The Chinese internet has finally ushered in a new era of scientists.

On the early morning of August 12, Lin Junyang announced on X the establishment of Pragmatic Tech, a company based in Shanghai with a focus on next-generation intelligent agents.

He further explained, "I got into linguistics because a friend recommended pragmatics to me. Later, I shifted to computational linguistics and natural language processing. The name means returning to where things happen. Moreover, it represents pragmatism, and we believe AGI should take this as its goal."

A few months after leaving Alibaba's Qwen team, Lin Junyang officially became an entrepreneur and quit drinking. Institutions including Gaorong Ventures, HSG, and Tencent appeared on his financing list, and market sources put the company's valuation at 2 billion US dollars. The reason why capital institutions are willing to offer such a high price largely stems from Lin Junyang's industry achievements and personal qualifications in the field of large models over the past few years.

Born in 1993, Lin Junyang holds a bachelor's degree from the University of International Relations and a master's degree from Peking University. After joining Alibaba Damo Academy in 2019, he entered the core Qwen team, led the R&D of the Qwen series, which became an open-source large model with global influence. On March 4, he announced his departure from Alibaba, writing "me stepping down. bye my beloved qwen", and now he is making his comeback as a legend.

Lin Junyang's entrepreneurial journey coincides with a highly anticipated moment. China's technology industry is re-elevating the status of scientists. More and more researchers are entering the core of model teams, heads of foundation models are directly connected to the highest decision-making system of companies, and young researchers can quickly obtain capital support based on their technical judgment after leaving large factories. The knowledge mastered by scientists is increasingly becoming a scarce industrial power again.

The "scientists" we are talking about here specifically refer to industry-oriented scientists. Most of them have received systematic training in computer science, artificial intelligence or related disciplines, have master's or doctoral research experience, or have long been engaged in cutting-edge scientific research. After entering the industry, they stand at the intersection of cutting-edge research and corporate decision-making. Their core competence is to judge technical directions when there are no ready-made answers, and translate research into engineering paths. The truly scarce value thus extends from academic ability to resource allocation, where computing power, R&D budgets and teams often follow technical judgments.

China's technology industry has experienced similar power shifts many times. Whenever technology enters a mature stage, the importance of the market, channels and scale will rise; when the industry re-enters uncharted territory, people who can find new answers will move back to the center of the stage.

Lin Junyang's establishment of the new company is just one recent episode, and similar changes are taking place in more companies. Baidu placed Wu Tian, head of foundation model R&D, directly in the reporting line of Li Yanhong; Tencent introduced Yao Shunyu, an undergraduate from Tsinghua University's Yao Class and a PhD in Computer Science from Princeton University, to be in charge of large models and AI infrastructure; at ByteDance, Wu Yonghui, who holds an undergraduate degree from Peking University and a PhD from the Hong Kong University of Science and Technology, leads the R&D of the Seed foundation model. Among startups, Yang Zhilin, founder of Kimi, graduated from Tsinghua University for his bachelor's degree and Carnegie Mellon University for his doctorate; Yan Junjie, founder of MiniMax, received his doctorate from the Institute of Automation, Chinese Academy of Sciences; Liang Wenfeng, founder of DeepSeek, also has a long-term background in technical R&D.

Whether the technical core of large factories is directly connected to the highest decision-making level, or scientists step forward to take full charge of startups, a group of industry-oriented scientists are moving from the R&D system to the center of corporate operation.

Problems have thus emerged.

Scientists once stood at the center of the industry, then were overshadowed by capital, channels, platforms, marketing and traffic, and regained power again in the era of artificial intelligence. Over the past decades, why have scientists repeatedly moved to the center of the industry in different technology cycles and then stepped back behind the scenes after technology matures? What exactly determines the flow of power within technology companies?

01 Scientists Stand Before The Birth Of Companies

On July 27, 1979, in a computer room at Peking University, a prototype of a precision Chinese character phototypesetting system completed debugging. After the keyboard stopped making sounds, the laser phototypesetter spit out a sheet of film the size of an eight-page newspaper. When Wang Xuan took over the film, China's reform and opening-up had just entered its first year, personal computers had not yet entered the lives of ordinary people, and modern Chinese technology companies had not really taken shape.

Wang Xuan had been working on the machine in front of him for four years.

In 1974, the state launched the Chinese Character Information Processing System Project, known as the "748 Project". Wang Xuan was responsible for the overall design of the precision Chinese character phototypesetting system. At that time, Europe and the United States were using the third-generation cathode ray tube phototypesetter, Japan was popularizing the second-generation mechanical phototypesetter, and the domestic relevant technical foundation was weak. Wang Xuan judged that if they followed the existing foreign routes to catch up, their products would still lag behind when launched, so he directly chose the fourth-generation laser phototypesetting that had no mature commercial products at that time.

Choosing an independently developed technical route carried far more weight in that era than it does today. China's press and publishing industry was still operating in the era of "lead and fire". To turn a manuscript into a printed newspaper, it required heavy processes such as type picking, typesetting, and plate casting, and formulas, symbols and complex layouts in scientific and technological books were particularly difficult to handle. By the mid-1980s, more than 3,800 scientific and technological books had been backlogged for more than two years without being published nationwide.

