Huawei has unveiled a three-year plan to problem Nvidia in AI chips, specializing in large chip clusters, sooner information hyperlinks, and powerful coverage help.
At its Huawei Join convention, Chairman Eric Xu detailed a technique to hyperlink as much as 15,488 Ascend AI chips utilizing a brand new UnifiedBus protocol. Huawei claims this setup can transfer information as much as 62 occasions sooner than Nvidia’s upcoming NVLink144 expertise. The objective is to offset weaker chip efficiency with scale and networking power.
The announcement indicators Huawei’s confidence in its native manufacturing provide regardless of US sanctions that minimize it off from high chipmakers like TSMC. Bernstein analysts say the transfer marks progress towards a self-reliant Chinese language semiconductor ecosystem.
China is pushing homegrown chip innovation as US tariffs and export curbs tighten. Huawei nonetheless lags in uncooked chip energy—its next-gen Ascend 950 is claimed to achieve solely about 6% of Nvidia’s upcoming VR200—however hopes to bridge the hole by clustering as much as 1,000,000 chips and bettering reminiscence structure.
Whereas consultants doubt whether or not Huawei can mass-produce its designs, the corporate believes its SuperPod cluster tech and high-speed interconnects might finally rival Nvidia’s dominance in AI computing.
Unicorn Alerts leverages superior AI expertise to give you highly effective market predictions and actionable inventory scans. Obtain the app at the moment and 10x your buying and selling & investing journey!
