As the industry shifts from 'brute force computing' to 'scenario-based cultivation', Megvii is undergoing profound changes in its transformation from security to intelligent driving in the era of large models.

Today, when mentioning the 'Six Little Dragons of AI', names like Zhipu, Dark Side of the Moon, Step Star, and Baichuan Intelligence are familiar. However, it has been less than three years since the term 'Four Little Dragons of AI' was coined.

Looking back from the midst of the generative AI wave, the once prominent 'Four Little Dragons' - Megvii, Yitu, SenseTime, and CloudWalk - have been excluded from the spotlight. This is not a normal situation if we consider the 10-year industry cycle of the AI sector.

As a technology-led industry, the previous Four Little Dragons of AI could not be underestimated in terms of talent density and technical prowess. So, what exactly caused them to gradually become 'low-key' in the new AI cycle? This is an interesting question.

Megvii's multiple failed attempts at IPO are well-known. However, in the long-term environment of AI industry driven by technology and investment, its success or failure is related to multiple factors.

For shareholders and strategic investors, this applies to both Alibaba in Megvii's early stages and Geely in its mid-to-late stages of transformation towards intelligent driving. For the team, after experiencing challenges and transformations, can this company, whose management team is rooted in technological entrepreneurship from Tsinghua University, become more mature and stable in the second wave of AI development?

01 Megvii's Bumpy Path to IPO

Megvii emerged early in the AI industry.

As early as 2015, a decade ago, Megvii developed China's first dynamic face recognition system, which was promoted in the summer of 2016 and quickly gained capital favor due to its product scarcity, completing a C-round financing of US$47 million shortly after.

A set of numerical comparisons is sufficient to summarize Megvii's expansion speed during this stage. A source close to Megvii told us that around 2016, Megvii's team size was around 300 people, but after the funding came in, the team expanded to over 1,000 people within one to two years.

Compared to SenseTime and Yitu, Megvii had a relatively higher talent density, which was also related to Sun Jian, the former dean of Megvii Research Institute.

Public information shows that Sun Jian was born in Xi'an, Shaanxi, in 1977. He graduated from the Institute of Artificial Intelligence and Robotics at Xi'an Jiaotong University in 2003 with bachelor's, master's, and doctoral degrees. After graduation, he joined Microsoft Research Asia until 2016, where his main research directions were computational photography, face recognition, and image understanding based on deep learning. In July 2016, Sun Jian joined Megvii as Chief Scientist and head of Megvii Research.

A former Megvii Research Institute employee said that after Sun Jian joined, he proclaimed the goal of "establishing the strongest research institute in the Eastern Hemisphere," with main research directions focusing on image classification, object detection, image segmentation, and more, which were Sun Jian's strengths.

Afterwards, related computing power and research processes were better regulated, and due to Sun Jian, more talented individuals joined Megvii Research Institute. An unconfirmed rumor is that at the time Sun Jian joined, around 70% of the research institute's researchers, including interns, came from Tsinghua University.

Looking back, when Megvii entered the security business in 2016, its first-mover advantage was undoubtedly significant: At that time, CloudWalk was still focused on its financial business, and while SenseTime was in the same race, its scale and volume were not comparable to Megvii at the time. This gave Megvii ample room for expansion in mobile phone-related fields. Also around the same year, Megvii began seriously considering the possibility of going public.

During this period, the continuously expanding security business was undoubtedly the ballast for Megvii's IPO financing. Its Face++ platform was the earliest face recognition platform in China. Launched around the 2015 Spring Festival, it was vigorously promoted starting in 2016 - by the end of 2016, Megvii's share in the public security industry exceeded 80%, with few competitors. This was also related to Ma Yuan, who was in charge of the security business at the time.

By the time of Megvii's Hong Kong IPO in 2019, the prospectus data showed that from 2016 to 2018, Megvii's revenue had a compound annual growth rate of approximately 358.5%. However, correspondingly, losses were also continuously increasing. Prospectus data showed that from 2016 to 2018, Megvii's net profit losses totaled approximately RMB 4.5 billion.

But this was not the main reason Megvii failed to list in Hong Kong. News indicated that Megvii had already passed the listing hearing in Hong Kong and was on the verge of going public, with an original plan to list on October 20 of that year. However, due to the turbulent political environment in Hong Kong at the time, some Hong Kong Stock Exchange employees requested Megvii to respond to related content, which Megvii chose to refuse and instead shifted to the STAR Market in 2021.

