Huawei Cloud has made a significant impact on the technology landscape with its recent announcements at the Huawei Developers Conference 2025 (HDC 2025). The company unveiled the latest iteration of its Pangu Models, version 5.5, along with its next-generation AI cloud service, solidifying its position as a leader in the development and application of AI technologies across various industries globally.
Zhang Ping’an, Executive Director of Huawei and CEO of Huawei Cloud, stated, «We are redefining the technological landscape with solutions that not only increase computational efficiency but also transform entire industries.»
Pangu Models 5.5: Comprehensive Evolution for Specific Industries
Huawei presented its new Pangu 5.5 models, which represent a complete overhaul of Huawei Cloud’s AI ecosystem, with significant improvements in five fundamental capabilities:
Natural Language Processing (NLP) Model
The new 718B deep-thinking model, with a Mixture of Experts (MoE) structure comprising 256 specialists, dramatically enhances reasoning, tool invocation, and mathematical capabilities. The adaptive integration technology between fast and slow thinking enables the model to provide agile responses to simple problems and in-depth analysis for complex challenges, multiplying inference efficiency by eight.
Pangu Multimodal Model
The innovative Pangu Global Model generates digital physical spaces for training robots and autonomous driving systems. A notable example is its application at Guangzhou Automobile Group, where it allows the reproduction of exceptional cases in complex scenarios within minutes, accelerating the iteration of autonomous driving models.
Pangu Prediction Model
It implements the industry’s first unified pre-training architecture, processing data from various industrial sources. For instance, China Baowu Steel Group has achieved a quality rating above 90% in molten iron temperature using this model, saving 2 kg of fuel per ton.
Pangu Scientific Computing Model
Applied in accurate weather forecasting, as demonstrated by its implementation at the Shenzhen Meteorological Office and the Chongqing Meteorological Service to predict extreme weather conditions.
Pangu CV Model
With 30 billion parameters, it is the largest computer vision model in the industry, enabling panoramic perception, analysis, and decision-making. CNPC has developed the Kunlun Large Model based on Pangu, achieving a 40% higher efficiency in detecting sub-millimeter defects in pipelines.
«The Pangu Models 5.5 are not just incremental updates; they represent a qualitative leap in our AI capabilities, especially in specific industrial applications,» emphasized Wang Yunhe, Director of Huawei’s Ark Lab.
It is worth noting that over the past year, Pangu Models have been implemented in more than 500 scenarios across 30 different industries.
Next-Generation AI Cloud Service
Huawei Cloud’s new AI cloud service utilizes a technology called CloudMatrix 384 supernodes. This innovation connects 384 specialized processors (NPUs) and 192 regular processors (CPUs) through a high-speed network, effectively creating a supercomputer that works four times faster than traditional systems, processing 2,300 information units per second.
These supernodes are highly flexible and efficient, able to better distribute work across different tasks simultaneously, reduce waiting times, and utilize up to 50% more of their total capacity. For very large tasks, such as training AI models with billions of data points, up to 432 supernodes can be connected, creating a giant system with up to 160,000 processing units in the data center.
Through these advancements, Huawei reiterates its leadership in the development and application of Artificial Intelligence technologies, which are being developed in parallel with progress in connectivity equipment and intelligent digital solutions, ranging from modular data centers to the latest-generation smartphones.
