Professor Li Zexiang’s robotics empire: Over 280 startups, a 50 billion RMB valuation

Series Preface: The Chinese embodied AI startup ecosystem is a martial arts world written by professors, academicians, and their students. This six-part series dissects the familial, geographic, and financial ties within this circle. In Part Two, we explore the martial arts world built over three decades by Li Zexiang and his robotics army nurtured in Songshan Lake.

In December 2025, two companies incubated by the same professor consecutively went public on the Hong Kong Stock Exchange.

One is Xidi Intelligent Driving, the first listed commercial vehicle intelligent driving company on the Hong Kong Stock Exchange; the other is Woan Robotics, the world’s first publicly listed AI embodied home robot company, with a market capitalization of nearly HK$18 billion on its first day.

This professor is Li Zexiang.

Looking back: In August 2023, Googol Technology listed on the ChiNext Board; looking ahead: In June 2026, Benmo Power passed its HKEX hearing; additionally, Hailu Innovation has submitted its application, and Yunjing Intelligence is advancing toward its Pre-IPO stage.

These companies are all closely related to Li Zexiang.

A university professor, over three decades, "produced" an entire IPO assembly line, having incubated more than 280 companies within the system, with a combined valuation exceeding 500 billion yuan. How did this system come into being?

The story begins in a laboratory in Clear Water Bay, Hong Kong, in 1992. To understand the past three decades, it’s essential to note a key context: Li Zexiang has remained continuously employed at HKUST since joining in 1992, without resigning or retiring.

His thirty years are divided into two phases—the HKUST period (1992–2014) and the Songshan Lake period (2014 to present)—two distinct approaches, yet fundamentally aligned.

In 1978, at age 17, Li Zexiang, a teenager from Lanshan, Yongzhou, Hunan, topped his county’s entrance exam and was admitted to Central South College of Mining and Metallurgy. The following year, after the establishment of diplomatic relations between China and the United States, Alcoa sent a delegation to China and left behind scholarship opportunities; Li became one of China’s first batch of government-sponsored undergraduate students sent abroad, traveling to the United States to earn his bachelor’s degree at Carnegie Mellon University. He later obtained his master’s and doctoral degrees from the University of California, Berkeley, completed a postdoctoral fellowship at MIT, and secured a faculty position at New York University.

If the story ended here, the world would have one more typical overseas Chinese professor.

But in 1992, 31-year-old Li Zexiang made a decision that puzzled his colleagues: he gave up his teaching position at New York University and returned to Hong Kong to join the Hong Kong University of Science and Technology, which had been established just a year earlier, as a faculty member in the Department of Electronic and Computer Engineering.

What he brought back was a teaching method considered highly unconventional at the time.

The Hong Kong University of Science and Technology’s Department of Electrical and Computer Engineering assigned him Laboratory 3126—he interpreted this number symbolically: “3 professors, mentoring 12 students, building a prototype in 6 days.” The Automated Technology Center (ATC) thus became the most hands-on hub at HKUST.

The rule for 3126 is: teachers don’t do projects for students; they only help identify real problems, find resources, and secure funding. Students must turn their ideas into working prototypes within a few days. This “student-led, teacher-supported” approach was later adopted by Li Zexiang in nearly every company he was involved with.

During this period, the companies incubated by Li Zexiang had extremely pure lineage—either colleagues or students he personally mentored.

The first company was co-founded by colleagues.

In 1999, he convinced seven or eight colleagues from the Hong Kong University of Science and Technology to move to Shenzhen and found Googol Technology, developing motion controllers—the underlying "nervous system" of industrial robots and CNC machines.

This is China's first company to possess fully independent intellectual property rights for motion controllers, and it subsequently held the top position in China's motion controller market for many consecutive years.

At this time, his identity was still that of a professor at HKUST—HKUST permitted professors to engage in industrialization activities for one day per week, and with weekends included, he could dedicate three days per week; he later recounted that during those years, after working in Shenzhen on weekends, he had to return to campus and “write self-criticisms.”

In August 2023, Googol Technology listed on the ChiNext Board, known as the "first stock in motion control"; to this day, Li Zexiang remains the chairman of Googol.

Gao Gu proved that "professors starting businesses" is viable, but it was a student named Wang Tao who made Li Zexiang truly understand "how education should be done."

Around 2003, Wang Tao was studying electronics at the Hong Kong University of Science and Technology.

