Behind the Rise of China’s Facial-Recognition Giants

Megvii, one of China's highly valued facial-recognition startups, has filed for an IPO. Most of its revenue comes from a unit selling surveillance and security systems.
screen showing a facial recognition system with people walking.
Gilles Sabrie/The New York Tim​es/Redux

Unfamiliar faces aren’t welcome at Beijing public housing projects. To prevent illegal subletting, many have facial-recognition systems that allow entry only to residents and certain delivery staff, according to state news agency Xinhua. Each of the city’s 59 public housing sites is due to have the technology by year’s end.

Artificial intelligence startup Megvii mentioned a similar public housing security contract in an unspecified Chinese city in filing for an initial public offering in Hong Kong last week. The Chinese company, best-known for facial recognition, touts its government dealings, including locking down public housing to curb subletting, as a selling point to potential investors.

Megvii’s filing shows the scale of China’s ambitions in artificial intelligence and how they could influence the use of surveillance technologies like facial recognition around the world. The company is one of four Chinese AI startups specializing in facial recognition valued at more than $1 billion, qualifying them as unicorns in Silicon Valley-speak. Now, the companies are looking to expand overseas, with help from public markets.

“These companies have benefited from China’s government making it a national priority to be the world leader in AI,” says Rebecca Fannin, author of the forthcoming Tech Titans of China and two previous books about China’s tech scene. That support has led to contracts and freed up government and private funds, she says. “Now you are starting to see these companies go global.”

Freedom House, a US-government-backed nonprofit, warned in a report last October that Chinese surveillance deals also export the country’s attitudes to privacy and could encourage companies and governments to collect and expose sensitive data. It argues that companies and products built to serve government agencies unconcerned about privacy are unlikely to become trustworthy defenders of human rights elsewhere, and can be forced to serve Chinese government interests.

Megvii’s filing says it has raised more than $1.3 billion, primarily from Chinese investment funds and companies, including ecommerce giant Alibaba. One of China’s state-owned VC funds also has a stake and a seat on the startup’s board. Other backers include US-based venture firm GGV and the sovereign wealth funds of Abu Dhabi and Kuwait. CB Insights said the company was valued at $4 billion earlier this year. Reuters reports that its public listing will raise at least $500 million, but the figure is redacted from the company’s filing. Fannin says larger rival SenseTime, valued at $4.5 billion per CB Insights and also expanding overseas, is expected to go public soon. Its international ambitions could be helped by having US investors in Qualcomm and Silver Lake Capital.

Demand for Megvii’s computer-vision systems is growing rapidly. The company reported revenue of 1.4 billion yuan ($200 million) in 2018, more than four times a year earlier. It posted losses of 3.4 billion yuan ($469 million). The “City IoT” segment of its business that provides surveillance and security systems, like access control for public housing, accounts for nearly three-fourths of its revenue and has customers in more than 15 “countries and territories” outside China. That division also offers software that can spot traffic offenses or changing traffic flows caught on video.

In the first six months of this year, Megvii says 4.9 percent of its revenue came from outside China, compared with 2.7 percent for all of last year. Now, it plans to establish joint ventures or offices in Japan, Singapore, Thailand, and the Middle East.

Megvii was founded in 2011 by Yin Qi and two friends from Beijing’s elite Tsinghua University. The company’s name is short for "mega vision"—also the rough translation of its Chinese name, 旷视. Their venture was perfectly timed to surf the swell of interest in AI prompted by the emergence of a technology called deep learning in 2012, which made software that interprets images much more accurate.

Since then, Megvii and rival AI unicorns SenseTime, CloudWalk, and Yitu have made facial recognition commonplace in China, where police scan public spaces for suspects, and citizens pay in stores and pay taxes with their faces. Recently, Chinese startups, along with some from Russia, have dominated the US National Institute of Standards and Technology’s rankings of facial-recognition accuracy.

One of Megvii’s early successes came in 2015, when Alibaba affiliate Ant Financial used its technology to launch a feature called Smile to Pay. Megvii now provides facial recognition to government surveillance projects in China, and face authentication to banks and smartphone manufacturers, including Oppo, which Counterpoint Research says is the world’s fifth-largest phone maker by shipments. Megvii’s researchers regularly win academic contests for computer vision algorithms and have defeated competitors from Google and Microsoft.

Facial recognition is the technology Megvii is known for and has offered longest, and AI-powered security and surveillance appears to be the company’s main business. The startup also has a warehouse robotics group, and it touts beautifying algorithms that can remove pimples and reshape your body in selfies.

Police in China use facial recognition to pluck persons of interest from concert crowds, and have even used wearable Google Glass-style devices that allow a cop to scan the face of anyone they’re looking at. The technology is part of the intense security apparatus assembled to watch over China’s northwest Xinjiang region, where an estimated 1 million Uighur muslims have been swept into internment camps. The New York Times reported in April that Megvii, SenseTime, and CloudWalk had helped create surveillance software that looks for Uighur faces. Megvii’s PR company said the company doesn’t design or customize its products to target ethnic groups.

Megvii’s filing mentions police use of its technology but not Xinjiang or algorithms that try to detect ethnicity. It presents case studies that make its surveillance offerings sound more cuddly.

One describes a 2018 incident in which police in northern China used Megvii’s technology to identify an elderly man who had forgotten his name and address and then escorted him home. The filing also mentions technology Megvii patented to identify dogs from their nose prints, used by Beijing authorities to manage strays.

US authorities are unlikely to buy from Megvii, despite the startup's having a research lab in the Seattle suburb that’s also home to Microsoft. US agencies are traditionally wary of security technology from places that aren’t close allies of America. Megvii’s filing warns that it could be harmed by the Trump administration’s tariffs on China or knock-on effects from its suspicion of telecoms company Huawei.

Fannin says the company will have an easier time in other parts of Asia, South America, the Middle East, and Africa. All have become important markets for Chinese technology companies, including Hikvision, the world’s largest vendor of security cameras.

Megvii surveillance technology is already available in Thailand, and its competitors are also beginning to find success outside China. Yitu provides facial recognition to police in Malaysia, and CloudWalk won a contract to build a national facial-recognition system for Zimbabwe through China’s Belt and Road program of international infrastructure deals.

Megvii’s filing devotes significant space to assuring would-be investors that it can be trusted to use AI appropriately. It includes the company’s AI ethics code of conduct and says that customer contracts forbid using its technology to infringe human rights. Megvii also lists an AI ethics committee that reports to its board, including company executives and directors, and some outsiders.

One member listed, Emmanuel Lagarrigue, chief innovation officer at France-based Schneider Electric, describes the committee as a “work in progress.” He says its membership has not been finalized but that Megvii deserves credit for creating the group. “It shows the company's willingness to be proactive and transparent on how it wants to operate and on how its technology is deployed,” Lagarrigue says.

Jeffrey Ding, who studies Chinese AI development at Oxford’s Future of Humanity Institute, says that on paper Megvii’s ethics structure appears stronger than that of some large US companies. Google and Microsoft have released statements of AI principles, but they are implemented only via internal review processes.

However, Chinese companies aren’t totally in control of their own destinies. “The economy is a relatively free market, but the party and state can potentially completely control any company at any time,” Ding says. “Ethical obligations from Chinese companies come with a little bit less weight.”

Updated, 9-4-19, 1:40pm ET: This article has been updated to reflect that Ant Financial used Megvii's technology to launch Smile to Pay.


WIRED Theme Week: How We Learn