Beijing-based Megvii—one of the darlings of China’s artificial intelligence (AI) industry—has long aimed to become the country’s first listed AI company. Yet, the stars have not aligned despite the firm’s solid leadership team and star technologies such as its online facial recognition platform Face++.
In August 2019, it sought approval for a listing on the Hong Kong Stock Exchange (HKEX) and submitted a draft prospectus. However, just two months later, Megvii landed on the US government’s Entity List, which led to concerns over the impact on the firm’s supply chain and planned Hong Kong IPO.
To add to the list of troubles, despite commentaries by HKEX’s CEO Li Xiaojia mentioning that the Hong Kong bourse should not be directly affected by the US Entity list, Megvii has had to face off campaigns lobbying for the HKEX to block its listing, owing to inadequate disclosure of sanctions risks.
Since Megvii’s first HKEX application and after its six-month regulatory deadline expiration in February 2020, the firm has kept stubbornly mum over the status of its fractured listing plans in the face of regulatory conditions and unpredictability regarding its presence on the US Entity List.
Now, Megvii is aggressively pushing the pedal to add more status to its brand via commercialization pathways, while hinting at aspirations of a future listing on the rocking Shanghai’s Stock Exchange Science and Technology Innovation, also known as the Star Market. A new spring may be imminent.
An AI expert
Megvii’s expertise in the AI field is remarkable, boasting over 1,100 patents and coming in as the top performer in at least 40 AI competitions. Megvii’s revenue growth metrics have also been positive, increasing from RMB 67.8 million (USD 9.76 million) in 2016 to RMB 1.427 billion (USD 219.8 million) in 2018, with a compound annual growth rate (CAGR) of 358.8% over this same time period.
Still, these positive numbers may not be enough for investors: Megvii’s valuation relies on its ability and demand for value realization through sustainable profit, which is especially difficult for AI companies due to lower gross margins and difficulties in scaling edge cases. This makes their business models more complex than traditional service-as-a-software (SaaS) companies. Despite solid revenue growth, Megvii has drawn net losses for three consecutive years, at a staggering RMB 5.6 billion (USD 806.1 million) as of the first half of 2019.
Why is turning a profit so difficult for AI companies such as Megvii? An answer might be conspicuous research and development (R&D) costs. Megvii has tried to defend its losses by citing continuous investment in R&D, stating that its net profit, after adjusting to excluding non-recurring gains and losses, was positive to RMB 32.7 million (USD 4.7 million) year-on-year as of the first half of 2019.
Yet, customer uptake and revenue realization may also be slow due to a lack of market understanding on how Megvii’s AI algorithms can be deployed for everyday use. For corporate clients, piloting AI trials may be easy, but deploying them at scale is difficult because non-AI specialist clients lack basic knowledge of identifying where AI opportunities lie.
According to a McKinsey survey, the most frequently cited barrier to AI adoption was a lack of strategy, with 43% of respondents indicating this was the case for their company. As of September 2019, the bulk of Megvii’s revenues—34%—came from five undisclosed customers. What’s more, 2019 was only the third year in which Megvii’s revenue exceeded RMB 100 million (USD 14.3 million), despite the company’s inception in 2011.
Customer adoption is particularly pertinent for Megvii, which sets it apart from other AI companies that stick to one vertical and go deep. Megvii provides a variety of AI solutions that can be adapted for specific problems, operating across three broad verticals of “personal, supply-chain, and smart city” Internet-of-Things (IoT) solutions. Yet, as of the first half of 2019, 73% of Megvii’s revenue came from its single smart cities product segment (Megvii’s solutions have been deployed across 260 cities), which is in part due to its heavy reliance on government entities for clientele, Bloomberg reported.
Achieving a more balanced revenue stream may be critical not just for listing success, but for longer-term business sustainability.
The leap to commercialization
On July 29, 2020, Megvii’s co-founder and CEO, Yin Qi, issued a rare statement addressing the company’s problems in this regard during a press conference, according to 36Kr.
“All AI companies have entered the valley of death. Everyone has high expectations, [but] when expectations fall, it will be very difficult to predict which companies will be able to cross this valley . . . Crossing this valley will happen over the next 18 to 24 months,” he said, referring to the live-or-die funding gap faced by startups, especially companies in the deep tech sector when making the leap to commercialization.
With this situation under the magnifying glass during its previous listing attempt, Megvii is now aggressively pre-empting customer needs and offering ready-made applications for potential use across all of its three verticals.
It frames itself as one of the “few” companies that can offer full-stack, end-to-end solutions rather than just niche algorithms. For instance, its public-facing product center doesn’t boast software products, but rather, “a large network of hardware.” This includes at least ten different kinds of AI cameras, sensors, robots, and other smart hardware devices integrated with IoT capabilities.
