It has been nearly four years since the launch of Lee Kai-fu’s widely read book on superpower rivalry in artificial intelligence. As a theme, it has rarely been out of the headlines since then.
A former executive at Google, Microsoft and Apple, Lee had returned to China to invest in the sector. His prediction was that the Chinese were going to be the pacesetters in deploying AI, even if the Americans were ahead in the raw science. But it’s another celebrated returnee – Yao Chi-Chih, a Harvard-trained scientist who came back to China in 2004 after nearly 40 years in the United States – that has been getting mention this month.
Yao is another of the most influential figures in artificial intelligence, Bloomberg reckons. Students from his much-revered classes at Tsinghua University have already created start-ups worth more than $12 billion (at their loftiest valuations), including the founders of the facial-recognition giant Megvii and Pony.ai, a leader in autonomous driving technology.
China’s advantages in AI are increasingly well understood, especially the world’s largest pool of internet users and the unprecedented amounts of data they create. But Yao has also talked about the sector in terms of national mission.
“China missed the microelectronics revolution 70 or 80 years ago, so today it is difficult to catch up with the advanced level of the international semiconductor industry,” he told China Global Television Network, a state TV channel, last year. “But in emerging fields such as quantum technology and artificial intelligence, China is expected to become an important player.”
Of course, these are major priorities for policymakers as well: Beijing has already laid out a strategy for China to be the top AI power in the world by 2030. But if the government is backing the sector and the conditions exist for domestic success, why aren’t more of China’s AI firms making any money?
That was one of the questions addressed by Helen Fang, HSBC’s head of industrials research, Asia-Pacific at HSBC, in conversation with WiC earlier this year (see WiC582).
Fang outlined a few of the reasons why profits have been elusive. Perhaps the key factor is the cost of R&D, where companies need to build software models with sufficient scale for commercial deployment. Getting to that point requires major investment, with some firms spending 70-90% of their revenues 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 explained. “Chinese engineers are already getting expensive, so much of the work is being distributed to India.”
Revenues have also been slower to materialise in China than rival markets. In part that’s because most of the sales are to government buyers, who take longer to pay their bills. Clients are also reluctant to commit to licensing deals that bring consistent payments over longer contracts. “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,” Fang explained.
Challenges like these are raising concerns among investors as a stream of AI companies try to convince them to buy their shares at IPO, despite an absence of profits.
There was another example in the media this month – AISpeech, a specialist in voice recognition, founded by two students that met during overseas study at Cambridge University. Launched in 2007 and supported by significant investment from backers including Alibaba, Foxconn, Midea Capital and Lenovo, AISpeech has just filed for a STAR Market IPO to raise Rmb1 billion. However, AISpeech reported losses of more than Rmb800 million over the last three years, ThePaper.cn notes, with spending on R&D in the period coming close to the entirety of its operating income.
The prospectus for the IPO also notes the possibility of further increases in R&D in the future, even at a time when fierce competition in its market segment may mean falling prices for its services, the company acknowledges.
The situation isn’t hugely different for some of the best-known firms in the AI sector. Sometimes referred to as the ‘little four dragons’, each of the group has run into problems reaching profitability in recent years.
CloudWalk, a developer of facial recognition software, finally arrived on the STAR Market in Shanghai in May this year, although investors couldn’t be persuaded to buy its shares anywhere near the top end of the range.
One of the reasons: net losses of over Rmb2.3 billion between 2019 and 2021, with R&D spend accounting for more than three-quarters of its costs.
Its capital raising was subsequently downsized but still merited mention in the local media as a so-called ‘blood listing’ in which the IPO price drops below the valuations of late-stage financings as a unicorn private company.
Megvii, another of the AI giants, is still to make a profit as well. With a focus on the Internet of Things (IoT) sector, it first applied for an IPO in Hong Kong in 2019 before switching its interest to the STAR Market in Shanghai. That debut was reported to have received formal approvals last year, although Megvii dragged its feet, probably in a bid to get its finances into better shape. There were reports in May that it had resumed its registration process.
Yitu, another of the computer vision heavyweights, had ambitions of its own to list on the STAR Market. But it dropped its IPO application last year, reportedly because of its financial results.
SenseTime, the biggest of the ‘four little dragons’, did make it to market in Hong Kong at the start of this year, despite the unwelcome intervention of the US Treasury Department, which added it to a sanctioned group of “Chinese military-industrial complex companies” while it was conducting its roadshow. SenseTime shrugged off the news and its stock price opened strongly in January. But its shares have traded steadily lower (admittedly in awful market conditions) before plummeting disastrously at the start of July when a lock-up period on its IPO expired. Investors raced for the exits, with its shares falling by almost half.
The shares recovered a little of that lost ground last week when SenseTime bought back some of its own stock. But they dropped another 8% on Tuesday after a stake matching the size of Softbank’s position in the company showed up in the stock exchange’s clearing system, triggering fears that the Japanese giant could be selling as well.
One of the criticisms of SenseTime is that it has been slow to move beyond its initial focus on image recognition into applications in machine learning, robotics and natural language processing. But others blame weaker sentiment in the wider sector for its declining value, as shareholders lose some of their earlier enthusiasm for the AI story.
“In previous years investors were optimistic about the prospects of the AI industry and eager to pump funds into lossmaking firms, which greatly pushed up their valuations. But the difficulties in implementing AI technology, the uncertain application scenarios and their inability to make money have led to huge losses that are getting worse and worse,” one investor told CBN, a local news site.
What all of this says for AISpeech and its own IPO is open to question. Yet if the leaders within Chinese AI can’t make money, the situation for companies in the next tier down is probably going to be challenging, unless they can find a niche in the market and dominate it.
One of the more positive prospects for the best of these firms is favourable treatment by policymakers, who don’t want their local champions to trail their peers in Western markets. In areas like facial recognition, that seems to be paying dividends, at least in how the technology is widely adopted. But whether the support from the state has served the AI champions well is more debatable if financial performance is the benchmark.
Equally, it’s not just a question of outpacing their international rivals. The leading AI firms are also going to have to fight off domestic challengers as they hunt for long-awaited profits.
For instance, for specialists in computer vision software there is always the fear that hardware makers like Hikvision and Dahua (firms that make the surveillance camera systems for ‘smart city’ deployments) could acquire more of the expertise in machine learning and neural networks themselves. And perhaps the threat to the ‘little dragons’ and their smaller peers is even greater as tech giants like Baidu and Huawei turn more of their focus to AI as well, deploying solutions of their own design. As a single example: Megvii offers a machine-learning framework called Brain ++ that helps businesses build their own algorithms and AI capabilities. But Huawei and Alibaba are both developing offerings with neural networks of their own.
© 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.