Huawei officially introduced a new artificial intelligence deployment framework for the manufacturing sector on April 28, during an industry summit held in Guangzhou. The methodology includes a three-layer infrastructure model, five stages of AI maturity, and an eight-step implementation roadmap—designed to help enterprises integrate AI across both strategic planning and operational workflows.
Alongside the framework, Huawei also released 20 AI-powered solutions spanning seven manufacturing scenarios.
Liu Chao, vice president of Huawei and CEO of its manufacturing and large enterprises business unit, said the company’s full-stack AI infrastructure is built to accommodate a wide range of manufacturing use cases, significantly lowering barriers to industrial AI adoption.
Working with partners, Huawei has developed modular AI solutions across segments including automotive equipment, mechanical electronics, pharmaceuticals, and light industry—as well as emerging fields like embodied intelligence.
One case in point is Huawei’s partnership with GAC Group. Together, the companies have built a smart R&D platform to streamline vehicle design and engineering. Traditionally, it takes two to three years to bring a new car from concept to production. By integrating Huawei’s large language models (LLMs) and AI-driven toolchains, that timeline could be halved—from 36 months to just 18. This agility marks a major leap forward in product development.
At the summit, GAC and Huawei launched the smart R&D solution and announced a pilot program to begin scaling the platform across operations.
The rise of LLMs has prompted Huawei to incorporate AI more deeply across its internal and customer-facing workflows. Feng Rui, deputy director of Huawei China’s smart manufacturing unit for government and enterprises, said Huawei has consolidated more than 13 million technical documents, 10,000 API manuals, and 8.5 million open-source code repositories into a unified R&D data platform.
This foundation supports Huawei’s proprietary AI assistant and development-focused LLM. Feng noted that tasks like releasing a new software version—which previously took 9 to 18 months—could now be completed on a monthly cycle.
The demand for AI model deployment has surged following the debut of DeepSeek. According to Guo Zhenxing, vice president of Huawei’s enterprise business in China, over 300 companies are planning to deploy large models in 2025.
Some have already moved fast. FAW Group has boosted coding efficiency by 30% using DeepSeek’s capabilities. Guangzhou Pharmaceuticals Corporation built a knowledge base that spans R&D, sales, and medical research, enabling faster knowledge reuse and applications like AI-powered search and competitive analysis.
Huawei has optimized DeepSeek across its Ascend platform, supporting everything from pre-training and fine-tuning to reinforcement learning and inference. The platform is designed for rapid secondary training and customization of large models.
Performance tuning has been a key focus. Huawei has aligned computing resources with model workloads, citing compatibility with advanced techniques such as model tensor parallelism (MTP), model layer aggregation (MLA), and scalable parallelism for expert models.
On the ecosystem front, Huawei has open-sourced its entire software stack for Ascend—from the CANN (compute architecture for neural networks) kernel to the high-level MindSpore AI framework. More than 100 partners in manufacturing have already built DeepSeek-powered applications on Ascend to address diverse industry needs.
The summit also featured breakout tracks focused on sectors like automotive, electronics, new energy, and computing components, offering industry-specific discussions on AI deployment.
Looking ahead, Guo said Huawei will continue supporting digital transformation across enterprises. In 2025, the company plans to deepen its end-to-end AI offerings—spanning data collection, transmission, storage, computing, management, and application—by delivering a unified intelligent infrastructure stack across connectivity, compute, storage, and platforms.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Deng Yongyi for 36Kr.