Huawei is lining up two new AI chips for 2027 as it tries to narrow Nvidia’s considerable lead in accelerator hardware.
According to Reuters, Huawei’s rotating chairman, David Wang, said the Chinese technology giant will launch the 960DT during the first quarter of 2027. The Ascend 960PR is due in the third quarter.
The silicon forms part of Huawei’s expanding AI computing push, where building another accelerator is only half the battle. Modern AI workloads need large numbers of chips working together without spending their lives waiting for data.
Huawei reckons its UnifiedBus technology can deal with that problem by connecting large numbers of processors into much bigger computing systems. Wang said it will be a key part of the company’s next-generation AI hardware.
The outfit has developed 11 semiconductors using UnifiedBus for its larger systems. Its most ambitious configurations, which Huawei calls superclusters, are designed to support as many as one million AI processors.
Huawei has already shipped more than 1,000 smaller linked systems, known as supernodes, to more than 370 customers. These combine multiple AI chips so they can chew through the same workload together.
Wang did not say how many processors could sit inside an individual supernode. Huawei declined to name the customers or provide a breakdown showing where those systems have ended up.
The ability to lash lots of processors together has become particularly important for Chinese technology companies. US export controls restrict Chinese access to some advanced computing chips and semiconductor manufacturing equipment.
That leaves outfits such as Huawei trying to squeeze more computing power from hardware they can obtain or build domestically. Joining more processors together can help, provided the interconnect does not become an expensive electronic traffic jam.
Those restrictions have helped Huawei become one of China’s more prominent AI accelerator suppliers. Nvidia remains the much larger international rival, with hardware and software that are already deeply embedded in developers’ AI infrastructure.
The software side could prove every bit as awkward as the silicon. Nvidia has spent years building a developer ecosystem around its hardware, meaning Huawei needs more than impressive processor counts and some very large boxes to prise users away.
Wang said Huawei’s AI chip ecosystem now has 5,270 monthly active developers. That gives the outfit a base to build on while it prepares the 960DT, Ascend 960PR and increasingly enormous UnifiedBus machines for 2027.