Technical problems are also industrial problems.

In 1985, the Huaguang Type II passed the national appraisal and user acceptance of Xinhua News Agency. Two years later, *Economic Daily* decided to use Huaguang Type III to publish its daily newspaper. Once put into actual operation, the machine soon exposed problems such as repeated characters, repeated lines, missing characters, difficult title movement, and scanning jitter, and the newspaper's publication was once delayed due to failures. The newspaper gave the R&D team ten days to solve the problems, otherwise they would go back to lead typesetting. Wang Xuan and his team worked continuously to fix the faults, and achieved stable newspaper printing in August of that year. In 1988, the printing house of *Economic Daily* sold all its lead characters, making it the first newspaper in China to completely bid farewell to lead typesetting.

The technology in the laboratory finally crossed a narrow threshold and truly entered the industrial world. Later, Wang Xuan developed a habit of carrying a magnifying glass with him on business trips. When he arrived in a city, he would buy a local newspaper, examine the pages carefully to judge whether Peking University's typesetting system was used, and find problems in the typesetting effect. Scientific research, product effect and market research were almost inseparable in Wang Xuan's work.

Another change was taking place in China in the 1980s. Intellectuals from scientific research institutes gradually moved to the market, and a number of technology enterprises with distinct scientific research institute backgrounds emerged in Zhongguancun. For the first time, technical personnel had to face prices, sales, customers and competition. Whether scientific research achievements could be turned into commodities gradually became a more realistic issue than winning awards.

Wang Xuan had a sense of "indebtedness" towards this. The state had invested about 10 million yuan in Chinese character laser phototypesetting. If the achievements stayed in the laboratory, no matter how many awards they won, the investment could not be recovered. He later repeatedly emphasized the concept of "standing on solid ground while reaching for the sky": "reaching for the sky" refers to cutting-edge technology, and "standing on solid ground" focuses on commercialization, promotion and services. Around 1988, the Peking University New Technology Company gradually developed into the Founder system; in 1992, Peking University Founder Group was officially established.

The market gave direct feedback. The price of domestic laser phototypesetting systems was once only about one-fifth of that of imported products. By the end of 1989, many foreign phototypesetting system enterprises that had entered the Chinese market withdrew from competition. By the 1990s, Wang Xuan's team's technology once covered the vast majority of the domestic newspaper industry market. Founder grew from a scientific research achievement into a large technology enterprise.

A similar story took place at another scientific research institution in Beijing.

In 1984, the Institute of Computing Technology of the Chinese Academy of Sciences founded the New Technology Development Company, which later developed into Lenovo. Lenovo's important early product, the "Lenovo Chinese Character System", came from the scientific research achievements of the Institute of Computing Technology. Ni Guangnan later served as chief engineer and continued to promote product development. In 1987, the Lenovo Chinese Character System had become the company's flagship product.

In the early stage of development, Lenovo soon encountered the other side of scientific research institutions moving to the market. R&D personnel pursued performance leadership and launched the immature Type IV Chinese card, which caused a large number of product problems and subsequent returns. Many years later, Liu Chuanzhi recalled that research institutes were used to evaluating work based on papers, awards and scientific and technological achievements. After running an enterprise, the evaluation criteria became whether the product could be sold and whether customers were willing to use it. Sales, maintenance, production, capital and market promotion gradually gained the same weight as R&D.

A new dividing line had emerged. Technology determines the starting point of a company, and the market gradually determines how far technology can go.

From the 1980s to the early 1990s, the primary difficulty facing China's technology industry was still "whether it can be made". How to input Chinese characters into computers, how Chinese newspapers can get rid of lead characters, how microcomputers can adapt to the Chinese environment - there were no ready-made answers to a large number of problems. Capital could buy equipment, sales teams could find customers, and the truly scarce resource was people who could find technical answers.

Scientists and engineers thus naturally had industrial power. Technical routes determine products, products give birth to companies, and companies then promote changes in the entire industry. Wang Xuan existed before Founder, and Ni Guangnan's technology also appeared before Lenovo was founded. Many technology enterprises of that era can be traced back along their products, all the way to universities, the Chinese Academy of Sciences, and a specific laboratory.

However, when machines could be manufactured and products could be produced, and the Chinese market was also expanding rapidly, a new problem emerged - after the products were made, who could sell them across the country?

The first large-scale power shift in China's technology industry also laid its groundwork here. The next group of people to step into the center of the times would less often wear white lab coats. They were more familiar with the market, channels, capital and users, and would redefine the most important capabilities of a technology company in the following two decades of the Internet boom.

02 The Internet Arrived, And Scientists Stepped Behind The Scenes

In January 2000, Ma Huateng and several colleagues were raising money everywhere for Tencent. QQ's user base was growing rapidly, and server and bandwidth costs were also rising. The more users there were, the faster the company burned money. At the end of 1999, Tencent received about 2.2 million US dollars in investment from IDG and Hong Kong PCCW; in June 2001, South Africa's MIH invested about 32 million US dollars in Tencent.

