Huawei Cloud releases AI-native cloud infrastructure CloudMatrix and Huawei Mainframe-to-Cloud Solution

Tao Jingwen, Huawei's Director of the Board and President of the Quality, Business Process & IT Mgmt Dept, recounted Huawei's own digital transformation journey to showcase the company's comprehensive intelligent upgrade strategies. In another highlight, Bruno Zhang, CTO of Huawei Cloud, presented insights on how Huawei Cloud is leveraging AI to reshape data centers, infrastructure, and cloud services. His keynote addressed the construction of an AI-native cloud designed to accelerate intelligence in industries.

Thinking AI-native and accelerating the reshaping of industries with AI

The intelligent era is coming fast. AI is being integrated further into industry applications to accelerate product time-to-market, transform scientific computing, and improve efficiency. Zhang states that enterprises need to seize opportunities in the intelligent era and use AI to build their competitive advantages. The key is to adopt an AI-native mindset now. Enterprises should position AI technologies and tools as core elements as they rethink and design their processes, IT architectures, and business innovations, so as to fully unleash the potential of AI, boost efficiency, innovate business models, and solve complex problems.

Zhang Ping'an, Executive Director of the Board of Huawei and CEO of Huawei Cloud

"First, we must actively embrace AI, open up industry scenarios, and develop enterprise AI platforms. This will enable AI to serve the core business of enterprises. Our practices have shown that the earlier we introduce AI, the faster we can benefit from it. Second, AI computing power is critical. Enterprises need to build AI-native cloud infrastructure that matches their requirements. Third, data quality determines the effectiveness of AI models. We need to build knowledge-centric data foundations so that data can serve AI better, other than just being used for business analysis. Fourth, we need to build suitable AI models based on business applications." Zhang went on to emphasize that, "We must abandon the misconception that larger models are better. It is not feasible to address all manner of needs with just one foundation model."

Building AI-native cloud infrastructure with diversified computing power, elasticity, and efficiency

Customers' demand for AI computing power is growing rapidly, and such demand is predicted to surpass the demand for general-purpose computing power by the end of this year. Zhang Ping'an said that the key to AI development is an infrastructure capable of providing the optimal diversified computing power with elasticity and efficiency. This means that building an AI-native cloud infrastructure is crucial.

In the intelligent era, model parameter scales have increased from tens of billions to hundreds of billions or even trillions. Zhang officially released CloudMatrix, a cloud infrastructure architecture designed to meet the explosive demand for computing capacity, high reliability, and high efficiency. CloudMatrix interconnects and pools all

resources such as CPUs, NPUs, DPUs, storage, and memory. It marks an evolution from monolithic to matrix computing power. CloudMatrix comprises an AI-native cloud infrastructure in which everything can be pooled, peer-to-peer, and composed, providing robust AI computing power to meet customers' needs.

AI-native cloud infrastructure CloudMatrix is officially released

Building knowledge-centric data foundations

Currently, the data platforms of many enterprises are not suitable for foundation models. A large amount of data needs to be cleansed and knowledge needs to be extracted. Zhang Ping'an pointed out that in the intelligent era, enterprises need to build knowledge-centric data foundations that serve AI models.

To simplify data use and improve model training efficiency, Huawei Cloud has fully upgraded DataArts to provide customers with AI-oriented and knowledge-centric data foundations. The updated features include AI + data convergence engines, data development and governance, knowledge services, and AI + data application enablement services. The upgraded DataArts significantly improves resource utilization and data supply efficiency.

Building multimodal models of different sizes

Zhang Ping'an also noted that the diversity of enterprise applications calls for the construction of multimodal models of different sizes. Only in this way can enterprises match applications and models and meet their requirements for cost-effective and professional foundation models.

In June this year, Huawei Cloud released Pangu Models 5.0, which is available in different sizes, such as with parameters in the billions, tens of billions, hundreds of billions, and trillions. The complete series of Pangu models, including NLP, CV, multimodal, prediction, and scientific computing models, meet practically all business application needs.

Huawei and the municipal government of Shenzhen Bao'an District jointly established the Embodied AI Industry Innovation Center in Qianhai. Through cloud-device intelligence synergies, embodied AI greatly reduces the complexity of design on the device side and improve the generalization and task execution of embodied AI in industrial applications. Through device-cloud intelligence synergies, the Innovation Center has improved the operation precision of robotic arms from millimeters to hundreds of micrometers, achieving a 99.99% success rate for dynamic insertion of rigid parts.