His graduation project was a remote-controlled helicopter flight controller—during the defense, the helicopter fell from mid-air, the demonstration failed, and he received only a C. But Li Zexiang remembered this student from the Robocon robotics competition: around 2005, Wang Tao, as team captain of the HKUST team, won the Hong Kong championship and third place internationally at Robocon.

Li Zexiang unusually admitted Wang Tao as a graduate student.

In 2006, Wang Tao, along with a few others, founded DJI in a warehouse of less than 20 square meters in Chegongmiao, Shenzhen—the building above was the Sunshine Golf Tower, and below were a restaurant and a machining factory. Later, Li Zexiang recalled: “In the early days, when Wang Tao came to Shenzhen for prototyping, the machining factory was right next to the restaurant—by the time he finished eating, the prototype was already done; it would take more than a week in Hong Kong and over a month in the U.S.”

DJI nearly died in its early days: after the release of its first mature product, the XP3.1, in 2008, the three co-founders besides Wang Tao left one by one, and the company ran out of funds.

Li Zexiang invested money and recruited the first cohort of graduate students from the teaching reform program at Harbin Institute of Technology’s Shenzhen Graduate School (co-founded by him in 2004, which trained over 300 students over six years) to bolster the team—these individuals later formed the backbone of DJI’s middle management. In 2012–2013, the Phantom all-in-one aerial photography drone ignited global demand; in 2013, Li Zexiang traveled to Silicon Valley and persuaded Sequoia’s Moritz to come to Shenzhen, leading Sequoia to invest tens of millions of dollars.

After that, DJI hardly needed external funding—it was essentially a money printer.

In 2019, Li Zexiang and Wang Tao jointly received the IEEE Robotics and Automation Award at the ICRA ceremony in Montreal.

After DJI became the global leader in consumer drones, Li Zexiang stepped back to the periphery. He explained himself: “The main thing was helping them cultivate and find talent. The day-to-day operations of the company have always been managed by them.”

The path of guiding students to start businesses doesn't always work.

In his early years, he partnered with eight students to start a business making LED wire bonding machines; the product was so successful that “farmers came with sacks of cash to pick up equipment.” Eventually, the eight founders argued for a month and shut down the company. His venture with students from the Harbin Institute of Technology Shenzhen Graduate School’s teaching reform class, Biri, also failed to survive.

Two failures convinced him that entrepreneurial education must begin at the undergraduate level and emphasize teamwork—“like walking at night: alone, you’re afraid; together with others, you have a flashlight and a stick to ward off dogs.”

At the end of the HKUST era, there were two companies that served as a bridge between eras.

In 2011, his doctoral student Shi Jinbo founded Liqun Automation at Songshan Lake to develop domestic high-end industrial robots—the company’s name derives from the mathematical concept of “Lie Group” (a continuous symmetry group named after Norwegian mathematician Sophus Lie), which forms the mathematical foundation for describing robotic motion configurations.

In 2012, Tao Shizheng, the captain of the Hong Kong University of Science and Technology robotics team he mentored, began developing electric outboard motors on campus and founded Yidong Technology.

The founders of these two companies are classic "1.0 lineage"—trained within competitive circles and mentored within their academic lineage—but they established their companies directly in Songshan Lake, serving as a bridge between the two eras.

Call the twenty-two years from 1992 to 2014 the "HKUST Era," whose model can be summarized as 1.0: driven by faculty-student relationships and incubated by individual labs.

These companies emerged from the projects and graduation designs at 3126: Googol was founded by faculty members, while DJI, Liqun, and Yidong were founded by students. Over 100 graduates from the 3126 program around that time, more than a third, went on to become entrepreneurs.

But this model has a natural limit: a single lab can accommodate only a dozen or so students at a time; the radius within which he recruits students extends no farther than the Robocon competition venue.

In 2014, Li Zexiang proposed an idea to the Dongguan municipal government: to establish a robotics industry base at Songshan Lake. The government would provide land and funding, while operations would follow a corporate model. Li Zexiang, together with two longtime colleagues—Gao Bingqiang, former Dean of the School of Engineering at HKUST (a renowned scholar in integrated circuits and later an independent director of澜起科技), and Gan Jie, professor of finance at Cheung Kong Graduate School of Business—sold portions of their shares in DJI to serve as startup capital, leading to the establishment of the XbotPark Songshan Lake International Robotics Base.

The campus faces Huawei’s Songshan Lake base. Covering over 60,000 square meters with 110,000 square meters of building space, it includes classrooms, laboratories, a shared factory, dormitories, a cafeteria, a library, a sports field, and even a children’s center—more like a miniature university than a traditional industrial park.

