AI Superpower China? — A Business Perspective

Without a doubt, the book “AI Superpowers: China, Silicon Valley, and the New World Order”, published by Dr. Kai-Fu Lee two years ago, represents one of the most read and commented books about China tech in the Western world. Simultaneously, there was a visible euphoria about AI in China, and in some cases, fear what this emerging industry would mean for the global business and political landscape.

Two years later, however, we see the AI euphoria fading. Not in China, but globally, there seems to be a realization that bold dreams such as fully autonomous driving, medical diagnosis, and humanoid robots seem to be further away than most observers expected a few years ago.

The Number of Companies and Funding Is Down After a Peak in 2017/2018

When Dr. Kai-Fu Lee wrote his book, China´s AI industry saw enormous growth, both in the numbers of companies established and funding. In the years before the deep learning revolution in the early 2010s, i.e., algorithms inspired by the structure and function of the human brain, AI startups presented a niche with only 10 to 30 new companies being established in China every year. The funding for AI startups was minimal compared to the investments in AI conducted by Chinese internet giants such as Tencent, Alibaba, Baidu, and JD. Even before the deep learning revolution, these companies were active adopters of AI, to provide tailored user experiences and maximize the monetization of their vast data.

The situation changed abruptly in the early 2010s, with the number of new AI startups skyrocketing to 63 in 2012, reaching a peak of 228 new startups in 2016. Many of the largest AI startups that can be seen in China today, including Megvii (2011), UBTech (2012), Sensetime (2014), Horizon Robotics (2015), and Cloudwalk (2015) have been founded in the first couple of years of the AI boom, benefitting from an enormously favorably investment environment in the years between 2014 and 2018. In that period, the industry has seen the emergence of more than 30 AI unicorn companies, with spectacular funding rounds: In October 2017, China’s second-largest AI company Megvii raised 460m USD at a valuation of 3.5bn USD. In May 2018, China’s largest AI company SenseTime raised 620 million USD at a valuation of 4.5bn USD. In February 2019, China’s largest AI Chip company Horizon Robotics raised 600m USD at a valuation of 3bn USD.

After the boom, reaching its peak in H1, 2018, even overtaking the US as the number one destination for venture capital globally, the euphoria seems to have decreased. In 2019 the number of new AI startups amounted to 53, and declined even further in 2019, to only 32.

The Commercial Prospects of AI Companies

When Dr. Kai-Fu Lee refers to AI companies, he refers to four types of companies: Internet AI, Business AI, Perception AI, and Autonomous AI.

Internet AI

Looking at Internet AI, Chinese companies have been rapid adaptors of AI.

In social media, Bytedance and Kuaishou have created a new generation of internet platforms automating the experience using advanced AI algorithms and moderation instead of just providing content discovery. The results speak for itself: Whereas traditional social media companies seemed to have reached its peak, Bytedance counts far more than 1bn active users globally and has reached a valuation of over 100bn USD in 2020, just eight years after being founded. In early 2021, the valuation is expected to be north of 200bn USD.

Above that, Chinese internet companies are also investing in AI technologies outside of their core business. For example, Baidu invested heavily in autonomous driving since 2013, establishing the so-called Apollo platform, one of the most adopted open development platforms for autonomous driving in the world. Another example is the e-commerce giant JD.com, which established JDx in 2016, a business unit with over 700 employees focused on developing autonomous technology in warehousing, retail, and delivery domains.

Business AI

In Business AI, China has produced several unicorns, including 4Paradigm and Yitu Technologies, with revenues of up to 400m USD each in 2020, which is a significant number when comparing to established AI companies such as Business AI company, Palantir Technologies, founded in 2003, and valued 15.8bn, only generated 739m USD in 2019. 4Paradigm became a company investors are particularly bullish about, closing a 700m USD round in early 2021, with expected IPO at the end of this year.

However, challenges remain: First, until today, the development of China’s B2B software sector is relatively limited compared to its consumer sector, indicating that there is a long way to go compared to consumer segments. Second, whereas American and European companies can expand abroad, data protection regulations, and limited trust in Chinese AI companies, might undermine their ability to expand to other profitable markets.

Perception AI

Perception AI represents the largest segment of AI startups in China, with most of the highest-valued startups falling in that category. That doesn’t surprise, as the deep learning revolution primarily fueled progress in developing perception algorithms. Within perception AI companies, there are three categories of companies that can be distinguished:

AI Software Giants

AI software giants including SenseTime and Face++, with each of them employing 2,000+ employees, and engaging in the development of AI algorithms for a wide range of use cases, be it smart city, healthcare, automotive, security, and education. With expected revenues of more than 600m USD each, they have created the most extensive commercial output to date, benefitting enormously from the demand for identification and smart city solutions in China. However, it remains to be seen if their valuations, that are a multiple of their annual revenues, 7, 8 years after foundation are justified, and whether there will be more uses cases companies can successfully tap into.

Specialized AI Software

There is a tier of smaller AI unicorns and companies that cannot chase every vertical and technology because of their more limited financing. One example is Mobvoi, founded in 2012, focused on voice recognition & natural language processing, developing consumer and business solutions, e.g., through its joint venture with Volkswagen established in April 2017. Another example is Unisound, founded in 2012, that enables other companies to develop speech recognition solutions.

However, the estimated revenues of many of the smaller companies seem to be disappointing. Even seven/eight years after establishment, many unicorns can be expected not even to have reached 100m USD in revenues. Xu Jinghong, Chairman of Tsinghua Holdings, the investment arm of Tsinghua University in Beijing, states that half of today’s unicorns might die if they cannot capture economic value from their developments.