At that time, the number of simultaneous online QQ accounts had exceeded 1 million. A new business rule gradually emerged: even without factories or core hardware, Internet companies could gain recognition from capital as long as they continued to accumulate users.

Users were becoming the new scarce resource in China's technology industry. At the end of 2000, the number of Chinese Internet users reached 22.5 million, an increase of 5.6 million from six months earlier, and the number of websites increased from about 27,000 to 265,000. At that time, more than half of the Internet users were under 24 years old, and the Internet was still a new world for young people. Twelve years later, the number of Chinese Internet users rose to 564 million, and mobile Internet users reached 420 million. The Internet completed its expansion from a niche communication tool to an infrastructure for public life.

The power structure of technology companies changed accordingly. Computers could already process Chinese characters, Internet protocols, servers, databases and personal computers gradually matured, and a large number of entrepreneurs no longer needed to find the reason for the existence of their companies from basic scientific research. New challenges were placed in front of everyone: after tens of millions or even hundreds of millions of users went online, how to retain them, how to generate transactions, and how to convert massive traffic into revenue.

The problems Ma Huateng faced were very representative. The real asset of QQ was not a technical breakthrough that could be written into a paper, but the increasingly huge social network. At the end of 2009, the number of active accounts on Tencent's instant messaging platform reached 520 million, with a peak of 93 million simultaneous online accounts; on March 5, 2010, the number of simultaneous online QQ accounts exceeded 100 million for the first time. After users connected to each other, the value of the existing network increased with every additional person. Scientific research achievements could be caught up by competitors, but the social relations of hundreds of millions of people were difficult to migrate again.

E-commerce demonstrated the power change even more clearly. In 2003, eBay already controlled Eachnet, which occupied most of China's C2C market share that year. Taobao entered the market in the same year, adopted a strategy of free store opening, and launched Alipay and Taobao Wangwang, embedding payment guarantee and instant communication between buyers and sellers into the transaction process. Eachnet's original charging system had a mature overseas business model, while Taobao chose to reorganize the platform around the transaction habits of Chinese sellers and buyers. In just a few years, the market positions reversed as their fortunes shifted.

Fewer and fewer problems that determine the outcome of competition are found in laboratories, and more are found in user acquisition, payment credit, merchant operation and platform rules. Ma Yun's most important work was not to solve a computer science problem, but to judge whether Chinese consumers dared to give money to unfamiliar sellers, why merchants were willing to move online, and who the platform should charge. Technology still supported transactions, but the judgments that really determined the direction of the company belonged to operators.

Liu Qiangdong took another path. After the SARS outbreak impacted offline business in 2003, JD gradually shifted to online retail. After getting the first round of financing in 2007, Liu Qiangdong decided to invest in self-built logistics. At that time, the industry generally questioned the heavy asset model of Internet companies undertaking warehousing and distribution. JD still continued to invest, and logistics eventually became an important part of user experience and the company's competitiveness.

The Internet thus redefined "technology companies". The core competitiveness of a company could come from social relations, merchant networks, logistics systems, or user habits accumulated by the platform. There were still a large number of engineers, and code still determined whether products could run, but scientists were less and less present at the very center of corporate power.

The people the era admired also changed.

The representative moment of Wang Xuan's generation was a national scientific research project solving long-term technical problems. Entering the golden age of the Internet, the names tracked by business media every day became Ma Yun, Ma Huateng, Ding Lei, Zhang Chaoyang, Li Yanhong and Liu Qiangdong. Social discussions gradually shifted from technical routes to business models, financing, listing, market value and the number of users. The most important figures inside companies were also increasingly close to registered users, active users, transaction volume and revenue.

The underlying law is not complicated. When technology is sufficient to support products, large-scale growth becomes a more difficult task. People who can find users, build channels, obtain capital and organize transactions naturally gain more power over resource allocation. Scientists did not leave the technology industry, but stepped back behind the increasingly huge commercial machine.

In 2012, the mobile Internet was still rapidly attracting new users. Mobile phones became an increasingly important entry to the Internet, and Tencent, Alibaba and Baidu were all looking for new positions in the mobile era. In the same year, in an image recognition competition on the other side of the ocean, a neural network called AlexNet suddenly widened the gap with traditional methods.

A problem that had been overshadowed by the prosperity of Internet business for many years re-emerged: when existing technology could no longer answer the future, who would decide where to go next?

Scientists regained the opportunity to enter the game.

03 The First Return Of Scientists

In March 2014, by the swimming pool of a hotel in Palo Alto, California, USA, Yu Kai, then executive deputy dean of Baidu's Deep Learning Research Institute, had a long talk with Andrew Ng.

Yu Kai hoped that Andrew Ng would join Baidu. Two months later, Baidu announced that Andrew Ng would serve as chief scientist, in charge of Baidu Research Institute and the "Baidu Brain" project. Andrew Ng had previously participated in the creation of Google Brain and was a professor at Stanford University. An artificial intelligence researcher with global academic reputation came from Silicon Valley to Beijing and joined the highest technical management sequence of a Chinese Internet company.

A few years ago,