Huawei’s rotating chair, Eric Xu, has called on Chinese artificial intelligence companies to accelerate their development efforts in order to close the gap with leading U.S. AI laboratories, emphasizing the need to balance innovation speed with the management of potential risks. Xu made these remarks during a press conference in Shanghai, contrasting with recent appeals from U.S.-based firms Anthropic and OpenAI, which have advocated for a coordinated slowdown in AI advancement amid growing concerns about the technology’s existential threats.
Xu stressed that Chinese researchers must increase their development pace not only to compete globally but also to gain firsthand awareness of the dangers associated with AI innovation. He highlighted the importance of striking an equilibrium between rapid progress and risk mitigation once potential issues become apparent.
The comments coincided with Huawei’s announcement of updated plans for its AI hardware, an area where it is seeking to offer an alternative to Nvidia’s dominant position in AI computing platforms. This strategic move takes place amid ongoing U.S. export restrictions that limit the availability of cutting-edge AI chips to China. Xu acknowledged that Huawei’s capabilities currently lag behind top U.S. rivals but noted that unpredictable export controls have driven many Chinese customers to prefer domestically produced technology.
“We can’t accept a destiny that we cannot control,” Xu said, underscoring the company’s and the Chinese government’s commitment to achieving full self-sufficiency in semiconductor chips and the broader supply chain.
Huawei has recently become more forthcoming about its AI ambitions, releasing a detailed multiyear roadmap for its Ascend series of chips last September. Yesterday, the company revealed that the Ascend 960 DT chip, intended for training AI models, is scheduled to launch in the first quarter of 2024, followed by another chip designed for AI inference later in the third quarter. Both new processors are expected to deliver double the computing performance of Huawei’s current highest-end offering.
A key differentiator in Huawei’s strategy is its focus on advanced networking technology, which enables the company to assemble larger clusters of AI processors than competitors without sacrificing bandwidth or stability. This design approach helps compensate for the relatively lower performance of individual Ascend chips, a limitation driven partly by export controls that restrict access to cutting-edge manufacturing technologies such as those provided by Taiwan’s TSMC.
The company’s latest data center product will support up to 4,096 Ascend processors within a single system, compared with 1,024 in earlier generations, allowing for the training and deployment of significantly larger AI models. This scalable architecture is central to Huawei’s efforts to enhance the competitiveness of its AI hardware ecosystem despite ongoing geopolitical challenges.