However, during the three-year reporting period from 2018 to 2021, Megvii generated revenues of RMB 854 million, RMB 1.26 billion, RMB 1.391 billion, and RMB 670 million, respectively; net profits of -RMB 2.8 billion, -RMB 6.643 billion, -RMB 3.326 billion, and -RMB 1.865 billion, respectively, for a cumulative loss of RMB 14.634 billion. As of the end of June 2021, Megvii's accumulated uncompensated losses were RMB 16.596 billion, with the magnitude of losses increasing rather than decreasing.

Megvii's approval process on the STAR Market was also very fast, taking less than half a year. However, during the registration phase, the regulatory incident involving Ant Financial Services Group erupted. As a major shareholder of Megvii, Megvii was inevitably affected.

It should be noted that before Megvii submitted its IPO application, Ant Group held a 15.08% stake in Megvii through API, making it the company's largest shareholder. In addition, Taobao China Holding Limited, wholly owned by Alibaba, held a 14.33% stake in the company, making it the second-largest shareholder. Ant Group and Alibaba are affiliated companies, so together, the Alibaba system held a 29.41% stake in Megvii.

In addition to holding company shares, Ant Group and Alibaba also had certain related transactions with Megvii, which also triggered inquiries from the regulatory authorities - the cooperation between the two parties began in 2014, with Megvii providing face recognition technology support and identity verification solutions for Sesame Credit, Alipay, and Zhejiang Internet Commercial Bank. During the reporting period from 2018 to 2020, Megvii's sales revenue from Ant Group and its subsidiaries was RMB 25.6662 million, RMB 9.4115 million, and RMB 22,200, respectively, with related sales revenue decreasing year by year.

This inevitably raised questions and inquiries from the regulatory authorities.

Therefore, as of November 14, 2024, Megvii chose to terminate its registration, falling at the entrance to the capital market for the second time.

During this period, founder Yin Qi was attempted to be extorted for RMB 3 million by his driver, and technical core Sun Jian, Ph.D., passed away suddenly in 2022 due to a sudden illness. Megvii, with its talent-intensive and promising business prospects, gradually drifted away from the IPO.

02 Stepping Out of the Comfort Zone

Remember in the spring of 2023, Yin Qi predicted at Megvii's enterprise business partner conference that AI would evolve in two major directions in the future:

One is AI in Digital: Technologies represented by ChatGPT bring new technological paradigm shifts to the digital world.

The other is AI in Physical: Companies represented by Tesla combine AI technology engines with hardware carriers to produce different types of intelligent machines such as autonomous vehicles and robots, transforming the physical world.

In a sense, the path chosen by Megvii in the era of large models also confirms Yin Qi's prediction two years earlier.

If we look at Megvii's main business during the IPO stage, it mainly focused on three segments: consumer IoT solutions, urban IoT solutions, and supply chain IoT solutions. Looking back two years after the explosion of generative large models, Megvii's once avant-garde business appears lagging due to the emergence of new technological singularities, which is undoubtedly a pity.

There are also some institutional factors at play. For example, from the approval to the listing, Megvii was delayed for 38 months on the STAR Market. According to China's listing system, Megvii could not conduct surprise fundraisings or transform into new businesses to "tell stories." In the second year of waiting for the listing, generative AI exploded within the industry, delaying Megvii by at least a year or more.

Performance factors also existed. Prospectus data showed that from 2018 to the first half of 2021, Megvii generated revenues of RMB 854 million, RMB 1.26 billion, RMB 1.391 billion, and RMB 670 million, respectively; net profits of -RMB 2.8 billion, -RMB 6.643 billion, -RMB 3.326 billion, and -RMB 1.865 billion, respectively, for a cumulative loss of RMB 14.634 billion. As of the end of June 2021, Megvii's accumulated uncompensated losses were RMB 16.596 billion.

Even for the emerging business of supply chain IoT solutions, the corresponding revenue share increased from 5.4% to 15.71%. For Megvii, it was also difficult to have sufficient cash flow to invest in basic model layer research and development, especially under the "high investment + high computing power" infrastructure model before Deepseek emerged.

However, this does not mean that Megvii lacks talent in the corresponding direction. For example, Zhou Xinyu, co-founder of Dark Side of the Moon, previously interned at Megvii when he was a senior at Tsinghua University and later joined Megvii through campus recruitment.

A source close to Megvii Research Institute revealed that Zhou Xinyu's work involved algorithms and he could be considered a "veteran" at Megvii. However, during the preparatory stage of Dark Side of the Moon, he and Yang Zhilin each took a group of "loyalists" from their previous companies, with some joining Step Star, further reducing Megvii's original talent density.

