This exemplifies how Huawei is developing agent-like AI systems that move beyond simple command-and-response interactions to platforms capable of independent planning, decision-making, and execution.
Huawei’s approach to building these agent AI systems is centered on a comprehensive strategy spanning AI infrastructure, foundation models, specialized tools, and agent platforms.
Huawei Cloud CTO Zhang Yuxin gave an overview of the framework at the recent Huawei Cloud AI Summit in Shanghai. At the summit, more than 1,000 political, business, and technology leaders explored practical implementations across finance, ports of entry, chemical manufacturing, healthcare, and autonomous driving.
This distinction is important because while traditional AI applications respond to user commands within a fixed process, agent AI systems operate with autonomy, fundamentally changing their role in enterprise operations.
Zhang characterized this as a “major shift in applications and computing,” noting that these systems will make decisions independently, adapt dynamically, and reshape the way computing systems interact and allocate resources. The question for enterprises is how to build infrastructure and platforms that can support this level of autonomous operations.
Infrastructure challenges drive new computing architectures
The computational demands of agent AI systems have exposed the limitations of traditional cloud architectures, especially as the requirements for training and inference of the underlying models rapidly increase.
Huawei Cloud support includes CloudMatrix384 supernodes connected through a high-speed MatrixLink network, creating what the company says is a flexible hybrid computing system that combines general-purpose and intelligent computing capabilities.
This architecture specifically addresses bottlenecks in Mixture of Experts (MoE) models through expert parallel inference and reduces NPU idle time during data transfer. According to the company’s technical specifications, this approach improves inference speed on a single PU by 4-5x compared to other popular models.
The system also incorporates memory-centric AI-native storage designed for common AI tasks to increase the efficiency of both training and inference. ModelBest, a company specializing in general purpose AI and device intelligence, has demonstrated a practical application of this infrastructure.
Li Dahai, co-founder and CEO of ModelBest, detailed how the company’s MiniCPM series, which spans basic models, multimodal functions, and full modality integration, is integrated with Huawei cloud AI computing services to improve training energy efficiency by 20% and performance by 10% compared to industry standards.
The MiniCPM model has applications in automotive systems, smartphones, embedded AI, and AI-enabled personal computers.
From basic models to industry-specific applications
The challenge of adapting basic models to specific industry needs has driven the development of more sophisticated training methodologies. Huawei Cloud’s approach includes three main components: a complete data pipeline that handles collection through management, an out-of-the-box incremental training workflow, and a smart assessment platform with preconfigured assessment sets.
Incremental training workflows reportedly improve model performance by 20-30% through automatic adjustment of data and training settings based on core model features and industry-specific goals. Evaluation platforms allow you to quickly set up systems to industry or corporate benchmarks, addressing both accuracy and speed requirements.
Practical implementations demonstrate the practical application of these methodologies. Shaanxi Cultural Industry Investment Group partnered with Huawei to integrate AI and cultural tourism business.
Huang Yong, chairman of Shaanxi Cultural Industry Investment Group, explained that the organization used Huawei Cloud’s data and AI convergence platform to combine diverse cultural tourism data and create a comprehensive dataset spanning history, film, and intangible heritage.
The partnership established what is called a “trusted national data space for cultural tourism” on Huawei Cloud, enabling applications such as asset verification, copyright trading, corporate credit enhancement, and creative development.
This collaboration resulted in the Boguan Cultural Tourism Model, which powers AI-driven tools such as Cultural Tourism Intelligent Brain, Smart Management Assistant, Intelligent Travel Assistant, and AI Short Video Platform.
International implementations show a similar pattern. Dubai Municipality collaborated with Huawei Cloud to integrate foundational models, virtual humans, digital twins, and geographic information systems into city systems. Mariam Almheiri, CEO of Dubai Building Regulations and Permits Authority, spoke about how the integration has improved urban planning, facility management and emergency response.
Introducing an enterprise-grade agent platform
The distinction between consumer-grade AI agents and enterprise-grade agent AI systems focuses on integration requirements and operational complexity. Enterprise systems must integrate seamlessly into broader workflows, handle complex situations, and meet higher operational standards than consumer applications designed for rapid interaction.
Huawei Cloud’s Versatile platform addresses this gap by providing the infrastructure for companies to create agents tailored to their production needs. The platform combines AI compute, models, data platforms, tools, and ecosystem capabilities to streamline agent development through the deployment, release, usage, and management phases.
Conch Group’s introduction into cement production provides specific performance indicators. The company partnered with Huawei to develop what it calls the cement industry’s first AI-powered cement and building materials model.
The resulting cementing agent predicts clinker strength after 3 and 28 days. The predicted value deviates from the actual result by less than 1 MPa, indicating an accuracy of more than 90%. To optimize cement calcination, the model proposes key process parameters and operational solutions that reduce standard coal usage by 1% compared to Class A energy efficiency standards.
Xu Yue, assistant general manager at Conch Cement, said the model’s success in quality control, production optimization, equipment management and safety has established the basis for end-to-end collaboration and decision-making through cement distributors, moving the industry “from traditional reliance on expertise to a complete reliance on data across all processes.”
In corporate travel management, Smartcom uses Huawei Cloud Versatile to develop a travel agency that provides end-to-end smart services across departures, connections, and flights. Kong Xianghong, CTO and Director of Smartcom Solutions at Shenzhen Smartcom, reported that the system combines travel industry data, company policies, and personal travel history to generate recommendations.
Employees adopt more than half of these suggestions and complete reservations within two minutes. Through predictive question matching, agents solve 80% of problems in an average of three interactions.
What’s next for autonomous AI?
The implementations discussed at the summit reflect a broader industry trend toward agent AI systems operating with increased autonomy within defined parameters. The advancement of technology from reactive tools to systems that can independently plan and execute complex tasks represents a fundamental architectural shift in enterprise computing.
However, migration requires significant infrastructure investment, advanced data engineering, and careful integration with existing business processes. Performance metrics from early implementations, such as increasing manufacturing efficiency, improving city management, and optimizing travel bookings, serve as benchmarks for organizations evaluating similar implementations.
As agent AI systems continue to mature, the focus appears to be shifting from demonstrating technical capabilities to operational integration challenges, governance frameworks, and measurable business outcomes. Examples from cement manufacturing, cultural tourism, and corporate travel management suggest that practical value arises when these systems address specific operational pain points, rather than acting as general-purpose automation tools.
(Photo provided by: AI News)
SEE ALSO: Huawei details open source AI development roadmap at Huawei Connect 2025
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
