Huawei’s Eric Xu says China should keep pushing AI while watching frontier risks

Huawei rotating chairman Eric Xu has added a distinctly Chinese perspective to the global debate over AI safety, arguing that domestic developers should keep advancing more powerful models while watching the risks that leading U.S. laboratories say they are beginning to encounter. Reuters reported that Xu made the comments on September 17 at Huawei Connect in Shanghai, a major company event focused on AI infrastructure, chips, cloud services and enterprise technology.

Xu’s central point was that China’s AI industry should not slow down before it has reached the frontier capabilities where the most serious safety concerns are expected to emerge. According to Reuters, he said Chinese developers may not yet be advanced enough to experience the risks that U.S. labs discuss when they talk about highly capable models. At the same time, he did not dismiss safety work. His argument was closer to a sequencing claim: China should pursue stronger models, observe the risk picture carefully and avoid allowing fear of hypothetical frontier behavior to become a reason for falling further behind.

The comments matter because AI safety is increasingly entangled with industrial competition. U.S. companies such as OpenAI, Anthropic and Google DeepMind have been publishing more information about model behavior, evaluation regimes and risk frameworks as their systems become more capable. Chinese firms are trying to narrow the capability gap while operating under export controls that restrict access to the most advanced AI chips. Huawei sits at the center of that tension because it is both a major supplier of Chinese AI infrastructure and a symbol of Beijing’s push for technology self-reliance.

Huawei Connect also comes at a time when the market is watching China’s alternatives to Nvidia hardware. Huawei has been promoting its Ascend ecosystem and large-scale AI computing systems for domestic cloud providers, model companies and enterprise customers. The company’s AI infrastructure strategy is meant to reduce dependence on foreign accelerators while giving Chinese developers more predictable access to compute. Compute access is one reason Xu’s safety comments carry weight: if frontier risk depends partly on scale, then hardware availability shapes both capability progress and the timing of safety questions.

There is a strategic message behind the remarks. Western policy debates often start from the idea that frontier models may soon produce dangerous autonomy, cyber capabilities or loss-of-control scenarios. Xu’s framing suggests that for Chinese companies the immediate issue is still catching up, improving infrastructure and making models useful for business and industry. That does not remove the need for safety governance, but it changes which problems appear urgent.

The risk for China is that a purely catch-up mindset could underinvest in evaluations, incident reporting and alignment research until problems appear suddenly. The risk for the global debate is that safety becomes another geopolitical slogan rather than a shared technical discipline. A more useful path would recognize both realities: capability gaps are real, and so are the incentives that push every major lab to move faster.

Xu’s comments therefore show how AI safety has become inseparable from the infrastructure race. The same chips, cloud systems and developer platforms that help countries build stronger models also determine when they confront the hardest governance questions. China’s answer, at least from Huawei’s stage in Shanghai, is to keep building while preparing for the risks that may come with more capable systems.