More than 80% of enterprises will use generative AI by 2026 to improve operational efficiency and enhance customer experience, according to Gartner. However, it can be challenging to build the right foundation models to power generative artificial intelligence (AI) to complete a variety of tasks.
This is why Huawei Cloud offers a two-pronged strategy to help organisations reap the full benefits of AI. “The first is ‘AI for Cloud’, which uses AI and foundation models to elevate their experience. They are reshaping industry applications and cloud services, including software development and digital content production. Secondly, ‘Cloud for AI’ makes AI adoption seamless and efficient. Our architectural innovation, AI-native storage, and data-AI convergence solutions will empower organisations to train and use AI like never before,” Huawei Cloud’s CTO Bruno Zhang shares at the Huawei Cloud Summit, on the sidelines of Mobile World Congress Barcelona 2024.
AI for Cloud
Under the AI for Cloud strategy, Huawei launched its large language models (LLM) called Pangu Models 3.0 to make it easier for organisations across industries to adopt generative AI.
Pangu provides a three-layer decoupled architecture to meet different needs. Zhang explains: “The L0 layer consists of five foundation models that offer general skills, while L1 contains industry-specific models trained using industry data. As for L2, it provides models for specific industry scenarios and tasks.”
For example, organisations can use the Pangu virtual human model to create AI characters that look and behave like humans. Trained with recorded live streams, the AI model can automatically generate scripts for virtual humans to sell products accurately and fluently. Pangu-based virtual humans only need to be trained in the organisation’s native language once for them to speak over 20 languages. They can also automatically capture on-screen comments and interact with viewers in real time.
See also: Huawei helps accelerate ICT talent cultivation for mutual benefits in intelligent education
Huawei also offers industry-specific LLMs, such as the Pangu automotive model. The model can automatically generate corner cases (or unexpected scenarios) for various complex driving scenes. It uses real-life photos and videos to create a digital twin space with movable objects, customisable weather, and lighting. This helps generate scenario samples for autonomous vehicles to cope with cases like overtaking and encountering animals on the road. Huawei claims that autonomous vehicles using the Pangu automotive model can learn new scenarios in less than two days.
Cloud for AI
The Cloud for AI strategy aims to help organisations ensure their cloud infrastructure can support the use of AI effectively. This is because cloud computing nurtures the development of AI models, but AI has complex demands for the cloud. According to Huawei, organisations want collaborative heterogeneous computing architecture, cloud-native compute with superlative performance, mass data storage, security compliance, lean governance, and flexible deployment to support AI.
See also: Incubating technology projects to foster an inclusive and sustainable Smart Nation 2.0
In response, Huawei Cloud delivers 10 innovations so that the cloud can provide an optimal AI backbone.
They are:
- KooVerse, which is Huawei’s global infrastructure covering compute, storage, networking, and security.
With 85 availability zones in 30 regions across more than 170 countries, Huawei Cloud offers low latency of 50 milliseconds. - Distributed QingTian architecture.
QingTian surpasses the limitations in compute, storage, and networking for a top-class AI compute backbone with heterogeneous, peer-to-peer, full-mesh computing. - AI compute.
AI Cloud Service supports trillion-parameter model training, and training jobs can run uninterrupted on a cluster over thousands of cards for 30 days, 90% of the time. It provides over 100 Pangu model capability sets and 100 adapted open-source large models out-of-the-box. - AI-native storage that can handle the massive volume of data that training models require.
Huawei Cloud uses expanded memory specification (EMS) service to store data parameters with ultra-large bandwidth and ultra-low latency; scalable file service (SFS) Turbo cache service for high throughput; and objects storage service knowledge lake to reduce 30% costs in storing training and inference data. - End-to-end security.
Huawei Cloud ensures security and compliance across the full lifecycle spanning model runtime environments, training data, AI models, generated content, and applications. - GaussDB, the next-generation database featuring high availability, security, performance, flexibility, intelligence, and simple and smart deployment and migration. It is certified CC EAL4+, the highest level in the industry.
- Data-AI convergence.
Huawei Cloud LakeFormation unifies multiple data lakes and data warehouses so that only one copy of data is used by multiple data analytics and AI engines. Three collaborative pipelines — DataArts, ModelArts, and CodeArts — then orchestrate and schedule data and AI workflows. They drive online model training and inference with real-time data. Smarter data governance is also enabled by the AI4Data engine. - Media infrastructure to enable efficiency, experience, and evolution.
In terms of efficiency, Huawei Cloud MetaStudio (a content production pipeline) can generate high-quality content quickly. Meanwhile, Huawei Cloud Live, Low Latency Live, and SparkRTC enable more seamless live experiences. As for evolution, Huawei Cloud provides generative AI-generated content and 3D space services with real-time user interaction. - Landing Zone, which offers unified account, identity, permissions, network, compliance, and cost management on Huawei Cloud.
- Flexible deployment to allow all Pangu model capabilities and services to work in the public, dedicated, or hybrid clouds.
Additionally, Huawei Cloud is building an open ecosystem to help partners capture opportunities from the cloud and AI. “We believe in helping our partners grow alongside us, and this goal is now backed by our GoCloud and GrowCloud programmes, wherein Huawei works with partners to develop scenario-based solutions for industries. We’re building a solid cloud foundation for everyone and every industry to accelerate intelligence,” says Jacqueline Shi, President of Huawei Cloud Global Marketing and Sales Service.
