Huawei’s deputy chairman saying Chinese AI isn’t good enough yet is not the kind of statement you hear often. But that’s exactly what Eric Xu, Rotating Chairman at Huawei, said at the 2026 Connect Conference. According to Huawei Central, Xu stated plainly that Chinese AI technology is not currently advanced enough to manage the safety risks it creates, and that China needs to keep building more capable models to actually address those problems.
That’s a striking admission. China’s AI sector has been under intense pressure to prove it can compete with American labs, and the domestic narrative has leaned heavily on progress stories, from DeepSeek’s efficient reasoning models to Baidu’s Ernie updates. For a senior Huawei executive to stand up at a major conference and say the technology still has a long way to go on safety is a meaningful departure from that tone.
The timing matters too. AI safety has become a central concern globally, with OpenAI, Anthropic, and Google DeepMind all publishing safety research and building dedicated teams around alignment and risk evaluation. Anthropic has made safety its core brand identity. OpenAI has a safety board that has already faced public scrutiny. These are not small investments. So when Xu frames China’s gap not just as a capability problem but specifically as a safety problem, it suggests Huawei is watching that conversation closely and recognizes the gap is real.
For developers and founders building on Chinese AI infrastructure or evaluating models like Qwen, Doubao, or Huawei’s own PanGu, this is worth taking seriously. Safety tooling, red-teaming capacity, and alignment research are still relatively thin across Chinese AI providers compared to what’s available through Anthropic’s Claude API or even OpenAI’s moderation endpoints. That gap has practical consequences for anyone deploying in regulated industries or consumer-facing products.
Xu’s comments also carry a policy signal. Huawei has significant influence over how China’s AI ambitions get framed at the institutional level. Calling for more advanced model development specifically to solve safety problems positions safety as an argument for accelerating research, not slowing it down. That framing will likely show up in how Chinese AI policy gets justified in the months ahead. Whether it leads to meaningful safety infrastructure or just faster model releases is the real question.
