Huawei sharpens AI edge in global cloud move

Huawei Cloud is counting on delivering artificial intelligence (AI) services aimed at various use cases to make an impact in global public cloud markets.

Speaking at the Huawei Cloud Summit in Singapore, Edward Deng, president for global markets at Huawei Cloud, said the company offers a full cloud technology stack powered by its own Kunpeng processors and Ascend AI chips, along with pre-integrated cloud services for different industries.

Huawei Cloud’s tightly engineered technology stack, particularly the AI pieces, may well be its competitive edge as it faces rivals Amazon Web Services (AWS), Microsoft and Google in global markets, where demand for public cloud services has been growing.

“We estimate that by 2025, almost 100% of companies will migrate their business to the cloud, 97% of large enterprises will use AI, and 77% of cloud applications will rely on AI,” said Deng, noting that AI will be the key to cloud competition.

Huawei Cloud, which recently opened a datacentre region in Singapore, has been making strides in AI, particularly in computer vision and image recognition. In March 2019, its AI development platform, ModelArts, came out top in both image classification and inference in the Stanford DawnBench deep learning competition.

In the contest, ModelArts took four minutes and eight seconds to train the ResNet-50 model on ImageNet with 128 V100 GPUs (graphics processing units) – more than double last December’s speed of nine minutes and 22 seconds, and four times the speed of running the fast.ai deep learning library on AWS.

In image classification, ModelArts was 70% faster than the second-placed Alibaba Cloud, three times faster than AWS’s performance and eight times faster than Google’s.

However, raw performance in benchmark tests has to be translated into real-world use cases. During his address, Deng described how one logistics company was using Huawei’s AI-powered Violent Sorting Recognition service to detect whether workers were loading packages onto delivery trucks without due care.

Underpinning such services are powerful AI models and algorithms which Huawei has developed on its own and through partnerships with universities. It has also opened a new cloud and AI innovation lab in Singapore to cultivate local AI talent and develop new AI applications.

“Most of the bottlenecks in AI computing today are not caused by computing itself – it’s the algorithms,” said Song Zhexuan, chief strategy officer at Huawei Cloud, in response to queries from Computer Weekly. “Huawei cannot address that alone, so we invest in universities to find new algorithms to solve this issue.”

Huawei is also taking steps to overcome the challenges posed by data protection and sovereignty rules that prevent AI models from being shipped across the globe.

“Data is a sensitive issue for many countries, as each country has its own data export and data controls like the GDPR [General Data Protection Regulation] in Europe, while China also has its cyber security regulations,” said Song.

“For that, there are certain services where we use local data to apply models, and due to data regulations, these models cannot be moved freely from region to region. However, we do have a huge amount of data from other services and partners, and that also comes with some really useful intelligence that we can use in these models.”

But it remains to be seen whether Huawei’s AI smarts and cloud services that address specific use cases will bear fruit.

For one, it is up against formidable rivals, particularly Google Cloud, an AI powerhouse that has been ramping up efforts to make its capabilities more accessible to businesses through services such as AutoML.

Meanwhile, analyst firm Gartner has just ranked Alibaba Cloud top for infrastructure as a service in the Asia-Pacific market.

To compete, Huawei – besides sharpening its AI edge – will need to speed up its time to market, as there is usually a delay between the time a cloud supplier launches services in its home market and the rest of the world.

Song said Huawei typically launches new services on a smaller scale before making them available to a broader customer base.

“If a service is proven to be robust and reliable, and without any issues, it will be immediately launched internationally,” he said. “But if there are security issues or issues due to some regulatory requirements, we would have to delay the launch in one region compared to another.

“Secondly, we have a large customer base in China, even for niche services. For international regions, the customer bases are smaller, and for niche services we often have very few customers, with requirements that are not as high as for China.

“That is the reality and another reason why there is a gap between launching a service in China and worldwide.”

That said, cloud suppliers are already narrowing the service gap. In 2018, Huawei, which operates 40 availability zones in 23 geographical regions, launched TaurusDB, a MySQL-based database service, in China and Singapore simultaneously. The high-performance distributed database is touted to deliver seven times the performance of a native MySQL database.

Colin Quek, an IT industry veteran with more than two decades of experience, told Computer Weekly that Huawei Cloud appears to have what it takes to compete against other cloud suppliers.

“The optimised technology stack is not what others have been talking a lot about,” he said. “And with AI cases still evolving, it helps that they have demonstrated how AI can be applied in different industries.”