During the HKUST era, incubation relied on mentor-student relationships—he only invested in people he had personally taught; during the Songshan Lake era, incubation relied on the platform—the hub opened its doors to all young people whose eyes sparkled with passion.

This is the key difference between the two eras: 1.0 was driven by teacher-student relationships, while 2.0 is driven by platform and fund dynamics. First comes the ecosystem, then the teacher-student relationships; anyone who enters the ecosystem, regardless of their background, is called “Teacher Li.”

Zhongtouwang summarized Li Zexiang’s complete ecosystem: competitions identify talent—labs cultivate talent—incubation bases nurture projects—funds provide the first investment—Pearl River Delta supply chains enable mass production—venture capital follows with additional funding—exit via capital markets.

The production line is set up—where do the "raw materials" come from? The case studies from the Songshan Lake period each illustrate the issue more clearly than the last.

The first company to move into the base was Yidong Technology—spun out from HKUST and nurtured by the base. The management committee specially built a dock by the lake for it to test boats. This marked the physical transition from 1.0 to 2.0.

The first person brought in from outside was Zhang Junbin from Yunjing—this path was the most extreme and best illustrated the openness of 2.0. In 2015, a 25-year-old young man from Chaoshan submitted his resume to Li Zexiang’s WeChat public account.

He earned his bachelor’s degree from the School of Mechanical Engineering at Huazhong University of Science and Technology and his master’s degree from the Michigan Institute at Shanghai Jiao Tong University. He has been tinkering with robots since childhood and won a gold medal at the International Youth Robotics Competition in 2006—note that he is not, in any sense, a “disciple of Li.” Li Zexiang invited him to dinner; after their conversation, he said to him: “Come to Songshan Lake—the base isn’t built yet, but you can come ahead.” Later, Zhang Junbin experimented with robots for picking up golf balls and navigation aids for the blind at the base, both of which failed, until his father complained at a family gathering, “Bending over to mop the floor is so tiring.” That’s when he began developing a robotic vacuum cleaner.

In October 2016, Yunjing was registered and established at Headquarters No. 1 in Songshan Lake. Zhang Junbin later said: "When I carried a backpack, Teacher Li was willing to support me, investing in me round after round."

Chen Yuqi of HAIRO Innovation represents the second path: a returnee. Born in 1989 in Wuhu, Anhui, he earned a first-class honors bachelor’s degree in Electronics from The Hong Kong Polytechnic University and a master’s degree in Robotics from ETH Zurich, with his graduation project focused on bin-robotics.

In December 2016, he co-founded Hairo Innovation in Shenzhen with two fellow alumni. He studied at PolyU and ETH, not as a student of Li Zexiang—but in multiple public sources, Li Zexiang is listed as a "mentor and chief advisor" to the Hairo team. This is typical of the 2.0 model: the mentor-student relationship is established only upon entering the ecosystem.

Woan represents a third path: proactively seeking acquisition. Woan’s founders, Li Zhichen and Pan Yang, are alumni of Harbin Institute of Technology. They founded the company in Shenzhen in 2015 and developed the SwitchBot Bot—a 35-gram, average-priced device at 126 yuan—that sticks onto switches as a "fingertip robot." To achieve an actuation accuracy of just 0.1 millimeters, the team devoted three months of relentless effort.

In 2018, the company hit a bottleneck: its successful crowdfunding failed to translate into stable revenue, and fundraising efforts repeatedly hit roadblocks. Li Zhichen made a critical decision: to proactively bring the project into the Songshan Lake XbotPark incubation ecosystem.

In December 2018, funds including XBOTPARK completed their Pre-A round investment. Later, Li Zexiang explained why: the team’s technical resilience, combined with a pragmatic approach—“actively avoiding the hype around humanoid robots and focusing on distributed, scenario-specific products.”

Li Zexiang calculated the metrics for Phase 1.0 of the incubator (2014–2020): 60 startup teams, with 80% surviving and a 15% rate of unicorns and near-unicorns. This survival rate is almost another species within the incubation field.

The factory is the visible production line; money is the invisible one.

First, let’s trace the origin of this funding pipeline: After the Songshan Lake base was established in 2014, a孵化 fund was launched the following year (2015)—initially named the “Hong Kong Shing Mun Bay Venture Fund,” later restructured into the XBOTPARK Fund, which exclusively invests in seed and angel rounds through the base.

