Huawei Cloud CEO Zhou Yuefeng: The core determinant of model competition has already shifted to "post-training".

Huawei Cloud CEO Zhou Yuefeng The focus of the global AI competition has shifted from general model capabilities to practical industry application capabilities. In this new business race, how should enterprises build their exclusive AI competitiveness? On March 20, at the Huawei China Partner Conference 2026, Huawei Cloud presented its own approach to solving the problem. Huawei Cloud CEO Zhou Yuefeng pointed out that public cloud is the optimal solution for enterprise AI implementation and the best platform to carry future AI productivity. In response to the current "hundreds of models competing" industry trend, he clearly proposed a path based on open-source openness and constructing core differentiation through "post-training," aiming to provide enterprises with practical, exclusive, and competitive AI solutions. New Winning Move: Post-Training? As the industry enters the era of intelligent agents, Zhou Yuefeng put forward a forward-looking industry insight: the core winning move of model competition has already shifted to "post-training" capabilities. Although general large models are "knowledgeable," they often lack deep industry know-how. The core value of post-training is precisely to enable these general large models to fully absorb exclusive industry knowledge and precisely adapt to specific business scenarios. Based on this, Huawei Cloud has developed a post-training suite covering the full technical capabilities from CPT (Continuous Pre-Training), SFT (Supervised Fine-Tuning), to RL (Reinforcement Learning). Through this suite, enterprises can deeply inject their industry knowledge into top-tier base models and integrate unique Huawei ecosystem "nutrients" such as HarmonyOS coding capabilities and Ascend operator optimization. It is reported that this combination not only enhances the model's accuracy in specific business scenarios, but also optimizes HarmonyOS's adaptability and the efficiency of Ascend's underlying computing power. At the same time, Zhou Yuefeng revealed that Huawei Cloud has a dedicated R&D team for reinforcement learning and post-training. This team conducts post-training and reinforcement learning for self-developed, open, and open-source large models across various industries and scenarios, granting them more differentiated capabilities. This is Huawei Cloud's strategy for models. Why Public Cloud Is the “Optimal Solution” Zhou Yuefeng pointed out that at the critical stage where AI is transitioning from technical exploration to large-scale application, the advantages of public cloud over self-built data centers are becoming increasingly apparent. Data shows that by 2025, 85% of global AI computing resources will be deployed in the cloud, and over 87% of enterprises will choose the cloud for AI business and innovation practices. Cloud has become the absolute mainstream for AI investment. The business logic behind this is not complex. In terms of cost and talent, offline data center construction faces heavy burdens of long build cycles and massive investment, while top AI talent is in acute shortage on the market. In contrast, the cloud gathers vast AI computing clusters and numerous AI engineers and algorithm design engineers. Accessing AI capabilities on the cloud not only achieves optimal costs, but also effectively alleviates severe talent anxiety, allowing enterprises to focus their precious energy on core business logic and AI innovation itself. In terms of security, a priority for enterprises, public cloud provides protection far superior to self-built environments. Taking Huawei Cloud as an example, relying on its intelligent unified operation solution, up to 99% of security threats can be closed within 5 minutes, and 99% of network attacks can be automatically handled, solidifying the security baseline for enterprise AI assets. More critically, faced with the rapid weekly iteration trend of AI large models, the traditional offline deployment model cannot keep up with frequent upgrades to stay abreast of technology. The public cloud model ensures enterprises always access the latest computing resources and continuously receive the most advanced AI capabilities. Adhering to Open Source Zhou Yuefeng also emphasized that Huawei Cloud always adheres to the principle of open-source openness. Regarding the large model strategy, Huawei Cloud has fully opened the self-developed Pangu large model, launched the full-size version matrix from 718B to 1B, and open-sourced it. It also supports more than 160 industry-leading SOTA models for out-of-the-box use, including DeepSeek, Qwen, and Zhipu GLM, among other mainstream high-quality models. Meanwhile, Huawei Cloud has achieved rapid response of top models "launching immediately upon release," such as Zhipu GLM-5's "Day 0" onboarding, providing enterprise developers with the richest and most cutting-edge arsenal. Facing the intelligent era, Huawei Cloud aims to become the "black soil" for enterprise-level AI innovation. It is reported that Huawei Cloud CodeArts intelligent code agent started public testing in February this year; Huawei Cloud’s one-stop enterprise-grade intelligent agent development platform AgentArts is slated for enterprise commercial beta in April, and the enhanced version of openJiuwen is scheduled for official open-source release in May. Risk Warning and Disclaimer The market involves risks, and investment must be cautious. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situations. Investing based on this article is at your own risk.