August 31, 2021
Image: Pavlo Gonchar/SOPA Images/LightRocket via Getty Images
SenseTime, a company that represents China's ambitions to pull ahead of the United States in the global artificial intelligence race, filed to go public in Hong Kong last Friday. Valued at $12 billion according to the Financial Times, SenseTime wants to prove to investors that it can turn cutting-edge AI research capacity into real cash.
Founded by the MIT-trained engineering professor Tang Xiao'ou in 2014, SenseTime is often regarded as the most valuable of China's "four AI dragons." It primarily sells its computer vision software to companies and governments to help them improve efficiency and offer services that weren't possible without AI.
SenseTime is a prime example of a research-oriented company. According to its prospectus, over two-thirds of its employees are trained scientists or engineers. It has won over 70 top positions in global AI contests and authored more research papers presented in international computer vision conferences from 2015 to 2021 than any competitor. It also has "the world's largest computer vision model with over 30 billion parameters."
But the company's success also comes from being able to find commercial applications for its research. SenseTime's prospectus reads like a compilation of all the buzzwords in today's technology world: Its AI algorithms are currently being used in products from enterprise software to smart city solutions to self-driving cars to consumer tech apps. And of course, don't forget the metaverse. To accelerate AI's development for so many different uses, the company says it has built a "centralized mass production" infrastructure that makes it possible to "develop AI models in hours rather than in weeks."
According to a Frost & Sullivan report commissioned by SenseTime, SenseTime is currently the largest AI software provider in Asia by revenue. Its annual revenue growth has been steady, but not exactly impressive, with a 13.8% uptick from $468 million in 2019 to $532 million in 2020. The company says COVID-19 is the main reason for the growth stagnation. (From 2018 to 2019, its annual revenue grew 63%.) Most of this revenue came from SenseTime's Smart Business (39.2% in 2021) and Smart City (47.6% in 2021) services.
Despite growing revenue, the company has always run a deficit. In 2019 and 2020, the company's annual net loss amounts to $768 million and $1.8 billion, respectively. Even after EBITDA adjustments, the company was still losing over one hundred million dollars each of the last two years. The operational costs mostly come from R&D and administrative expenses.
Also noteworthy: The company still receives a significant amount of Chinese government subsidies. In the past three years, SenseTime has received $31 million, $38 million and $54 million in subsidies respectively.
SenseTime will continue to invest heavily in research. The company pledges to invest 60% of the amount raised in this IPO to enhance R&D capabilities, with 35% going to algorithm mass production infrastructure and 25% to other technologies.
Another important task for SenseTime is figuring out more ways to convert its research progress into real-world applications. That means expanding into more industry verticals and finding more enterprise-level application scenarios, both domestically and internationally.
SenseTime has decided to go public at an interesting time when so many things, both in and out of the company's control, could go wrong.
First of all, as a company that relies on tons of data to train its algorithms, it is in a precarious spot as Beijing ramps up its control over business data. DiDi's IPO in the U.S. has already seen share prices fall well below their initial level due to data security concerns, and China has released a slew of data regulations in the past month whose effects are yet to be felt. SenseTime will need to be very careful about its data storage and transfer practices to stay in the safe zone.
There's no less hostility on the other side of the Pacific Ocean. In Oct 2019, the U.S. Commerce Department added SenseTime to a trade blacklist over national security concerns and foreign policy interests, blocking the company from purchasing components and technologies from U.S. companies. But SenseTime argues that only Beijing SenseTime, a corporate subsidiary, was blacklisted, while the other subsidiaries like Shanghai SenseTime and Shenzhen SenseTime should remain unimpacted. It also says it will fight for Beijing SenseTime to be removed from the list "as soon as an appropriate opportunity arises."
One of SenseTime's core businesses is Smart City, which involves using AI algorithms to help municipal governments better manage their residents. While that includes innocuous technologies like traffic flow optimization or fire detection, some Smart City technologies, like facial recognition cameras, are also accused of being used by China's government to surveil minorities. As a service provider, SenseTime is no stranger to global criticism that it empowers state surveillance. It pulled out of a joint venture in Xinjiang in 2019 for this very reason.
Last but not least, AI is a crowded sector in China, where SenseTime not only faces competition from the other AI dragons, but also Big Techs like Baidu and Huawei.
"As 'smart security' integrates image recognition technology into [the] security system, SenseTime and all pure AI solution providers are too 'soft' in the field — they need to apply their image recognition products and services to hardware to provide a comprehensive solution." — Fu Yingwen, analyst at investment research firm EqualOcean
On why SenseTime doesn't have an economic moat: "In the field of computer vision, [the competition] comes down to algorithms and processing power. In terms of algorithms, many of them are open source, so the top 20 companies aren't quite different from each other when it comes to applying their technologies to daily scenarios; in terms of processing power, it depends on the microchips, where SenseTime also doesn't have an advantage." — AI industry observer Yu Ming, in Chinese publication Lanjinger