- Huawei promises annual progress on its AI chips while rivals remain on slower development cycles
- Nvidia now faces a competitor who is accelerating the expansion of its infrastructures
- Huawei already operates large computing clusters capable of supporting millions of connected vehicles
On June 5, Huawei Vice President Chen Lin spoke at the Huawei Cloud INSPIRE Innovators 2026 conference, where all eyes were on one announcement in particular – the Ascend 950DT chip, expected on Huawei Cloud later this year.
The 950DT features improved vector computing power, increased memory bandwidth, and native support for low-precision formats like FP8.
According to Chen, the chip is simpler to program and better suited to intelligent driving than anything that has come before, but the rest of his talk deserves much greater attention than the chip itself – especially from competitors like Nvidia
One generation per year, with computing power doubled each time
“The Ascend chip is evolving at a rate of one generation per year, with a doubling of computing power,” Chen said, without nuance or reservation.
This is a public commitment to a launch pace aggressive enough to challenge how progress in AI chips is measured.
Nvidia has long imposed this tempo, each new architecture raising the level for all the competitors who try to follow – and a player who announces annual generational leaps, publicly, on stage, does not behave like a company still catching up.
Huawei’s ability to maintain this pace without advanced Western lithography tools remains an open and legitimate question.
This announced pace only has weight because it is based on a very real infrastructure.
Huawei Cloud has deployed large-scale computing clusters in Gui’an, Wuhu and Inner Mongolia, with a global network covering 34 regions and 102 availability zones.
More than 100,000 Ascend compute units currently support continuous iteration of algorithms for paying customers via Huawei Cloud.
Every day, more than two million intelligent driving vehicles and 60 million connected vehicles operate stably on this same infrastructure. This is operational data, not projections from a roadmap.
More than 30 automakers and OEMs have established in-depth partnerships with Huawei Cloud around intelligent driving and intelligent manufacturing.
This growing customer base absorbs each new generation of chips as soon as it arrives, giving Huawei a real testing ground that strengthens each subsequent version.
The transition from functionality to real ease of use
Huawei’s position goes beyond just hardware characteristics to venture into territory that is more difficult to challenge quickly.
Chen emphasizes that systems engineering capabilities are just as crucial as raw computing power in improving the effectiveness of intelligent driving training among manufacturers.
With its Lingqu architecture, Huawei Cloud ensures high-speed interconnection within supernodes, which significantly improves large-scale training efficiency.
Its AI DataLake platform produces hundreds of thousands of data sequences every day.
Huawei Cloud also worked directly with key players in intelligent driving across the entire model iteration cycle – from the integration of computing power to the adaptation and optimization of algorithms.
This level of involvement transforms Huawei from a chip supplier to an integrated infrastructure partner.
The stated ambition, in Chen’s words, is to move from chips that are simply “usable” to a complete stack that is truly “easy to use.”
Via Guancha
