China’s biggest tech company is not just talking anymore. It is shipping. The new embedded Huawei AI chip push is built around the upcoming Ascend 960. It is Huawei’s boldest move yet to challenge Nvidia’s hold on AI hardware, both in China and beyond. US export limits are still in place. Beijing still wants chip self-reliance. Huawei’s chip roadmap shows how much the Nvidia China market is already changing.
Huawei’s Ascend 960: The Next Embedded Huawei AI Chip
Huawei’s rotating chairman, Wang Tao, unveiled the Ascend 960 at a summit in Shanghai. It will replace the current Ascend 950 family. Huawei plans to make it available in 2027. Analysts want to know one thing: can this chip train the biggest AI models, not just run them? Huawei’s Guo Ping says the plan is smart design, not just raw chip power. He points to how other hardware makers overcame limits with better architecture.
Each new Ascend chip aims to double the compute power of the last one. The Ascend 970 is already planned for 2028. This steady pace is central to Huawei’s three-year push to close the gap with Nvidia’s top chips.
Traction in the Nvidia China Market
Huawei is not just building chips on paper. DeepSeek, one of China’s top AI labs, is reportedly using about 160,000 Huawei Ascend 950DT chips. These will run at a new data center in Inner Mongolia. Huawei is also looking at markets outside China, including Malaysia and Egypt.
This matters. It shows Chinese AI firms trust Huawei hardware enough to build real infrastructure on it, not just test it. That is a real shift in the Nvidia AI hardware story, where Nvidia has long been the default pick.
Huawei AI Chips vs Nvidia: Where the Gap Still Exists
Huawei AI chips vs Nvidia are still not a close race. Top Chinese labs, including DeepSeek, still train their hardest models on Nvidia chips first. They switch to domestic chips later, for simpler tasks. This shows how hard it is to fully drop Western hardware. Nvidia’s newest Blackwell chips are still not sold directly in China. Yet some still slip in through smuggling, in small numbers.
Huawei also faces supply problems. It recently raised prices on the Ascend 950DT chip by 60%. The reason: tight supply of parts. Ambition and factory output do not always move at the same speed. Morgan Stanley expects China’s AI computing market to hit about $96 billion by 2030. That is a huge prize, and it explains why both sides are racing so hard.
SuperPod Clusters and the Bigger Nvidia China News Story
Huawei built a new system called SuperPod to make up for any gap at the chip level. It can link up to 100,000 Ascend chips into one cluster. It uses Huawei’s own UnifiedBus network. Huawei also uses a method called logic folding to boost chip performance. The bet is simple: strong networking and scale can offset weaker single chips. Morgan Stanley analyst Charlie Chan agrees. He says system-level design, like packaging and rack-scale setups, is closing the gap fast.
The Bottom Line
The rise of the embedded Huawei AI chip marks a real shift in China’s AI hardware world. Nvidia still holds the technical edge today. But the real question is different. Can Huawei’s bet on scale and smart design close that gap for good? The answer will shape not just the Nvidia China market but the global AI hardware race for years to come.