Guo Zhenxing of Huawei: After the wave of DeepSeek, AI will rapidly unleash huge manufacturing production dividends | The Frontline

On April 28, the Huawei AI + Manufacturing Industry Summit 2025 was officially held in Guangzhou. The theme of this summit is "Accelerating Industry Intelligence". More than 900 opinion leaders, enterprise representatives, industrial partners, scholars and experts from the manufacturing industries such as automotive, mechanical and electronic, pharmaceutical, heavy industry, and light industry participated in the summit on-site.

At the industry summit, Huawei officially released the "Three - layer, Five - stage, Eight - step" methodology and shared 20 solutions for seven scenarios in the manufacturing industry.

Liu Chao, the vice president of Huawei and the CEO of the Manufacturing and Large Enterprise Business Group, said that Huawei's full - stack AI infrastructure can be flexibly adapted to multiple manufacturing scenarios, reducing the threshold for enterprises to use AI. The intelligent applications incubated based on Huawei's own production practices enable front - line workers to easily use AI.

Huawei, together with its partners, has created a series of solutions covering new fields from vehicle equipment, mechanical and electronic, pharmaceutical and light industry to embodied intelligence.

In the automotive field, taking the cooperation between Huawei and GAC Group as an example, Huawei helped GAC Group build an intelligent R & D platform.

In the past, traditional automobile manufacturing enterprises generally took two to three years to iterate a new model. However, with the help of large AI models and R & D toolchains, the R & D time for a whole vehicle can be optimized from 36 months to 18 months. The iteration cycle from concept design to the entire manufacturing process has been greatly accelerated.

At this summit, GAC Group and Huawei jointly released an intelligent R & D platform solution and announced the official launch of the construction of a demonstration site for the intelligent R & D platform.

After the wave of large models arrived, Huawei has started an AI transformation and efficiency improvement, both in its own business and with its customers in the manufacturing industry.

Feng Rui, the deputy head of the Smart Manufacturing System Department of Huawei's China Enterprise Business Group, said in an interview after the summit that in the R & D field, Huawei has imported more than 13 million high - value basic documents, more than 10,000 API (interface technology documents), and more than 8.5 million open - source code repositories into the R & D data platform.

With the help of large R & D models and AI assistants, the development cycle of Huawei's software versions, which used to take about 9 - 18 months to release a version, can now be reduced to one month per release.

After the DeepSeek wave, there has been a large number of deployment requirements in the manufacturing industry. According to Guo Zhenxing, the vice president of Huawei's China Enterprise Business Group, it is estimated that more than 300 enterprises will formulate plans for large model deployment in 2025.

Many enterprises have quickly implemented these capabilities. For example, FAW has relied on DeepSeek's code capabilities to increase the overall code output efficiency by 30%. Guangzhou Pharmaceutical Group has built a pharmaceutical knowledge base with DeepSeek, enabling cross - departmental knowledge reuse and applying it to scenarios such as intelligent retrieval across different departments including R & D, sales, and medical research, as well as industry trends and competitor analysis.

Huawei has carried out all - round solution adaptation for DeepSeek based on its Ascend solutions. From pre - training to fine - tuning, reinforcement learning scenarios, and inference scenarios, targeted adaptation computing solutions have been proposed, which can help customers quickly complete the secondary training of DeepSeek.

Huawei has done a lot of optimization work on the matching between computing resources and models with mainstream models, including DeepSeek, to ensure that each model can achieve optimal performance in the Ascend environment, such as DeepSeek's MTP, MLA, and large EP parallel technologies.

Currently, at the ecological level, Ascend's full - stack software platforms have been open - sourced, such as the underlying CANN and the upper - layer MindSpore. Now, more than 100 manufacturing partners have built DeepSeek solutions based on Ascend, covering various industry needs.

At this summit, Huawei also set up sub - forums for vehicle equipment, electronics and new energy, computing component partners, etc., to conduct in - depth exchanges around specific industries and key scenarios.

Guo Zhenxing, the vice president of Huawei's China Enterprise Business Group, said that in 2025, Huawei will seize the AI opportunity and provide end - to - end full - stack new infrastructure for intelligent connection, storage, computing power, and platforms around the needs of data "collection, transmission, storage, computing, management, and application", continuously helping enterprises with digital and intelligent transformation and creating value for customers.