China’s push to develop its own AI chips has gained momentum with announcements from Alibaba and Huawei, strengthening President Xi Jinping’s position as he meets with President Donald Trump for a multi-day state visit to Washington.
The new technology releases demonstrate China’s determination to reduce its reliance on U.S. semiconductor technology, and to overcome U.S. restrictions on advanced AI chips and semiconductor manufacturing equipment. Yet significant questions remain about whether Chinese chips can match the performance of NVIDIA’s leading GPUs.
“China’s chip self-sufficiency push is a bluff backed by half-strength silicon,” said Brendan Burke, Research Director at The Futurum Group. “Huawei cut back its SuperPoD and still delivers about half of NVIDIA’s compute per chip. Refusing H200s doesn’t close that gap. It hands the lead to U.S. labs. Beijing isn’t beating export controls. It’s absorbing them and calling it strategy.”
Still, Alibaba recently unveiled what it touts as China’s “most powerful” AI chip. And the company has just announced a roadmap that includes accelerator cards, agentic AI and new AI models.
Extensive AI Roadmap
A leading aspect of Alibaba’s roadmap is its new Zhenwu V900 processor. The chip offers three times the performance of its predecessor, the Zhenwu M890, which debuted in May.
Designed for AI training and inference, the V900 includes 216 GB of GPU memory and 1,200 GB/s of inter-chip bandwidth. It supports FP8 and FP4 precision formats, enabling developers to balance computing performance with the costs of running AI workloads.
Alibaba expects the processor to enter mass production and become commercially available in the first quarter of 2027.
Alibaba also unveiled a supernode server architecture that it says can support computing clusters of up to 500,000 accelerator cards. This hardware is an example of Chinese efforts to overcome a major limit in its AI development efforts: how to deliver competitive computing power without access to NVIDIA’s most advanced chips.
Alibaba, meanwhile, is pursuing a broader strategy that extends beyond hardware. The company announced that its Qwen 4 AI model is currently in training, with future Qwen 4.5 and Qwen 5 models expected to reach 5 trillion to 10 trillion parameters.
Alibaba also reported that its Qwen3.8-Max model completed 33 automated improvement cycles over one month, raising its Artificial Analysis benchmark score from 40 to 45. In a separate chip design experiment, the model made more than 10,000 electronic design automation tool calls and reduced chip area by 42% without sacrificing performance, according to Alibaba.
Furthermore, Alibaba CEO Eddie Wu announced a goal of operating more than 20 gigawatts of global data center capacity by 2032. The company’s roadmap also includes AgentCore, an enterprise platform for developing and managing AI agents, and proprietary Yitian processors designed for agentic AI workloads.
Similarly, Huawei recently announced that its Ascend 960DT AI processor will begin shipping in the first quarter of 2027, accelerating its original release schedule by three quarters. Yet Huawei faces technical obstacles. Its latest SuperPoD architecture connects fewer processors than originally planned, while individual Huawei chips deliver approximately half the compute performance of NVIDIA’s leading chips.
Additionally, Huawei Chairman Eric Xu acknowledged that the company may struggle to manufacture enough processors to satisfy demand from Chinese AI developers.
