Chinese tech giants critical of Huawei accelerator overheating

Chinese tech leaders express skepticism towards Huawei´s Ascend Artificial Intelligence accelerators amid overheating and software hurdles.

Chinese technology heavyweights such as Tencent, ByteDance, and Alibaba are scrutinizing Huawei´s efforts to challenge NVIDIA´s dominance in the artificial intelligence accelerator market. Against the backdrop of new trade restrictions, Tencent recently confirmed amassing a large supply of NVIDIA H20 chips—hardware difficult for domestic companies like Huawei to match in scale. Despite the limitations imposed by sanctions on imported chips, Chinese companies reportedly spent billions securing advanced NVIDIA GPUs during early 2025, leaving Huawei to navigate a challenging and competitive environment.

Huawei´s Ascend accelerator line, notably the Ascend 910C model and its anticipated successor, has been positioned as a potential alternative to NVIDIA´s restricted products. Preliminary data suggested solid performance, fueling initial optimism. However, according to industry insiders speaking to The Information, real-world testing at ByteDance and Alibaba uncovered persistent issues. Specifically, Huawei´s trial units experienced overheating, raising reliability questions for large-scale deployment. Additionally, feedback pointed to a lack of maturity in Huawei´s Compute Architecture for Neural Networks (CANN) software stack, which lags behind the well-established CUDA ecosystem powered by NVIDIA.

The combination of hardware instability and software shortcomings has led major players in China, including DeepSeek, to accelerate their own artificial intelligence chip development projects. This broader shift threatens Huawei´s position as a leading domestic alternative and reflects a subtle fracturing of China´s technology self-reliance strategy. As these companies build up proprietary capabilities, the competitive field for artificial intelligence accelerator supremacy is likely to intensify, with more firms opting to reduce dependence on both foreign and local giants alike.

76

Impact Score

New test generates an immune health score

Researchers at Yale University created an immune health metric by profiling blood cells, gene expression, and more than 1,300 proteins, then using machine learning to correlate those signals with health. The experimental test aligned with responses to disease and vaccines but is not ready for clinical use.

How muscles remember movement and exercise

Research shows skeletal muscle stores a lasting epigenetic memory of both training and atrophy, shaping how quickly we regain strength or lose it, and that exercise can help reset negative imprints.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.