Huawei Cloud Aims to Accelerate Intelligence for Industries

AI took center stage inside Huawei’s booth at Mobile World Congress 2024 (MWC), where the company highlighted new capabilities using its Pangu models.  These updates aim to make it easier for industries including telecoms to deliver cloud services and solve industry challenges.

There are stark differences between Huawei Cloud’s Pangu Models and ChatGPT, according to Bruno Zhang, CTO of Huawei Cloud.

“Unlike LLMs such as ChatGPT, Pangu provides a three-layer decoupled architecture to better meet the different needs: The L0 layer consists of five foundation models, which provide general skills to power an endless possibility of industry-specific applications.  L1 contains industry-specific models, trained using industry data. L2 provides models for specific industry scenarios and tasks.”

Huawei launched its cloud business seven years ago.  In the telecom industry, it offers an open platform to telcos to innovate and deliver new services.  To enable that, the company has implemented a two-pronged approach: AI for Cloud and Cloud for AI.  Pangu is at the heart of both.

AI for Cloud

“AI for Cloud” uses AI to reshape software development and cloud services, democratizing the specialties for business personnel so that they can become developers. The latest release of Pangu 3.0 enables that by building models and generating code.  

According to Zhang, telcos can generate code using the Pangu R&D model with just one click.  He adds that the Pangu telecom model can automatically troubleshoot 90 percent of network faults in just minutes; and the Pangu virtual human model boasts 95 percent lip sync accuracy for customer service and livestreaming.

Cloud for AI

“Cloud for AI” uses an intelligent cloud infrastructure to address challenges related to storage, database, data and deployment model, etc.  The strategy focuses on AI-native storage, GaussDB database, Data AI-convergence and a distributed architecture called QingTian.  According to Huawei Cloud, QingTian surpasses the limitations in compute, storage, and networking for a top-class AI compute backbone with heterogeneous, peer-to-peer, full-mesh computing..  

To train industry specific models, exabytes of data is needed. Huawei Cloud solves this challenge by providing anAI-Native storage system with a layered structure. “We first offer a memory service called EMS. It stores petabytes of parameters with ultra-large bandwidth and ultra-low latency. We also offer SFS Turbo cache services with high throughput and concurrency. With IOPS in the tens of millions, data no longer needs 100 hours to prepare, just 5. Finally, a knowledge lake built on Object Storage Service reduces the cost of storing training and inference data by 30%”, said Bruno Zhang.

To develop telecom models specifically, Huawei has unified data from the BSS, OSS, and MSS domains into one data lake.

According to Bruno Zhang, “With LakeFormation, one copy of data can be used for both data processing and AI training without migration. DataArts Studio, makes the entire data governance process more intelligent, from data integration, development, to quality and asset management. And then our three pipelines – DataArts, ModelArts, and CodeArts – orchestrate and schedule data and AI workflows to drive online model training and inference with real-time data.”

Huawei Cloud GaussDB is an enterprise-class distributed database that's widely used in finance, telecom, and government sectors with over 100,000 nodes. GaussDB ensures dual cluster strong consistency with zero RPO, which can meet high availability requirement of telco. It is certified CC EAL4+, the highest security level in the industry. To automate migration, GaussDB provides one-stop tools to convert 95% of common syntax to reduce the cost and improve the efficiency of migration.

The AI Cloud Opportunity

According to Gartner, more than 80 percent of businesses will harness AI generated content in production by 2026.  It is a rapid pace of transformation that demands enterprises to invest in intelligent cloud infrastructure, which Huawei said is already starting to take shape.

More than 800 government clouds and more than 300 financial institutions are built on Huawei Cloud. Specifically in China, 90% of the top 50 e-commerce companies and 90% of top 30 automotive companies have migrated their services to Huawei Cloud.

Huawei Cloud is expanding its business globally. Last year, localized cloud services were launched in Türkiye and Saudi Arabia. Up to now, Huawei Cloud has 85 AZs in 30 regions around the world, serving customers in more than 170 countries and regions. Enterprises can access Huawei Cloud services without building their own data centers to expand the global market. Huawei Cloud will also launch localized cloud services in Egypt and the Philippines. Huawei Cloud will launch AI cloud services in Hong Kong to provide large-scale and efficient intelligent computing.