Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

aiweekly.co.newsletter.semianalysis.com+1.finance.yahoo.com.A new SemiAnalysis report says the partnership between Huawei and AI startup DeepSeek is closing the hardware and software gap with Nvidia. The South China Morning Post reports the research firm described it as a scenario where Jensen Huang's "nightmare is starting to come true". Separately, Huang said in an interview at Nvidia's GTC Berlin conference that the US must treat China as "very, very serious competition".scmp.com+1
The report, titled "The Huawei-DeepSeek Symbiosis," was published on October 21 by Bryan Shan and Dylan Patel. It argues that "neither can do this alone." In its telling, Huawei has no frontier lab to prove its chips can train and run models, and DeepSeek has models but cannot buy the computing power it needs. "Huawei provides the hardware and native software stack; DeepSeek brings the workloads and engineering needed to make that hardware useful," the authors wrote.newsletter.semianalysis.com
The analysis follows DeepSeek's release, just before China's National Day holiday, of Ascend support for six software projects, including TileLang, DeepGEMM and FlashMLA. DeepSeek had originally built these tools for Nvidia chips. SemiAnalysis predicts that DeepSeek will be running primarily on Huawei's technology stack by 2027 or 2028.scmp.com+1
The firm also ran its own benchmark tests on the Ascend 950PR chip, and the results were mixed. On compute-heavy workloads, Huawei's own kernels reached 99.7% of the chip's theoretical matrix-math ceiling. Overall, though, the chip's performance falls between Nvidia's older A100 and AMD's Advanced Micro Devices, Inc. MI300X. On workloads limited by memory speed, Ascend reached a median 41.5% of its attainable performance, compared with 46.8% for Nvidia's H100. The report also found that DeepSeek's current optimizations depend on Huawei's supernode systems and do not yet work well on x86 servers.aiweekly.co+2
Yahoo Finance's Deirdre Bosa asked Huang about reports that DeepSeek and Huawei are working together on chip designs. He replied, "We should always take China seriously". "If there's one country that could manufacture things at scale, whether it's manufacturing, of course, chips at scale, manufacturing open models at scale, or even very importantly, manufacturing AI researchers at scale," it is China, he said. Huang added that American frontier labs, computing systems and infrastructure are still the strongest in the world.benzinga.com+1
Huawei has been gaining ground in China's AI accelerator market. An IDC analysis found that Nvidia held 34% of that market in the first half of 2026, compared with 28% for Huawei, and that Nvidia's shipments in China fell 44% from a year earlier. Nvidia shares fell in trading on Thursday, October 22.benzinga.com