Regardless of Washington’s efforts to curb Chinese language entry to modern AI {hardware}, Huawei has made important progress. “Huawei’s HiSilicon 910 sequence is an efficient final result for China’s native AI server chip ambitions, which have been accelerated by US export restrictions,” mentioned Neil Shah, VP for analysis and accomplice at Counterpoint Analysis.
Faisal Kawoosa, founder and lead analyst at Techarc, famous that China’s longstanding perception in constructing its total tech stack and ecosystem is now bearing fruit. “Huawei growing Ascend 910D is a step on this path. It’s a major transfer and can open up a twin AI entrance on the worldwide area,” he mentioned. “Chinese language tech corporations have traditionally led mass adoption of applied sciences, and this might now speed up international AI deployments as nicely.”
Whereas international locations just like the US and India stay cautious of Huawei, Kawoosa added that there’s nonetheless “important alternative for Huawei to popularize its tech in markets the place it stays welcome,” delivering totally localized options with out reliance on international suppliers.
A robust transfer — however how near Nvidia?
Regardless of the fanfare, Huawei’s journey to rival Nvidia on technical grounds stays uphill. In accordance with Shah, the Ascend 910D is a significant development however nonetheless lags behind Nvidia in vital areas.
“From an total system-level design — compute, reminiscence integration, networking scalability, and crucially, software program orchestration — Nvidia stays three generations forward,” Shah defined.
Huawei’s Ascend 910D will reportedly ship with two generations outdated high-bandwidth reminiscence (HBM) in comparison with Nvidia’s newest choices. Furthermore, with out a sophisticated CUDA-like ecosystem to optimize AI workloads throughout GPUs and networks, Huawei’s chips will face inherent disadvantages in scalability and effectivity.