- Huawei is preparing to test his most advanced Ai chip to date
- The reports indicate that ASSED 910D aims to challenge Nvidia H100 in the performance of the AI
- Analysts are skeptical, due to software shortcomings and disadvantages of the ecosystem
Huawei has intensified its ambitions to be considered the Chinese rival of Nvidia in recent weeks – first, by launching a new set of IA infrastructure architecture, the Cloudmatrix 384 supernod, to compete with the NVL72 system of the American flea giant.
Then, one day after the United States announced its intention to further strengthen the exports of ia flea markets to China – including the H20 processor of NVIDIA – Huawei revealed Ascend 920, its new generation AI chip, which should enter mass production in the second half of 2025.
Now a report in The Wall Street Journal Huawei claims are preparing to test its most advanced AI chip to date, Ascend 910D, and has approached several Chinese technological companies to start technical assessments. Lots samples should be available from the end of May 2025.
Sives plus on the Nvidia H100
Ascend 910D is the last iteration of Huawei AI processors, based on existing 910B and 910C models. Although the chip is not yet available in the trade, Huawei hopes it could correspond – or even exceed – the performance of the H100 of Nvidia.
Independent analysts are naturally skeptical about the fact that Huawei is able to fill the performance and the ecosystem gap between himself and Nvidia of any time.
Hpcwire Neil Shah reports, Vice -President of Counterpoint Research, saying: “According to a global design in the system – calculation, memory integration, networking and, above all, software orchestration – Nvidia remains three generations in advance.”
This gap is due in part to the established software battery of Nvidia, in particular Cuda, which plays a major role in the acceleration and management of workloads of the AI. Huawei, on the other hand, does not have a similar software platform, which makes GPU or scale optimization more difficult in complex AI infrastructure.
Despite these obstacles, Huawei continues to send its 910C chips in volume to Chinese data centers and research laboratories and the decision to test the 910D suggests that the company doubles efforts to create an autonomous EA material ecosystem.
Whether the 910D becomes a real rival of the H100 of Nvidia or remains a “good enough” option for domestic use, it signals a continuous change in the world material competition of AI.