The challenges for smaller companies seem to be both on the supply and demand side: On the one hand, the boom made the hiring of AI talent costly, in many cases, salaries are even higher than in industrial countries like Germany, UK and France, while the best talent would like to join either internet companies or larger AI companies. Simultaneously, there are large additional costs, including in the annotation of data, the usage of computing power, and the procurement of other technological resources. On the other hand, clear use cases for perception AI other than smart city or safety scenarios seem to be lacking. Companies possessing the data themselves, especially big internet companies, might prefer developing the technology in-house rather than relying on specialized AI companies.

AI Chip Companies

The AI boom has also created new companies in hardware, developing semiconductors to run inference operations in edge devices or cloud, which is an area that is currently dominated by Western companies. Examples include Horizon Robotics, valued at 4bn USD in 2021 , recently raising over 1bn USD recently closing its 1bn USD Series C round with a focus on edge computing for automotive and IoT use cases, and the already IPOed Cambricon, valued with a market cap of over 8bn USD as of April 2021 with a focus on cloud computing.

The commercialization of Chinese AI chip startups seems to be relatively limited compared to the revenues generated by software startups, especially comparing the enormous funding required to operate a chip company; this, however, is not surprising: First, the development cycle for semiconductors takes years, compared to software companies, which can quickly adapt, change, and scale successful products & solutions.

Second, in the first years, even if a semiconductor might offer superior products, customers will likely continue to rely on established companies such as NVIDIA, Intel, Qualcomm, that are proven in the market and provide an advanced ecosystem of support. Nevertheless, the trend towards local procurement and continuing development of these comparably young chip companies might create enormous commercial potential in the medium to long-term (5 to 15 years) explaining its high revenue multiples.

Autonomous AI

Autonomous AI has probably been the most ambitious field for AI commercialization, with autonomous driving has been one of the hottest topics between 2012 and 2018. Several unicorns emerged in that period, including the software company Momenta, the robotaxi companies Pony AI, AutoX, and the autonomous truck company TuSimple.

Yet, the industry is facing challenges: First, to compete at a global level seems challenging, given the sophistication of especially American competitors, such as Waymo, Nuro, and Cruise, that seem significantly ahead compared to their Chinese peers.

Second, there appears to be a significant market for autonomous driving neither in China nor abroad. Even with support from national, state, and local governments, safety concerns, commercial viability, and cultural barriers won’t be overruled in the short- to medium term. Thus, a refocus on lower levels of automation, i.e., automated driving appears to be a more feasible way for near term commercialization than the focus on fully automated vehicles.

Next to autonomous driving, robotics might represent another area that is more likely to be commercialized in the near-term, with UBTech, founded in 2012, being valued between 5 and 10bn USD. UBTech is expected to already today generate revenues of 500m USD, and its robots can be found in hotels, restaurants, and public buildings across China.

Conclusion

Artificial intelligence, and deep learning have already impacted our lives substantially, and Chinese and Western companies, especially in the internet industry, can be seen as the key adopters of that technology. At the same time, Chinese academic institutions have established themselves as a key hub for AI research and production of talent next to the United States as we have already outlined in our article “Chinese Undergrads Will Determine The Outcome of US-China Arms Race in Artificial Intelligence”.

However, in terms of business potential the AI industry in China, especially in the field of perception, and autonomous AI has been relatively disappointing, especially when comparing against the era of internet companies.

1. Limited Company Founding: After massive founding of new AI companies, progress seems to be stalling, and the number of new AI companies has reached small double digit figures. Instead of new players, the industry has produced a few unicorns that are dominating the industry across verticals such as surveillance, automotive, and IoT sectors.
2. Limited Investor Appetite: While 2021 has seen several multiple hundred US pre-IPO rounds, company valuations did not increase significantly over the last 2, 3 years, reflecting that many companies have not lived up to their promises. Use cases for AI that can be well monetized are lacking, while development costs are high, and unique competitive advantages are lacking.
3. Limited Internationalization Potential: Whereas some Chinese technology companies had the luxury to expand abroad, e.g. Huawei, Xiaomi, and Bytedance, this luxury will be much harder to be attained by AI companies whose data has national boundaries, and whose products and services face more scrutiny than of nearly no other industry.

Nevertheless, the industry is still at its beginning and complex technological, entrepreneurial and political factors are interplaying and over time will show whether the AI industry in China will become key force within the global technology industry. From a commercialization perspective it currently seems unlikely that AI companies will itself become as big of an industry as other high-tech industrial sectors, yet its impact on the broader technology ecosystem might not be understated.

1. Top Talent: The top AI talent and cutting edge developments conducted in AI companies, can be utilized in internet, and traditional companies, such as automotive and fuel the development of Chinese companies, which can then leapfrog competitors in the West that might be lagging behind in the development and deployment of cutting edge AI technology.
2. Consumer Habits: The creation of pioneering companies and use cases, has made Chinese citizens more demanding and accepting of AI technologies than citizens of most other countries. Their demands will fuel further developments and put China into a spot where new AI applications, can be first tested and deployed.

Concluding, the term superpower would usually signify that a technological capacity would be used against other countries, or have some kind of destructive potential. Yet, AI companies seem to have more of an enabling nature, rather than the nature of monopolistic competition. Instead of emphasizing a zero sum games, companies in East and West should see how they can benefit and learn from the technologies in China, and leverage China next to the US as a training ground to develop the most cutting edge technology.