Huawei and East China Normal University have developed a time-series foundation model that topped the international TIME-leaderboard and GIFT-Eval benchmarks.
DeepSeek has released an open-source software stack for Huawei Ascend chips, providing an alternative to Nvidia tools and supporting the Ascend 950 accelerator.
DeepSeek and Huawei are partnering to develop open-source programming tools for Ascend chips, aiming to reduce reliance on NVIDIA technology.
DeepSeek has launched an open-source suite of infrastructure tools for Huawei Ascend platforms, including the TileLang compiler and various high-performance computing libraries.
Huawei has unveiled the Atlas 960E SuperPoD, an AI cluster connecting 4,096 Ascend chips to support 10-trillion-parameter models.
Huawei has launched the Atlas 960E SuperPoD, an AI computing cluster featuring 4,096 Ascend NPUs designed to compete with Nvidia in large-scale infrastructure.
China Telecom and Huawei have launched the Xing4.0-29B-A4B model, the first large-scale agent model trained on Ascend Atlas 900 A3 SuperPoD liquid-cooled infrastructure.
Huawei aims to overhaul its AI training infrastructure by 2027 by launching three new generations of Ascend processors with doubled performance.
Huawei has accelerated its AI accelerator roadmap, moving up the release of the Ascend 960 series and significantly boosting the FP4 performance of its upcoming hardware.
Huawei Technologies will launch two new Ascend AI accelerators in 2027, focusing on cluster scaling and proprietary interconnect technology to overcome hardware limitations.
Huawei is focusing its AI strategy on computing power and hardware, with over 1,000 Ascend super-nodes already deployed and new models scheduled for 2027.
Huawei has accelerated the release schedule for its Ascend 960 AI chips, with the first models expected to launch in early 2027.