According to another informed source, whether it was later Dark Side of the Moon or Step Star, most of the employees originally from Megvii united under the banner of Sun Jian. However, with the fall of Sun Jian's banner, some employees who still held onto some "idealism" chose to respect "reality" and embrace the latest AI wave.

On the one hand, this may be a matter of opportunity, which is also quite regrettable.

However, after terminating its IPO on the STAR Market, Megvii was finally able to let go and step out of the comfort zone of security.

03 Horse Racing Under Geely

Almost simultaneously with the termination of the IPO, on November 11, 2024, A-share listed company Lifan Technology announced the election of Yin Qi as chairman of Lifan. From this, it can be seen that compared to devoting himself to "large models," he chose intelligent vehicles, a track that many people believe has already passed its peak.

However, from a historical business perspective, Megvii has always followed a hardware-software integrated business model, from the hot security industry to mobile phone-related businesses that have continued for more than a decade, to previously self-developed mobile robots such as AMR robots, SLAM navigation intelligent forklifts, AI stackers, and shuttle cars.

Most of Megvii's product directions utilize advanced AI technology combined with intelligent terminals to serve corresponding business scenarios.

In the eyes of some industry insiders, among the Four Little Dragons of AI 1.0, Megvii's technology is relatively more "grounded." One piece of evidence is that Megvii's related AI papers, in terms of both citation rate and indicators of conversion into productivity, are second only to OpenAI.

As early as 2021, Megvii was involved in mass-producible L2+ assisted driving solutions. However, in terms of timing, the same departments were already established by companies such as Momenta, Huawei, and DJI as early as 2016. The reason Megvii chose to enter was its patents related to visual perception technology, which it excels at.

In 2023, Megvii released its intelligent driving solution and began collaborating with Geely the same year. It is worth noting that Geely is also one of Lifan's shareholders, leading to a series of intersections between Megvii, Yin Qi, and Lifan.

Lu Jiu Business Review found that whether it is an entire vehicle brand or an intelligent driving supplier, Geely excels at internal horse racing.

Currently, Geely's brands are also in a period of integration, with brands like Geometry and Radar being integrated into Geely Galaxy and Geely Auto Group, respectively, while Zeekr and Lynk & Co have merged.

In terms of intelligent driving, in addition to Megvii, don't forget that Geely also has a team called Meizu, which is engaged in intelligence and in-vehicle integration through acquisitions and investments in earlier years.

Therefore, it can be foreseen that this round of integration is also a process of resource concentration and optimized allocation for Geely, while it is also a rare window of opportunity for various intelligent driving suppliers.

As of now, Megvii's intelligent driving solutions have been mass-produced in multiple Geely models, meaning that in the field of new energy intelligent driving, Megvii already has its own "basic plate." Whether Megvii can stand out in internal horse racing is also a question for the future.

Looking back at Megvii's more than ten years of development, the company has almost completely experienced every cycle of China's AI industry. From the technological breakthrough in dynamic face recognition in 2015, to the failed attempt to list on the Hong Kong Stock Exchange in 2019, to the termination of registration on the STAR Market in 2024, its fate has always been closely linked to the tides of the times.

As the flood of generative AI sweeps in, Megvii's hesitation in the capital market and the pain of business transformation have become a typical sample for observing the changes in China's AI industry.

Under the dual pressure of technological iteration and capital cycles, Megvii's transformation choices are quite enlightening. When most AI companies choose to go All in on large models, Yin Qi is betting on the tracks of intelligent driving and robotics.

This seemingly "retrograde" strategy actually coincides with its underlying logic of "AI in Physical" -

From security cameras to mobile robots and then to intelligent vehicles, Megvii has always been exploring the interface between AI technology and the physical world. Its deep ties with Geely and the acquisition of control over Lifan Technology demonstrate its determination to build a new growth curve through automotive intelligence. However, this transformation path is not smooth.

In the field of intelligent driving, companies such as Huawei and DJI have already formed technological barriers; in the robotics sector, companies like UBTECH and Pioneer Robotics have gained first-mover advantages. How Megvii's proud visual perception technology can maintain competitiveness in a multi-modal fusion intelligent driving system remains an unsolved proposition. Moreover, the pressure of losses that have persisted for more than a decade requires it to prove itself in commercialization.

So far, Megvii still has many opportunities. Its accumulated over 4,000 AI patents, mass production experience in the L2+ assisted driving field, and the technical legacy left from the Sun Jian era may all become leverage points for the future. As the industry shifts from 'brute force computing' to 'scenario-based cultivation,' Megvii's accumulation in hardware-software integration and vertical integration may usher in a new valuation reassessment in the post-AI era.