After the release of the deepseek-V3.2-Exp large model, Huawei Ascend, Cambricon, and Hygon information have completed adaptation. In the cloud computing sector, major internet companies such as Tencent and Alibaba are also actively adapting domestic chips. At the end of consumers, foreign giants like Apple and Openai are accelerating the commercial deployment of AI technology.
On September 30, Star Market reported every day that the ecological synergy of domestic AI software and hardware showed an accelerated trend. Yesterday, shortly after the release of the large DeepSeek-V3.2-Exp model, several chipmakers completed their “Day 0 adaptation.”
According to WeChat's official account “Huawei Compute”, Ascend quickly completed adaptation and deployment based on inference frameworks such as VLLM/Sglang, achieving 0-day support for DeepSeek-V3.2-EXP, achieving all inference code and open source operator implementations for developers.
Cambricon has announced on its official WeChat account that it has achieved compatibility with DeepSeek's latest model, DeepSeek-V3.2-Exp. It can greatly reduce the use of DeepSeek sparse's attention mechanism and combine it with Cambricon's extreme computational efficiency, training and inference costs in long-term sequencing scenarios.
Hygon Information has announced that DCU has achieved seamless adaptation and deep optimization of DeepSeek-V3.2-Exp. This model demonstrates excellent performance in Hygon's DCUs, while examining the high versatility, strong ecosystem compatibility, and self-controllable technical advantages of Hygon's DCUs.
As domestic hardware manufacturers gradually strengthen their efforts, major internet companies are also engaged in “two-way collaboration.” On September 16th, Tencent Cloud announced that it is fully adapted to mainstream domestic chips, actively participating in the open source community, contributing to them and giving back. At the recently held 2025 Alibaba Cloud Conference, Wu Yongming pointed out that Tongyi Qianwen is aiming to take an open approach and build an open computing power system that attracts more third-party GPUs and switch manufacturers to support uneven computing.
Guotou Securities believes that domestic computing power is witnessing new changes in both supply and demand aspects. On the one hand, Huawei Ascend's representative domestically produced computing chips will steadily improve performance and provide supply support to the domestic artificial intelligence industry. Meanwhile, domestic internet giants are gradually adapting to domestic chips, and capital expenditures are expected to continue to grow and provide demand-side support to the domestic computing power industry.
Previously, Guotai Haitong Securities Research showed that artificial intelligence is the key driver behind new technological revolutions and new rounds of industrial transformation. Large domestic models and domestic chips are evolving towards a unified ecosystem of software hardware synergies, which is expected to further accelerate the development of the domestic AI industry.
Perhaps, as Openai CEO Sam Altman stated, “The AI age requires a paradigm shift similar to the iPhone. Hardware, operating systems, and models need to be co-designed from the start of AI.” Overseas, commercial implementation of AI technology has moved beyond the end of chips, and is now moving towards consumer terminals. For example, Apple is working to provide real-time translation and other features to its users through large device-based models. Openai has officially entered the home appliance market by issuing assembly contracts for AI devices to domestic manufacturers.
Regarding terminal products, the National Development and Reform Committee said yesterday that it will promote accelerated application and widespread use of next-generation smart terminals and intelligent agents. First, we will strengthen the creation of a policy environment with clear guidance, respect for patterns, standardized development, and policy guidelines. Second, it promotes technology breakthroughs in coordination and open source sharing, supporting AI companies and leading companies in various industries to carry out joint innovation. Third, we will strive to expand our markets on both supply and demand aspects.
Guotai Haitong Securities said that large-scale models and generation AI are moving rapidly from the cloud to consumer endpoints, accelerating the creation of an integrated hardware software AI ecosystem with key players. In the long run, anyone who can first create a killer application that combines “AI + hardware” will have an advantage in the next wave of the computing platform competition. This also serves as a key indicator of capital markets to measure the growth potential of technology companies.
Chinese merchant securities also pointed out that the main bottlenecks for deploying large-scale models on the device side are computing power, memory, energy consumption, thermal dissipation, sensing, and system adjustments. Associated hardware manufacturers are expected to secure new ecological positions, presenting rare opportunities to move from “existing competition” to “gradual feasts.”