Megvii runs in deep contrast to computer vision competitors such as China’s SenseTime, which markets products and services under one umbrella, with a significantly lower emphasis on hardware, or the US’s Athena Security, which presents its products as modular technology solutions.
Simply put, Megvii is looking a lot more like a hardware company than a software solutions company, despite its aggressive insistence on the preeminence of its core algorithms.
Over the past four months, Megvii has sought to throw off its rarefied image and has accelerated the roll-out of a diverse suite of products, ranging from beauty to supply chain solutions, to appeal to different consumer bases. Since February, Megvii’s announcements have pivoted to commercial solutions or products in the field, while previously, it focused on showing off how it had succeeded at various international competitions or attained prestigious rankings.
Perhaps Megvii has learned that transforming its original academic style to more consumer-facing products was necessary for success. For instance, in April, it launched FaceStyle, an AI-powered makeup and beauty solution platform that allows end-users to personalize their shopping experience by virtually trying on makeup through computers or mobile devices.
The solution was implemented by e-commerce platform Haitunjia, to help marketers keep track of product popularity, as well as to help coquettish shoppers on Haitunjia to recreate the makeup experience down to every wrinkle and spot.
To add more blush to its strategy, Megvii made its first global push to supply AI solutions for storage provider Store Friendly in Singapore, which allows customers to drop off and pick up personal belongings without the need to communicate with human clerks at the company’s mini-warehouses.
“We have intended to leverage our highly scalable business model to expand globally,” said a Megvii spokesperson to KrASIA. “We hope to provide more tailored solutions for international customers . . . This unmanned warehouse has been in operation for nearly one year, which gives us confidence that our technology could solve real-life problems and pain points encountered by partners and customers across the globe.”
These new global products are not novel or deep tech solutions that Megvii is capable of producing, given its reputation as one of the smartest companies in the world. Yet the sudden global push overlaps with a pivot to more accessible products and revenue streams, in a simultaneous bid to throw off its enigmatic image.
Another major change made by Megvii, in the wake of its tumultuous year, involves the decision in March to open-source its deep learning framework MegEngine, one of the three components of Megvii’s hallmark platform Brain++, alongside data management platform MegData and cloud computing platform MegCompute. The platform is scheduled to launch officially in September this year.
Deep learning frameworks are interfaces, libraries, or tools that enable deep learning models to be built more efficiently, by enabling developers to utilize pre-built techniques when training their own AI systems. Open-sourcing MegEngine could enable Megvii to capitalize on crowdsourcing to accelerate the development of new use cases. More importantly, it could help Megvii to develop a more complete ecosystem around Brain++, challenging the world’s two top deep learning frameworks, Google’s TensorFlow and Facebook’s Pytorch.
Casting its eyes toward another bourse
Aside from its attempt to push the pedal on commercialization, Megvii is actively looking at the more “relaxed” Star Market for a potential listing. First announced by Chinese President Xi Jinping in November 2018 at the China International Import Expo, the board is part of a series of capital reforms aimed to enhance market inclusivity and spur growth, targeting innovative high-tech companies such as Megvii.
Like the HKEX, the Star Marked can accommodate dual-class structure companies, which holds huge appeal to founders who wish to retain control over the company while dispersing economic rights more widely.
“The Shanghai Science and Technology board supports and encourages ‘hard technology companies’ to go public, which is a good development opportunity for Chinese technology companies. Megvii is actively considering this,” said Yin during the press conference, despite insisting that plans to go public in Hong Kong had not been fully shuttered yet.
Recently, tech media platform 36Kr reported that Megvii had been in talks with CITIC Securities for sponsorship, while it also analyzes the option of using multiple underwriters for a Star Market listing, according to industry sources.
When approached by KrASIA, Megvii’s spokesperson stated that the company will not comment on market speculation. “Megvii is weighing up a range of options in terms of listing location and no firm plans have been made for now.”
Yet, other tech companies seem to race faster than Megvii when it comes to going public. AI chip maker Cambricon (688256.SH), also backed by Alibaba (NYSE: BABA, joined the Star Market in a record time of 68 days, while there are speculations of a Star Market IPO by competitor SenseTime.
Despite fears of overheating and untested valuations for loss-making companies, stocks have performed well on the Star Market so far. QuantumCTek’s (688027:SH) shares closed at 924% above the IPO price on its opening day, while Cambricon and Kingsoft Office’s (688111.SH) shares shot up 288% and 205% respectively.
However, Megvii has tried to play down the perceived desperation for an IPO. “Listing is a means rather than an end,” Yin said in July, adding that the firm maintains a healthy cash flow.