In July 2016, Li Zexiang, together with Sequoia’s Shen Nanpeng and Tencent’s Ma Huateng, co-founded the Hong Kong X Technology Fund (initial capital of HK$300 million, investing in over 50 companies over five years). In August 2022, Dongguan Qingshuiwan Phase II Venture Investment Partnership was established (completed registration in 2023), with a registered capital of RMB 727.5 million. The total capital scale of funds within the system exceeds RMB 2 billion.

Lead Fund One: The operating entity of the XBOTPARK Fund is the Dongguan Songshan湖 International Robot Research Institute. Lead Fund Two: Tsing Chuen Capital—named after Tsing Chuen, the location of HKUST, specializing in succession.

According to Li Zexiang’s interview with China Entrepreneurs, the LPs of these two funds consist of five types of investors:

● Top VC firms: Sequoia China, Hillhouse Capital

● Industrial Capital: Tencent

● State-owned capital and government-guided funds: Shenzhen Capital Group, Dongguan Industrial Investment Master Fund

● University funding: The University of Hong Kong invested HK$70 million to acquire equity in Qingshuiwan Phase II—this is a rare instance of a university directly investing in a private equity fund on the mainland.

● Faculty personal contributions: Professor Gao Bingqiang: RMB 20 million, Professor Gan Jie: RMB 12 million, and several other professors

In other words, Li Zexiang used his own credit as the GP, convincing Sequoia, Tencent, Shenzhen Capital Group, the University of Hong Kong, and Dongguan’s state-owned assets to become his LPs—so-called “the mentor’s money” is backed by the entire Chinese venture capital ecosystem.

This division of labor emerged naturally: XBOTPARK Fund handles “launching” — providing the initial funding and securing the earliest equity stake; Qingshuiwan Capital handles “mentoring” — following through from the seed round all the way to the Series D, guiding promising startups all the way to the IPO doorstep.

As of early September 2026, according to the IT Juzi database, XBOTPARK Fund - Songshan Lake Robotics Institute has made a total of 76 investments, while Qingshuiwan Capital has made 91. When these two investment curves are plotted out, three clear trends emerge.

Trend One: The XBOTPARK Fund does only one thing—makes the first investment. The vast majority of its 76 investments are in seed or angel rounds, with relatively small amounts: the seed round investment in Benmu was RMB 300,000 (April 2020), the seed round in HAIROBOTICS was several million RMB (January 2017), and the angel round in Sunlin was several million RMB (August 2016). Its very first external investment occurred in May 2015—RMB several million to Zhang Junbin, who applied via a WeChat public account, marking the beginning of Cloud Whale.

From an investment stage perspective, XBOTPARK’s strategy is to take only 5%–13% early-stage equity and generally does not follow up after the Series A round, leaving room for mainstream VCs.

Trend Two: Qingshuiwan Capital led investments from the angel round through Series D. Its first investment in 2017 was in EcoFlow—the first case of spillover from the DJI ecosystem. Although founder Wang Lei was not a direct disciple of Li Men, he embodied the far-reaching influence of this system best.

Subsequently, Qingshuiwan appeared repeatedly within the system’s portfolio companies: SIDAI Intelligence’s Series A (March 2018, $30 million) and Series A+ (August 2020, RMB 100 million), Yidong Technology’s Series D (August 2024, hundreds of millions of RMB), Shenyuan Biology’s Series B+ (RMB 300 million), and Buting Technology, from Pre-A all the way through to Series A+.

According to IT桔子 data, the two institutional entities have jointly invested in over 25 companies (e.g., Yunjing, Woan, Benmo, Wuqiong Innovation, Zhuoyi Technology, Yinghansi Power, Hengzhi Future, etc.). The investment sequence is clear: XBOTPARK leads the first round, followed by Qingshuiwan—the mentors’ capital accompanies the company from the angel stage all the way to the IPO doorstep.

Re-investments by these two institutional entities in the same company are standard practice: Yunjing received three rounds of investment from XBOTPARK (Angel → Series A → Series C+), Woan received two rounds (Pre-A: RMB 5 million → Series A: RMB 12 million), Wuqiong Innovation received three rounds, and Zhuoyi, Yinghansi, and Aikesibote each made two rounds (Angel + Series A).

The most extreme case was XbotPark: a 300,000 RMB pre-seed round (April 2020), followed eight months later by a 3 million RMB angel round (December 2020), with continued support all the way to the HKEX hearing—on a diluted basis, that initial 300,000 RMB investment returned approximately 400 times.

