This market research report was originally published at Tractica's website. It is reprinted here with the permission of Tractica.
The recently concluded Huawei Analyst Summit 2019 (HAS 2019) in Shenzhen didn’t have much to offer in terms of new announcements around artificial intelligence (AI). It did, however, bring to light Huawei’s increased focus on AI as it embeds AI into its products and services and across devices, equipment, and the cloud.
The company first announced its full-stack AI portfolio at Huawei Connect 2018, which had much more substance around AI. Most of the discussions at HAS 2019 were a continuation of that full-stack AI strategy, providing a few details on product roadmaps and specifications. Overall, HAS 2019 provided a better understanding of how Huawei’s customers are starting to use the company’s AI capabilities and how its strategy is progressing on the ground.
5G Networks Will Ultimately Drive Network AI
Not surprisingly, 5G was a key theme at the summit. Huawei announced that it has 40 commercial contracts for building 5G networks, but no contracts yet in mainland China. This surprised most analysts, as Huawei has been facing a lot of headwind from the U.S. and European governments around security issues.
From an AI perspective, the contracts for 5G networks are possibly good news. 5G is seen as an enabler for AI because it helps networks transition more quickly to software-defined networking (SDN) and network functions virtualization (NFV), which are drivers for AI-driven network monitoring and optimization. Huawei’s SoftCOM AI solution, which specifically targets AI network monitoring and fault detection among other use cases, should eventually get traction as the deployed base of Huawei 5G equipment increases. As reported in his detailed blog covering the SoftCOM AI roundup from HAS 2019, Mark Beccue opines that, “… there appears to be a lack of urgency around the primary pain point – the significant reduction of operating expenses. Driving down costs is the key selling feature of the near-term use of AI-driven network automation.”
This is not news to Tractica, as the current predicament has been accounted for in our market forecasts in Artificial Intelligence for Telecommunications Applications. Our central hypothesis has been that 5G and SDN will be the key drivers for AI in telecom networks, and we see an inflection point coming around 2021-2022. While Huawei seems to be ahead of its competition in driving network AI and building a comprehensive portfolio of SoftCOM AI solutions, the traction on the ground has been slow, which matches Tractica’s hypothesis.
Huawei Envisions AI Chips in All Devices and Equipment
Huawei is bold in placing a bet on AI, as it sees AI chips being embedded in every device and equipment that it sells. Its portfolio of Ascend AI chipsets ranges from the Ascend 910 to Ascend 310 and the recently announced Kunpeng 920 ARM-based central processing unit (CPU) processor. The Ascend 910 is a training-focused chip for the cloud and data center while the 310 is an inference chip for end-user devices. Huawei told Tractica that the Ascend 910 will be shipping in the next month or so.
The Ascend 310 is embedded into the Atlas AI platform that comes in different form factors like accelerator modules for cameras and drones. Huawei’s security camera portfolio of software-defined cameras (SDCs) uses the Ascend 310 chipset. The company also mentioned to Tractica that it is seeing activity around Ascend 310 for drones and smart autonomous retail. However, security cameras represent the largest use case, and Huawei’s security camera business is growing faster than its server business. The inclusion of the Ascend 310 in the cameras allows them to process 200 faces or objects on the camera itself, which is an industry-leading metric according to Huawei.
More later about Huawei’s growth for its public safety business. AI-enabled cameras are not just being used for security. For example, a construction company in China is using Huawei’s SDC solution for construction site monitoring and a utility provider is using it for power line monitoring.
Huawei also anticipates its base station and other wireless equipment, both radio access equipment and core network equipment, will have AI chipsets in them eventually. Although its not clear whether the Ascend 310 will be used for that purpose, these chipsets will drive new AI edge capabilities for Huawei. In one of its presentations, Huawei talked about its development of the edge cloud and how AI will increasingly become one of the workloads being deployed on the edge cloud. This squares well with Tractica’s own analysis that at some point in the next 1 to 2 years, we will see the 5G AI edge data center proposition develop into reality.