In addition, Huawei Cloud's Pangu Models 5.0 enhance spatiotemporal controllable generation (STCG) in the field of multimodal generation. Take autonomous driving as an example. Pangu Models 5.0 can generate not only videos of normal driving scenarios, but also videos reflecting random road conditions, as well as accidental and aggressive driving. This greatly reduces the dependency of autonomous driving training on massive road data collection.

"The intelligent world is approaching fast. We need to think AI-native," Zhang Ping'an said. He called for enterprises to actively embrace AI, make high-value enterprise applications available, and set the flywheel of AI models, computing power, and data in motion to drive business forward and unleash the boundless potential of innovation.

Mainframe migration: The last mile in enterprise cloud migration

Core systems are the most important business systems for all enterprises. In recent years, enterprises such as banks have continued to accelerate the shift from core system deployment on mainframes to deployment on the cloud, in order to respond to the changes in the market and the intelligent era with greater agility. Zhang Ping'an pointed out that core system migration is the last mile on the journey of enterprise cloudification and the way forward to embrace full-scale intelligence.

At this conference, Zhang Ping'an announced the global official release of the Huawei Mainframe-to-Cloud Solution. This solution helps customers build new core systems on the cloud with high availability, easy O&M, and better agility. Based on the distributed cloud foundation of Huawei Cloud, this solution provides the capability of detecting hardware faults in seconds. Based on Elastic Load Balancing (ELB) and financial-grade distributed middleware, this solution supports cell-based service deployment. Based on the GaussDB distributed database, the solution implements fast and large-capacity service expansion and multi-site multi-active disaster recovery (DR), building 99.999% financial-grade high availability on the cloud. The solution is designed to detect faults

within 1 minute, demarcate faults within 5 minutes, and rectify faults within 10 minutes. In this way, Huawei has built intelligent fault demarcation capabilities to make large-scale cloud infrastructure O&M more efficient. Huawei also pioneers the in-place seamless upgrade technology of operating systems. During the upgrade, VMs do not need to be migrated. In addition, batch concurrent upgrade is supported. The upgrade time of thousands of mainframe nodes is shortened from dozens of days to less than 5 hours, greatly improving the platform evolution efficiency.

The Huawei Mainframe-to-Cloud Solution is officially released

Mainframe cloudification is not only a simple replacement of mainframe or midrange computers, but an architectural innovation and user experience upgrade for core systems. Currently, most banks in China have chosen Huawei to build their new core systems on the cloud.

Intelligent upgrade: Driving ongoing quality growth for enterprises

AI presents the greatest revolution of our times, and AI and cloud are two super-powerful tools reshaping countless industries.

Tao Jingwen, Huawei's Director of the Board and President of the Quality, Business Process & IT Mgmt Dept, stated that any enterprise that wants to make effective use of AI must first upgrade their AI data governance system. The large amounts of structured data accumulated in the information era and the many documents previously generated from corporate operations cannot be directly recognized and studied by AI. "Enterprises must deeply integrate AI models with their data toolchains and build on their traditional data governance and data platforms to develop an AI security governance mechanism and an AI digital pipeline. Then, supported by Huawei Cloud's new AI computing platforms, they will be able to develop new corporate AI services," said Tao.

Tao Jingwen, Huawei's Director of the Board and President of the Quality, Business Process & IT Mgmt Dept

AI will usher in a new revolution. Enterprises that want to make the most of AI must deeply integrate it into their processes, organizations, IT, data, and business scenarios to provide services that can create business value. Based on its own experience, Huawei has developed a methodology for this, which can be broken down into three layers, five phases, and eight steps. The three layers are redefining intelligent business, AI model development and delivery, and ongoing optimization of intelligent applications. The five phases they must go through are identifying suitable scenarios, reshaping business processes, transforming organizations, optimizing corporate data, and adopting AI applications. Tao pointed out that not all scenarios are suitable for AI adoption, which is why Huawei has developed 12 criteria to determine which scenarios are ready for and will benefit from AI.

Tao emphasized that while implementing the intelligent upgrade, Huawei focuses on scenario-driven AI adoption for numerous, repetitive, and complex tasks that require lots of time and effort. "Multi-modal systems engineering built with multiple models is the key to AI adoption in enterprises," Tao continued.

