HSBC Partner Content

Fang’s A-Z of China AI

Industry expert on how Chinese firms are doing in artificial intelligence


Helen Fang: AI expert

By far the greatest danger of Artificial Intelligence is that people conclude too early that they understand it,” warned Eliezer Yudkowsky, a leading theorist in the science of AI, nearly fifteen years ago.

Advances since then haven’t always made understanding the sector any easier, as far as WiC is concerned. Fortunately, Helen Fang – Head of Industrials Research, Asia Pacific at HSBC – has done an excellent job of breaking down what’s going on in China’s AI industry in a piece of research titled Artificial Intelligence: Mapping out the value chain published earlier this year.

Fang’s analysis splits the industry into five layers. The foundation block is the AI computing chips that power the algorithms. These chips are housed in a second layer of infrastructure, made up of cloud computing networks and data centres. Next is the machine learning frameworks that train the data in the AI algorithms, which underpin the software models that create AI-enabled solutions for customers. Together, they support the applications of AI-powered technology that are sold into the marketplace, such as traffic management systems in smart cities or the anti-collision capabilities in self-driving vehicles.

WiC spoke to Fang for background on the headline items in her research piece, as well as what they signal for the prospects for China’s best AI firms.

In chipmaking, at least, Chinese companies are still gearing up

It’s commonly reported that Chinese firms are behind their international peers in chipmaking for smartphones and more sophisticated electronics. The situation is similar for the design of the different kinds of high-performance chips that support applications of artificial intelligence.

“AI companies need to train different data sets into something that works as an application” Fang explains. “First they need to collect the graphical information and transform it into digital data. That relies on GPUs, or graphics processing units, the production of which is mainly dominated by American companies like Nvidia.”

After transforming the graphs into digital information, the AI providers need to train the data so that it recognises patterns. The training chips that power this process are another area where Chinese firms are yet to establish a strong position. Once the data has been trained, it starts to make inferences from the rules within the predictive models. And in these kinds of inference chips the Chinese are more advanced, Fang says.

Another Chinese company is also developing its own chips for its AI software models by tailoring them to its intelligent video analysis, which speeds up the inference process.

“In general terms what we are seeing is that Chinese companies are progressing in inference chips but they are behind in the other kinds of chips, perhaps by as much as a few generations,” Fang acknowledges.

“Investors often ask how long it will take for them to catch up but it’s not an easy question to answer. The ‘generations’ are best understood as the sizes of the modules in the chips – in semiconductors, it’s a case of the smaller the chip, the better the performance.”

Companies are active across different parts of the value chain

Businesses are positioning themselves in different areas of the industry.

Some companies are concentrating on chipmaking with a focus on ASIC (‘application specific’) formats that are expected to replace GPUs as the standard over the longer term.

Others are choosing to concentrate more on the processes that support machine learning, while model producers are working on the software deployed to drive the applications for end users.

Companies also try to differentiate themselves by targeting specific sectors as customers. Some are focusing more on applications for supply chains and the retail sector, for example, while others are targeting clients at the Smart City level.

AI firms need to scale before they can profit

Companies need to produce AI applicable models on a massive scale before they can get beyond breakeven, Fang says. That’s not happening yet.

One of the main challenges in getting to profitability is that the costs for training the data are significant, with companies spending 70-90% of their revenues on research and development, primarily in training their datasets.

“Each new task requires a different set of data and all of it needs to be labelled, processed and trained,” she explains. “Chinese engineers are already getting expensive, so much of the work is being distributed to India. But there are still all the other costs to consider, like buying the GPU chips and powering them through specialised data centres. It’s very costly.”

Another roadblock to profitability is that every time an AI firm wants to introduce a new task to a learning process, it goes through the same data loop all over again. “In our analysis we emphasise how companies need to build base models that take on a role similar to that of libraries,” Fang explains. “If companies can come back to the core modules and remodel them, it is quicker and cheaper to get new applications into the market.”