AI Software Platforms Becoming Edge Compatible
In addition to the hardware, Huawei also provided more detail on the software and compiler stack, including ModelArts, which is its AI development platform. The company claims to already have 1 million developers using ModelArts, but most of these developers look to be from academic research groups in China. Huawei hasn’t provided any official breakout of the developers by sector, but based on our discussions, it looks like enterprise AI developers represent a small share.
Huawei also mentioned use of the open-source Kubernetes edge-based software complier known as KubeEdge, which has its origins in Huawei’s Cloud Intelligent Edge Fabric (IEF) service. This compiler provides the infrastructure and support to orchestrate the running of container-based applications at the edge and allow for app deployment and meta data synchronization between cloud and edge. Huawei specifically discussed and showcased applications using KubeEdge, including quality inspection, optical character recognition (OCR) application for delivery agents, face recognition, and driver behavior analysis.
As AI edge applications increase over time, and as Huawei drives AI chipsets into every device and equipment, we shall see greater usage of KubeEdge. This is in line with other edge-based development solutions like Google Coral and NVIDIA JETPACK, although KubeEdge adds additional capabilities of containerization to edge computing. It is also open source and supports multiple hardware chips, including those from NVIDIA and ARM.
Huawei AI Extending to the Cloud, Enterprise, and Public Safety
Huawei has reorganized its enterprise cloud business unit into Cloud and AI, which now includes its existing enterprise cloud business and the full-stack AI portfolio of solutions. The company’s advantage of owning and developing its own hardware and software stack allows it to offer cost-effective Cloud AI services. The Kunpeng ARM processor further strengthens Huawei’s ability to provide lower cost cloud compute, as many enterprises find graphics processing unit (GPU) instances on the cloud too expensive. Huawei’s cloud services extend across public, private, and hybrid cloud, including providing full-stack AI container services.
There were several enterprise customers at HAS 2019 that showcased the use of Huawei’s Cloud and AI services, including Vanke (real estate developer), SF DHL (logistics), and KingMed (medical diagnostics). However, it seems like Huawei is getting maximum traction and growth from its Cloud services through public sector and government contracts. Among Huawei’s largest cloud customers, the Chinese government plays a big role across all levels of government, at the national, provincial, and city/county levels. The percentage of those government customers using Cloud AI is not clear, although Huawei’s largest growth for enterprise has come from public safety (38% growth year-over-year with 6,000 employees and 4,000 R&D staff), which includes many government customers. Interestingly, the company told Tractica that the internal use of Huawei Cloud is significant, although those are not counted within its cloud revenue.
One of the highlights from the enterprise track at HAS 2019 was the presentation by Shenzhen Big Data Research Institute of how it is transitioning from a city services aspect to a data-driven economic infrastructure. It is breaking down the silos of hierarchical government departments into a more collaborative and flat data-sharing structure. This is what is happening across China, and Huawei’s full-stack AI solution is playing a major role within this transformation. The company is applying its elastic cloud capabilities powered by containers to drive AI from the cloud to the edge.
Huawei also elaborated on the public safety and surveillance projects that it has done in Shenzhen. Cameras are not just being used for facial recognition, jaywalking, and other public safety use cases, but also to check for fake license plates and cars not following rules. Using its Ascend 310-based cameras, Huawei can add new traffic rules to its cameras and control traffic lights based on traffic congestion. As a result, Shenzhen has been able to reduce traffic congestion by 70%, and Huawei claimed the crime rate in the city has fallen by 28%. The company also showcased video surveillance projects in Brazil and Argentina and talked about growth in other Asian, African, and Middle East markets.
Video surveillance continues to be one of the largest AI use cases in Tractica’s taxonomy of more than 300 AI use cases, and Huawei’s growth and success in this space only strengthen our outlook. However, it looks like video surveillance has many more aspects to it than just safety and security, which is something we will need to watch carefully.
Aditya Kaul
Research Director, Tractica