He then shared three best practices from Huawei. The first is Huawei's digital solution for contract-related objects, processes, and rules. The solution allows employees to concurrently and efficiently handle numerous contracts, and supports the intelligent extraction and comparison of key contract terms and conditions in different languages, reducing the time taken to complete contract risk reviews from two hours to 5 minutes.

Second, Huawei has equipped its 110,000-plus R&D engineers with a development assistant. This tool uses foundation models to automatically extract task-context information for line-level code writing, function generation, and code interpretation and annotation. Now, 7 million lines of AI-generated code are accepted every year.

Third, in its manufacturing domain, Huawei adopts a multi-model systems engineering approach and combines decision-making and generative AI models to develop a range of intelligent applications, such as plan solvers, AI-assisted visual quality inspection, predictive equipment maintenance, and manufacturing knowledge enablement. This has greatly improved the company's overall manufacturing productivity, and shortened its order delivery periods by more than 30%.

Huawei has developed an AI digital pipeline to better integrate AI with its business processes and data. Tao emphasized that enterprises must load their own data into foundation models and conduct model optimization training to develop their own business-specific models. An enterprise's datasets are its proprietary and business-specific data, knowledge, and information, so the best way to address the uncertainty surrounding AI model results is to constantly improve data quality until it reaches "textbook" level.

Tao concluded his speech by saying that transparency breeds trust and trust promotes sharing. Huawei has developed a digital space governance model for enterprises based on its own experience in digital and intelligent operations as well as industry best practices. This model provides an efficient, intelligent, transparent, and trustworthy digital governance blueprint that is built upon unified data and language. This blueprint can help enterprises achieve digital transformation and intelligent upgrade and drive their ongoing quality growth.

Cloud for AI x AI for Cloud: Forging an AI-Native Cloud

Huawei Cloud's AI-native strategy has been a cornerstone of the company's innovation over the past year. This strategy is twofold: AI for Cloud and Cloud for AI, marking significant advancements in Huawei Cloud's capabilities.

In the realm of Cloud for AI, Huawei Cloud has transformed traditional data centers into AI compute centers on the cloud, offering customers efficient, green, and reliable AI compute on a large scale. A highlight of this transformation is the introduction of CloudMatrix, a next-generation AI-native matrix compute architecture where everything can be pooled, peer-to-peer, and composed, instead of traditional monolithic and general-purpose compute.

Bruno Zhang, CTO of Huawei Cloud, detailed CloudMatrix's innovative use of a new high-speed bus to interconnect multiple resources, achieving resource pooling from the cluster level to the data center level, which aligns with the demands of the AI era for compute in terms of scale, efficiency, stability, and scalability.

Bruno Zhang, CTO of Huawei Cloud

In AI for Cloud, Huawei Cloud integrates Pangu models to build intelligence into cloud services. Pangu Doer, Huawei Cloud's intelligent assistant, provides precise responses based on enterprise data, streamlining customer interactions with Huawei Cloud.

The "1+N" Pangu assistant system, unveiled by Mr. Zhang, signifies a new era in cloud service interaction. The "1" represents Pangu Doer, which supports the entire cloud journey, from planning, use, maintenance, to optimization, with intelligent services like knowledge search, data query, and automated execution. The "N" denotes the myriad use cases in product R&D, data analytics, security, and office collaboration, where Pangu models are trained using the scenario-specific data, know-how, and practices to enhance cloud services and boost efficiency with tailored AI.

Launch of Huawei Cloud Pangu Doer

A case in point is the Postal Savings Bank of China's utilization of Pangu Doer in CodeArts to build an intelligent development platform that streamlines the work of over 4,000 R&D engineers. This platform leverages AI-powered software analysis, code RAG, and other cutting-edge technologies to provide features like code generation, unit test (UT) generation, and technical Q&A. With over 30% of AI-generated code and over 60% AI-generated UT code being ready-to-use, the platform has automatically generated more than 290,000 lines of high-quality code, which efficiently supports the development of over 200 application systems. Pangu Doer is now available on the Huawei Cloud website.

Huawei Cloud's commitment to industry collaboration was evident at HUAWEI CONNECT, where a variety of sessions, roundtables, and showcases took place. These events, spanning three days, highlighted innovative technologies and practices in cloud infrastructure, core systems to cloud, application modernization, data governance, and AI enablement across industries, all aimed at amplifying industry intelligence.