{"id":182498,"date":"2025-10-01T12:19:11","date_gmt":"2025-10-01T12:19:11","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/182498\/"},"modified":"2025-10-01T12:19:11","modified_gmt":"2025-10-01T12:19:11","slug":"deepseeks-new-ai-model-debuts-with-support-for-china-native-chips-and-cann-a-replacement-for-nvidias-cuda-chinese-chipmakers-huawei-cambricon-and-hygon-get-first-class-support","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/182498\/","title":{"rendered":"DeepSeek\u2019s new AI model debuts with support for China-native chips and CANN, a replacement for Nvidia’s CUDA \u2014 Chinese chipmakers Huawei, Cambricon, and Hygon get first-class support"},"content":{"rendered":"

Chinese AI firm DeepSeek has released<\/a> its latest large language model, DeepSeek-V3.2-Exp, with first-day optimizations for Huawei\u2019s Ascend hardware and CANN software stack. The launch marks a shift in priorities to ensure leading-edge models run on domestic accelerators rather than relying on Nvidia\u2019s CUDA ecosystem<\/a>.<\/p>\n

DeepSeek announced the model on September 29, posting code and checkpoints to Hugging Face alongside a technical report. The company describes V3.2-Exp as an \u201cintermediate step toward our next-generation architecture,\u201d designed to cut costs on long-context inference. It features a sparse attention mechanism that trims memory and compute requirements while maintaining output quality.<\/p>\n

<\/p>\n

Huawei\u2019s Ascend team and the wider vLLM-Ascend community moved swiftly to integrate DeepSeek-V3.2-Exp. In the vLLM-Ascend repo, a new issue outlines custom operator installation steps and kernel packaging for Ascend NPUs to support V3.2-Exp. The CANN team also published an inference recipe, positioning the model for immediate deployment across Huawei hardware.<\/p>\n

You may like<\/p>\n

Other Chinese chipmakers have joined in, including Cambricon, which released an update to its vLLM-MLU fork with compatibility for V3.2-Exp, claiming the combination of its inference engine and the model\u2019s sparse attention cuts costs for processing long sequences. Hygon also announced that its DCU accelerators had been tuned for \u201czero-wait\u201d deployment through its DTK software stack.<\/p>\n

Increased collaboration bw DeepSeek & Ascend\/CANN team in supporting V3.2-Exp w\/ gitcode updates to Cann as well as GitHub updates into vLLM & SGLang + TileLang support.Also Cambricon had updates into vLLM (vLLM-MLU) to support its inference.DS is really dealing w\/ reality of\u2026 https:\/\/t.co\/Unsvyxw9b6 pic.twitter.com\/CBgk7pVZrxSeptember 29, 2025<\/a><\/p>\n

Meanwhile, SGLang confirmed V3.2-Exp support across multiple back ends, including Ascend, while DeepSeek\u2019s GitHub notes suggest parity with vLLM at launch. DeepSeek itself publicly references both TileLang and CUDA kernels in its announcements, urging researchers to use TileLang for prototyping. Practically, that means the same model artifact can be deployed across Nvidia and Chinese accelerators with only minimal graph changes.<\/p>\n

The sheer speed of adoption<\/a> here illustrates how China\u2019s AI ecosystem is undeniably preparing for a future in which access to Nvidia hardware cannot be taken for granted. Nvidia\u2019s CUDA remains dominant for both training and inference, but DeepSeek\u2019s latest release is one of the first from a major Chinese company to apparently arrive optimized for non-CUDA stacks on day one.<\/p>\n

The coordinated effort across Ascend, Cambricon, and Hygon is the clearest sign to date that Chinese firms are taking Beijing\u2019s demands for AI sovereignty seriously<\/a>, not just making their hardware compatible after the fact, but positioning domestic platforms as first-class targets.<\/p>\n

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.<\/p>\n

Follow Tom’s Hardware on Google News<\/a> to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.<\/p>\n