Another challenge: fixing the revenue problem

China’s AI firms will also have to counter shortcomings in their business models, including an overreliance on revenues from clients in the public sector. Similar to the market in China, the leading AI providers in the US make many of their sales to government customers, Fang says. But they benefit from much quicker cash conversion cycles in payment terms, sometimes of no more than a day. In China it typically takes longer.

Another difference is that companies in the United States are often paid through an IP licencing process in which customers shoulder the initial installation fees but then make periodic payments over longer contract periods.

That’s not normally the case in China, where the business is more likely to be billed on the basis of project fees. “As an example, a local government wanting to improve its flood prevention systems would normally expect to pay a single, project fee for the work rather than make ongoing payments in an ‘AI-as-a-service’ model,” says Fang.

Another feature of the sales landscape in China is that government customers rely on third parties to manage their contracts with AI firms, rather than dealing directly. That also has an impact on market dynamics. “If you look at the leading AI firms in China, more than 30% of their revenues come from their top five customers, who are third-party integrators, or middlemen, working on behalf of government clients,” Fang reckons. “The AI firms often lack bargaining power on pricing, which is another hindrance to getting to profitability.”

The four main areas where AI is being deployed in China

Fang identifies four main verticals for applications of artificial intelligence. The biggest is Smart City – where most of the customers are municipal governments – covering sales in areas including traffic management, environmental protection and emergency response.

Next is Smart Business, where AI is being deployed by companies to improve performance in areas including transportation and management of residential property.

Another focal point in the Smart Business sector is the manufacturing industry, where AI companies see new opportunities to win business in areas like machine vision and robotics.

Fang isn’t quite as confident about their prospects, however, pointing out that the AI firms need to understand a lot more about the production processes of their customers first.

“They need more knowhow in order to train the data but the manufacturers are generally quite reluctant to share their expertise,” she explains. “These companies don’t want to reveal their secrets to outsiders, seeing it as their competitive advantage. Some even worry the AI firms could leverage their insights to build software models they then sell to rival manufacturers.”

Applications of artificial technology on consumer devices like smartphones and tablets make up a smaller source of revenue in another area often defined as Smart Life. This segment is “more commercial” as a market than Smart City sales, Fang says, and the sales opportunity is growing because of the proliferation of IoT devices, as well as new demand for AI-enhanced experiences from consumers.

Finally, AI firms have also been investing heavily in applications for Smart Auto, where they are excited about the prospects of emerging trends such as autonomous driving technologies. As Fang points out in her report, as many as 90% of the vehicles sold in China by 2030 are expected to be equipped with some kind of autonomous driving capability, according to market research specialist Frost & Sullivan.

Superpower tensions could lead to a segregated market…

Geopolitical tensions between the US and China will continue to be a factor in the future, it seems. Superpower rivalry sets the scene for many of the bigger-picture assessments of the sector, including the view that companies from the United States are ahead in the raw science of artificial intelligence but that the Chinese have the edge in sourcing the data that finetunes the algorithms behind the applications of the technology.

Fang agrees that companies in China are going to have some advantages in data, particularly for deployments of artificial intelligence in crucial industries such as self-driving cars. “Mileage matters because the more data you have, the more precise the AI models can become,” she explains. It is also an area where Chinese firms seem likely to benefit from a supportive stance from policymakers. “If the government allows AI firms to have the data, they will benefit from it,” she says.

However, she looks at the situation from a different angle in highlighting potential bottlenecks, not least in how GPU chips are still a fundamental part of the industry value chain and that this is an area where American firms are the dominant players.

That means that the prospects for some of China’s AI firms could turn unpredictable if sentiment in the sector starts to be shaped more by politics than profits.

© ChinTell Ltd. All rights reserved.

Sponsored by HSBC.

The Week in China website and the weekly magazine publications are owned and maintained by ChinTell Limited, Hong Kong. Neither HSBC nor any member of the HSBC group of companies ("HSBC") endorses the contents and/or is involved in selecting, creating or editing the contents of the Week in China website or the Week in China magazine. The views expressed in these publications are solely the views of ChinTell Limited and do not necessarily reflect the views or investment ideas of HSBC. No responsibility will therefore be assumed by HSBC for the contents of these publications or for the errors or omissions therein.