning that decisions made today could determine who controls the next generation of artificial intelligence by 2028.

In its report titled “2028: Two scenarios for global AI leadership,” Anthropic lays out two different outcomes for the global AI race. In the first, the US and allied democracies successfully maintain their lead by tightening export controls, limiting what Anthropic described as “distillation attacks” by Chinese AI firms, and accelerating the global adoption of American AI technology. “In this world, democracies set the rules and norms around AI,” the company wrote. In the second scenario, Anthropic warned that if Washington fails to close loopholes around chip exports and remote access to computing infrastructure, Chinese AI firms could catch up or surpass US companies. “In this world, AI norms and rules are shaped by authoritarian regimes, and the best models enable automated repression at scale,” Anthropic wrote.

Chips and Compute Are at the Center of the Fight

A major focus of the paper was access to advanced chips and semiconductor manufacturing tools. Anthropic argued that China’s AI sector remains constrained primarily by limited access to cutting-edge compute resources rather than talent or research capabilities. The company claimed Chinese firms have stayed competitive by exploiting export-control loopholes, accessing overseas data centers, and using distillation techniques to mimic American AI models. The report pointed to companies including Huawei, Alibaba, and ByteDance as examples of Chinese firms advancing AI efforts despite US restrictions. Anthropic also cited reports that Chinese AI models, including those from DeepSeek, were trained on advanced US chips that are technically restricted from being sold to China. According to the paper, “distillation attacks” have become another major concern for US AI firms. Anthropic described the practice as Chinese labs creating fraudulent accounts to extract outputs from American models and replicate their capabilities at lower cost. The company said OpenAI, Google, Anthropic, and the Frontier Model Forum have all publicly criticized the practice.

AI Seen as Military and Political Tool

Anthropic repeatedly framed AI as both an economic engine and a national security technology. The company warned that frontier AI systems could reshape cyber warfare, military planning, and surveillance. It claimed the Chinese government already uses AI for censorship, monitoring, and cyber operations, and argued that more advanced systems could dramatically expand those capabilities. Anthropic also argued that a close US-China AI race could weaken safety standards because companies and governments may feel pressured to release increasingly capable systems faster. The report highlighted concerns about safety practices at Chinese AI labs, claiming only a small number publicly disclose safety testing results for high-risk areas such as chemical, biological, radiological, and nuclear threats. This lack of transparency, combined with the rapid pace of development, raises the risk of catastrophic accidents or misuse. Historical parallels can be drawn to the nuclear arms race, where both sides prioritized speed over safety, leading to near-disasters. In the AI context, the stakes are even higher because the technology can be weaponized or accidentally cause harm at a global scale. The US government has already established the AI Safety Institute, but Anthropic argues that more stringent international norms are needed to prevent an uncontrolled arms race.

A Call for Tougher US Policy

The paper arrives as debate intensifies in Washington over semiconductor export restrictions and America’s long-term AI strategy, including how much access Chinese firms should have to advanced Nvidia hardware. Anthropic urged US policymakers to tighten controls on advanced chips, crack down on chip smuggling and on overseas compute access, and expand efforts to promote the adoption of American AI systems worldwide. The company said the US currently has a rare opportunity to lock in a significant lead. “There is a high likelihood that we will look back on 2026 as the breakaway opportunity for American AI,” Anthropic wrote. Still, the company said it supports AI safety dialogue with Chinese experts where possible and stressed that its concerns are directed at the Chinese Communist Party rather than the Chinese people or broader AI research community.

Anthropic’s report builds on a growing body of research that examines the geopolitical implications of advanced AI. For example, a 2025 analysis by the Center for Security and Emerging Technology (CSET) found that China has already surpassed the US in the number of AI research papers published annually, though the US still leads in high-impact patents and foundational model development. Another study by the RAND Corporation highlighted that China’s AI progress in military applications, such as autonomous drones and cyberattacks, could erode the US technological edge within five years if current trends continue. These findings underscore why Anthropic’s call for action is being taken seriously by policymakers across the political spectrum. The Biden administration’s chip export rules, updated in 2024, have slowed but not halted Chinese access to advanced semiconductors. Smuggling networks have been documented routing chips through third countries, a challenge that Anthropic says requires stronger enforcement and international cooperation. The company also emphasizes that export controls alone are not enough; the US must invest in building a robust domestic semiconductor industry, as evidenced by the CHIPS Act, and in training the next generation of AI talent.

The debate over distillation attacks highlights a unique vulnerability of the open AI ecosystem. While US companies like OpenAI and Google have closed certain proprietary models, many foundation models are still accessible via APIs or through open-source releases. Chinese researchers can use these interfaces to extract training data or model weights, then reuse them to bootstrap their own systems. Anthropic proposes that companies implement more rigorous user verification, limit API call rates, and employ advanced detection techniques to identify suspicious activity. Additionally, international agreements could establish norms against such practices, similar to how the World Trade Organization governs intellectual property. However, critics argue that heavy-handed restrictions could stifle collaboration and slow down global AI progress. Proponents counter that leadership in AI is a zero-sum game when it comes to national security, and that democratic values must be protected. The outcome of this policy tug-of-war will likely shape not only the future of AI but also the balance of power in the 21st century.

Expanded Analysis: The Stakes of 2026

Anthropic’s warning that 2026 is a decisive window is rooted in several technological and economic factors. First, the next generation of AI chips, such as Nvidia’s B200 and beyond, will offer exponentially more computational power. Who gets access to these chips first will determine the speed and scale of model training. Second, the accumulation of data and user feedback loops means that the leading AI platforms become harder to catch up with over time. Third, the development of AI safety infrastructure, including red-teaming, audits, and regulatory frameworks, is still in its infancy. If China or the US moves faster on capabilities without parallel safety investments, the other may feel forced to do the same, creating a race to the bottom. Anthropic’s own track record, including its recent surpassing of OpenAI in enterprise AI adoption according to Neuron AI, adds credibility to its analysis. The company has a vested interest in a regulatory environment that favors responsible innovation, but its arguments are supported by independent data on chip flows and patent filings.

In conclusion, Anthropic’s paper ultimately frames 2026 as a narrowing window for US policymakers. Whether that view proves prescient or self-interested, the company argues that decisions made now could shape not only who builds the strongest AI systems but also whose rules govern how they are used. The global community now watches to see if the US will rise to the challenge or allow the opportunity to slip away.

Source: eWEEK News