Unlock the publisher’s digest free
Roula Khalaf, editor -in -chief of the FT, selects her favorite stories in this weekly newsletter.
The founder of Huawei said the United States exaggerated the capacity of the Chinese flea manufacturer and had played his business technology in the midst of commercial talks between Beijing and Washington which include discussions on export controls.
In a rare interview with the daily newspaper of the Chinese state on Tuesday, Ren Zhengfei said that the Huawei Ascend chip, the main rival of Nvidia products in China, “is still lagging behind the United States by a generation”. He added that “the United States has exaggerated Huawei’s capabilities – we are not yet strong”.
REN’s comments arise as Nvidia’s managing director Jensen Huang has sounded the alarm in recent weeks on Huawei’s progress with his artificial intelligence tokens, saying that Washington’s limits on sale to manufacturers of American Chiples have given birth to a “formidable” competitor who threatens to undermine American domination of AI.
The United States and China began a new series of trade discussions on Monday in London, which included a discussion on Washington export controls on key technology.
During the first series of talks in Geneva, the United States did not discuss export controls. However, the recent use of restrictions by Beijing on certain rare criticisms and minerals used in the manufacture of cars – which threatens to close factory lines in the United States, Europe and Japan – has brought the problem in commercial discussions.
Huawei benefited from the Washington ban on Nvidia flea expeditions to China while Chinese technology giants accelerated ascend flea purchases and were preparing to adopt Huawei technology.
However, the majority of Chinese AI companies, including Deepseek, use Nvidia chips to train large language models that feed AI tools. Domestic alternatives are increasingly used for the less complex task of calling models to generate answers in tools such as chatbots.
Huawei analysts and researchers have already complained about technical problems in using the company’s fleas for LLM training, citing difficulties in working with fleas together and distributing computer workloads.
On Tuesday, Ren hinted that the company had made progress to solve these problems, saying that Huawei could “compensate” the lowest performance thanks to cluster computer, which involves linking several chips to increase the power of the AI server.
“Using clustering and stacking, our IT results are comparable to the best in the world,” he said.
Huawei’s new AI Cloudmatrix 384 server is the centerpiece of its strategy to compete with Nvidia by assembling a large number of fleas to strengthen bandwidth capacity and manage more data. The server connects 384 AI processors using Huawei optical technology.
Several Huawei customers are already testing the Cloudmatrix server, working with business engineers to solve heat -generated heat problems by the large number of chips running in unison and the weight of the machine, said a person involved in tests.
Huawei’s biggest challenge has been to create “an ecosystem” of developers who will use its platform to build AI models, added the person. One of the advantages of NVIDIA is its CUDA software platform, which, according to the developers, is easy to use.
Ren said that Huawei was investing RMB180 billion RMB (25 billion dollars) per year in research and development, with RMB60 billion RMB which did not aim to develop products but to make revolutionary discoveries.
He added that China had separate advantages when creating its technological capacities.
“AI depends on abundant electricity and advanced network infrastructure,” he said. “China’s electricity and network production systems are world -class. Our telecommunications infrastructure is the most advanced in the world. ”
Additional Demetri Sevastopulo reports in Washington
