Switch off the editor's digest freed from charge
Roula Khalaf, editor of the FT, selects her favorite stories on this weekly newsletter.
Huawei's founder said that the USA had overdated the abilities of the Chinese chipmach and the technology of its company in the course of trade talks between Beijing and Washington, which contained discussions about export controls, depressed.
In a rare interview with China's state people day-after-day on Tuesday, Ren Zhengfei, Huaweis Ascend Chip, the predominant competitor of Nvidia's products in China, said “still stays from a generation behind the USA.” He added that the “Huawei's skills have exaggerated us – we usually are not that strong yet”.
Ren's comments come when Jensen Huang, Managing Director of Nvidia, has spoken to Huawei up to now few weeks together with his artificial intelligence chips and said Washington's borders for the US chip maker who’s sold to China have led to an “impressive” competitor who mixed the American dominance of AI technology.
The USA and China began a brand new round of trade talks in London on Monday, which included the discussion about export control in Washington in relation to key technology.
During the primary talks in Geneva, the United States didn’t discuss export controls. Beijing's recent use of restrictions for some critical rare earths and minerals which are utilized in automotive production that threatens to shut the factory lines within the USA, Europe and Japan has brought the issue into gender discussions.
Huawei advantages from Washington's ban on shipping from Nvidia chips to China, since Chinese tech giants accelerated the purchases of Ascend chips and made preparations for the takeover of Huawweist technology.
Nevertheless, the vast majority of the Chinese AI firms, including Deepseek, Nvidia chips, uses the massive language models that provide AI tools with electricity. Domestic alternatives are increasingly used for the less complex task to call up the models to generate answers in tools reminiscent of chatbots.
Analysts and Huawei researchers previously complained about technical disorders when using the corporate's chips for training LLMS. You have given difficulties in working with the chips and the distribution of workloads of computing in all cases.
Ren indicated on Tuesday that the corporate had made progress to unravel these problems, and said Huawei could “compensate for a poorer performance” through cluster computing, wherein several chips are linked to increasing the AI server.
“With the assistance of clustering and stacking, our computer results are comparable to one of the best on the planet,” he said.
The recent CloudMatrix 384 -ai server from Huawei is the guts of its technique to compete with Nvidia by compiling a lot of chips to strengthen the bandwidth capability and to address further data. The server connects 384 KI processors using Huawei's optical technology.
Several Huawei customers are already testing the CloudMatrix server and are working with the engineers of the corporate to unravel problems related to the heat that generated by the massive variety of chips which are operated in harmony and the burden of the machine, said a one who is involved within the tests.
Huawei's biggest challenge was to create “an ecosystem” of developers who will create KI models with their platform, the person added. One of some great benefits of NVIDIA is the software platform Cuda that developers say that they’re easy to make use of.
According to Ren, Huawei invested in research and development for a 12 months 180 billion RMB (25 billion USD).
According to Ren, China had different benefits when expanding its technological skills.
“AI is dependent upon loads of power and a complicated network infrastructure,” he said. “China's electricity generation and grid systems are world class. Our telecommunications infrastructure is widespread worldwide.”
