AI lead – Anthropic argues the US could secure a 1–2 year AI edge over China, if chip export controls and anti-distillation enforcement move fast.
A narrow window is opening for the United States to extend its edge in frontier AI against China, and it may close sooner than many policymakers expect.
Anthropic said the US could lock in a one- to two-year lead in frontier capabilities over China. but only if it acts quickly.. In a detailed post. the company laid out two potential paths for the AI landscape by 2028—one in which the US and its allies restrict China’s access to American AI compute. and another in which they do not.
At the center of Anthropic’s argument are two drivers of China’s progress in AI. at least as it characterizes the competition.. The company pointed to “loose controls” on chip exports. and also to techniques it described as “distillation attacks. ” where developers train smaller “student” models using outputs or knowledge generated by a more capable model.
Anthropic said that addressing both issues could make it possible to secure a 12–24 month lead in frontier capabilities. It also warned that the opportunity to build and preserve that advantage is not guaranteed to last.
The company linked the race for capability to a broader stakes question: AI safety and governance.. Anthropic argued that a “neck-and-neck race” between US and Chinese AI labs could make efforts to test. regulate. and govern advanced systems harder for both governments and the private sector—because intense competition tends to increase pressure to release faster. sometimes before safety measures are fully validated.
To reduce that risk, Anthropic called for policy steps designed to tighten the flow of advanced hardware and to limit workarounds. It urged stronger chip export controls, higher enforcement budgets, and action to stop distillation attacks by Chinese AI labs.
Earlier this year, Anthropic also raised concerns about certain Chinese companies using its Claude technology.. In February. the company said three of China’s biggest AI firms—DeepSeek. MiniMax. and Moonshot AI—were “illicitly” using Claude to develop their own models. framing the issue as one involving unauthorized use rather than ordinary competition.
The policy backdrop for Anthropic’s recommendations stretches across multiple US administrations.. The Biden administration first imposed export controls on American chips to China in 2022.. Under the Trump administration, those rules were expanded further, including prohibitions on Nvidia and AMD selling chips to China.
The chip rules, however, have not been static.. Last August. the Trump administration partially backed away from the earlier prohibitions by allowing Nvidia to sell its H200 chips if the company paid a 25% levy on those sales.. Anthropic’s new remarks land against that evolving enforcement reality, emphasizing that gaps can be exploited.
Enforcement has also remained a live issue.. In December. the US charged several parties over attempts to smuggle Nvidia’s most advanced chips by relabeling them as “SANDKYAN” chips.. Those allegations underscore the practical challenge of preventing unauthorized access, even when official controls exist.
Meanwhile, the picture is contested by observers who say China’s position may not actually be improving as quickly as the competitive framing suggests. In April, an ex-ByteDance engineer, Zhang Chi, said in a podcast interview that China was falling further behind rather than closing the gap.
Zhang’s argument centered on constraints he attributed to China’s AI development. He said Chinese AI lacks high-quality data to train models and also lacks access to advanced chips—two factors that, if accurate, would affect the pace at which performance can scale.
Anthropic’s post arrived on the same day as President Donald Trump met with Chinese leader Xi Jinping. the first visit to China by a US president since Trump’s 2017 trip.. The meeting included a business delegation featuring US corporate leaders such as Tesla CEO Elon Musk. Apple CEO Tim Cook. and Nvidia CEO Jensen Huang. reflecting how deeply technology and trade are intertwined at the highest level.
For investors and companies watching the AI supply chain. these developments point to a key risk: rules that govern chips and compute are now shaping strategic outcomes as much as algorithms do.. When export controls tighten. the competitive advantage can tilt quickly toward labs with stable access to hardware and with the ability to comply. adapt. or contest enforcement.
The dispute over whether China is closing—or widening—the capability gap also suggests that “progress” in AI is not one-dimensional.. Even if training techniques can be leveraged through methods like distillation. meaningful gains still depend on inputs such as data quality and compute access.. That means policy choices around hardware. combined with enforcement capacity. can influence both the technical tempo of model development and the long-term competitive trajectory.
There is also a governance angle that extends beyond the US-China contest.. Anthropic’s warning about safety testing under deadline pressure highlights a broader structural tension in frontier AI: as competition intensifies. the incentive to move quickly can collide with the time-consuming work of evaluation. monitoring. and risk reduction.. In that environment. how governments calibrate controls and how companies demonstrate safety discipline may end up determining not just who leads. but how safely leadership is exercised.
Finally. the existence of multiple policy pathways described by Anthropic—ranging from tighter restriction of compute access to more permissive access—illustrates that the next stage of AI competition is not inevitable.. It is. at least in part. shaped by decisions about export rules. enforcement funding. and the practical ability to deter workarounds such as those involving distillation strategies and attempted chip rerouting.
Anthropic AI lead chip export controls frontier AI China US AI distillation attacks Nvidia H200