The industries of these portfolio companies also reveal the strategies of Li Zexiang’s two institutional entities: beyond robotics and advanced manufacturing, the portfolio includes Homan, which focuses on pet intelligence; Desert Farmer, which specializes in agricultural breeding; Keyi Surgery, which develops medical devices; and Walnut Intelligence, which produces sports equipment... This confirms Li Zexiang’s “category thinking”—he doesn’t invest in “robot companies,” but in categories that can be reinvented through hardware.

At the other end of the pipeline are the products—and also the report cards. Ranking Li Zexiang’s companies by their level of recognition in the capital market yields exactly three tiers: publicly listed (including those in the IPO pipeline), unicorns, and potential unicorns. Let’s first take a broad look, then examine a few in detail.

Tier 1: Listed and Upcoming IPOs — Stretch out the timeline to clearly see the capitalization process:

Tier 2: Unicorn (unlisted, valuation of $1 billion or more / RMB 10 billion or more):

Tier 3: Potential Unicorn (valuation between RMB 1 billion and USD 1 billion):

Adding up all three categories, what’s most remarkable about this diagram isn’t the quantity, but the coverage: from motion control (GaoKong) to joints (BenMo, LiQun), from chassis (SongLing) to complete systems (YunJing, WoAn, HaiRou), from water surfaces (YiDong) to mines (XiDi)—Li Zexiang’s ecosystem has embedded the very concept of “embodied intelligence” into the capillaries of both industrial and household applications.

I've finished reviewing the list; now let's look at the real secret behind this IPO machine—how the money moves.

The two institutional entities from Chapter 4 participated in nearly every first round of funding for the above companies: HAIRO’s first investment (Seed round, Jan 2017), Benmo’s first funding (Apr 2020, RMB 300,000), Cloud Whale’s angel round (May 2015, lead investor), Woan’s Pre-A round (Dec 2018, in a dire situation), Songling’s first funding (Angel round, Aug 2016), and Xidi’s Series A round (Mar 2018, Qingshuiwan)—all were theirs.

Mainstream VCs enter only after the mentor’s capital. Names like Sequoia, 5Y Capital, Hillhouse, Yuanma, Jinri Capital, and Tencent only appear once a company has a product, revenue, and “proven certainty.” VCs buy certainty; Li Zexiang’s fund buys “people”—betting the smallest amount of money at the most uncertain stage on competition champions, doctoral theses, and students whose eyes light up.

This model has a counterintuitive feature: it deliberately avoids taking large equity stakes. It typically holds only 5%–13% of early-stage shares, steps back after the Series A round, and gets diluted to 0.2%–6% before IPO (Wo An’s 12.98% is a rare exception, as it includes investment from the Songshan Lake Research Institute’s incubation efforts; Benmo’s underlying stake is 3.39%). In 2024, funds within its system conducted a “exit drill”—transferring partial secondary shares from flagship projects like Yunjing to new funds, recycling capital to invest in new ventures.

Looking together at Li Zexiang’s thirty years—the 1.0 era at HKUST and the 2.0 era at Songshan Lake—this is a victory of craftsmanship, not a victory of model.

The essence of the master-apprentice system and incubator model cannot be captured in any manual—it is tied to a specific individual.

The deepest moat in Li Zexiang's system is not a specific technology, but the three layers of assets accumulated over thirty years:

A talent funnel centered around Robocon/RoboMaster (recognized by the competition community as the “godfather”); a talent development pipeline from the HKUST reform class to the Shenzhen Institute of Science and Technology (where his former students return to mentor new ones); and the campus in Dongguan’s Songshan Lake, directly facing Huawei—where classrooms, factories, funds, and dormitories form a walkable, self-contained ecosystem.

Shenzhen Institute of Science and Technology selects 80 students each year from over 1,300 applicants; at the graduation pitch event, the judges seated in the audience may be partners from Clear Water Bay. In this system, entrepreneurship is not just a line on a resume—it’s a craft broken down into structured courses.

When the wave of embodied intelligence arrived, they found themselves standing on the sidelines.

After 2023, when "embodied intelligence" became the hottest term in the venture capital market, the three most scarce assets suddenly became: hardware capable of mass production, real-world data, and global distribution channels.

And this is precisely the accumulated expertise of Li Zexiang’s ecosystem over the past two decades: Yunjing’s robotic vacuums serve as a home data entry point (with over one million units deployed featuring binocular vision), Songling’s data collection bases sell training data, Benmo’s joints are components purchased by every humanoid robot manufacturer, and Woan controls household distribution channels in over 90 countries.

They don’t need to switch industries to become embodied; they just need to add an AI brain to their existing assets.

This article is from the WeChat public account "IT Juzi" (ID: itjuzi521), authored by Judy.