- Huawei promises annual leaps in AI chips as rivals follow slower development cycles
- Nvidia now faces a rival that is accelerating its infrastructure expansion
- Huawei already operates large computing clusters that support millions of connected vehicles
On June 5, Huawei Vice President Chen Lin spoke at the Huawei Cloud 2026 INSPIRE Innovators Conference, where all eyes were focused on one announcement: the Ascend 950DT chip, coming to Huawei Cloud later this year.
The 950DT offers improved vector computing power, broader memory bandwidth, and native support for low-precision formats such as FP8.
According to Chen, the chip is easier to program and better suited for smart driving than anything that came before it, but what Chen said next deserves much more scrutiny than the chip itself, especially for rivals like Nvidia.
One generation per year, computing power doubled each time
“The Ascend chip is evolving at a rate of ‘one generation per year, doubling computing power,'” Chen said, unreservedly and unreservedly.
This is a public commitment to a release cadence aggressive enough to call into question how the progress of AI chips is measured.
Nvidia has controlled that pace for a long time, with each new architecture raising the bar for every competitor that pursues it, and a rival that locks in annual generational leaps – publicly, on stage – is not behaving like a company that is still playing catch-up.
Whether Huawei can maintain that pace without advanced Western lithography tools remains a fair and open question.
The announced cadence only has weight because behind it there is a genuine 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.
Currently, more than 100,000 Ascend computing units support continuous iteration of algorithms for paying customers through Huawei Cloud.
Every day, more than two million smart driving vehicles and 60 million connected vehicles circulate stably on that same infrastructure. Those are operating numbers, not projections from a roadmap slide.
More than 30 automotive OEMs and suppliers have built strong partnerships with Huawei Cloud in smart driving and smart manufacturing.
That growing customer base absorbs each new generation of chips as they arrive, giving Huawei a live testing ground that refines each subsequent launch.
The change from usable to really easy to use
Huawei’s comment goes beyond hardware specifications and becomes something more difficult to quickly counter.
Chen emphasized that systems engineering capabilities are as decisive as raw computing power in helping automakers improve the efficiency of intelligent driving training.
Through its Lingqu architecture, Huawei Cloud achieves high-speed interconnection within super nodes, significantly improving training efficiency at scale.
Its AI DataLake platform supports the production of hundreds of thousands of data clips every day.
Huawei Cloud has also worked directly with major intelligent vehicle manufacturers throughout the entire model iteration cycle, from integrating computing power to adapting and optimizing algorithms.
That level of deep involvement transforms Huawei from a chip supplier to an integrated infrastructure partner.
The stated ambition, in Chen’s own words, is to move from chips that are merely “usable” to a full stack that is genuinely “user-friendly.”
Via Guancha (Originally in Chinese)
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds.
