Each generation has its own Whampoa Military Academy.

In the context of the Chinese Internet, being regarded as the "Whampoa Military Academy" by peers is often the beginning of a company's decline into mediocrity, or rather, a dignified and tragic farewell.

This sounds like a paradox. The more hardcore your business is and the higher the density of talents, the more lethal the chips you cultivate for your competitors.

When an organization starts to output top - notch talents to the industry in batches, while enjoying its high reputation, it often faces a structural failure. This failure does not stem from strategic myopia, but rather from the overly dazzling elitism it has created. The existing business framework can no longer contain the ambitions of these people.

From a sign in Zhongguancun, Beijing in 1998, to the obsessive engineer culture in DJI's office building in Nanshan District, Shenzhen, and to the systematic departure of ByteDance's Seed team today, the three Whampoa Military Academy models correspond to three fates in China's science and technology history.

But the thread running through them has never changed: How do talents come, how do they leave, and what do they take away after leaving?

1. The First Batch of Seeds in Utopia

In 1998, Li Kaifu hung a sign at Sigma Building on Zhichun Road, Haidian, Beijing - the Microsoft Research China.

In an era when people still relied on the "squeaking" sound of telephone lines to access the Internet and most people used 56K modems, no one knew what this sign meant, let alone that it would become the first real - sense Whampoa Military Academy for China's AI industry later.

In 2001, after the Microsoft Research China was upgraded to the Microsoft Research Asia (MSRA), it marked the beginning of the enlightenment era of China's AI.

At that time, Sigma Building had an almost sacred academic atmosphere. Li Kaifu, Zhang Yaqin, Shen Xiangyang and other returned elites brought back not only codes but also an elite sense of mission to change the world with technology.

Bill Gates' original intention was simple: find the smartest people and conduct the most cutting - edge research.

Each name, such as Li Kaifu, Zhang Yaqin, Shen Xiangyang, Zhang Hongjiang, and Wang Jian, later became a microcosm of an era. Moreover, a long list of people who changed the industry emerged from this building. Tang Xiao'ou founded SenseTime, Yu Kai founded Horizon Robotics, and Zhou Jian, who once worked at Microsoft China, founded Ubtech Robotics.

It's no exaggeration to say that when you open the early financing plans of China's AI industry, half of the resumes bear the MSRA logo.

The Microsoft Research Asia was an outlier at that time. It didn't care about output but only about whether the research was cutting - edge. It allowed every technologist to waste time in the laboratory to explore topics ten years later.

In that rough - and - tumble era of pursuing quick monetization, this experience of being allowed to conduct luxurious research was a unique utopia in China's science and technology circle.

But utopia can't keep people.

Li Kaifu went to Google and founded Sinovation Ventures. Wang Jian went south to Alibaba and created the Apsara system. Zhang Yaqin joined Baidu. When they left, it was the golden decade of China's Internet's transformation from replication to original creation.

This is an overflow of idealism. These people, with their belief in basic sciences, became the underlying technology and talent pool for the most crucial decade of China's Internet.

However, ten years later, the story in Shenzhen was much more intense.

2. Entrepreneurs Emerging from the Battlefield

DJI is not a research institute but a battlefield.

If MSRA is an ivory tower, then DJI is a high - pressure workshop. DJI's founder, Wang Tao, built an engineering culture in an almost obsessive way. In DJI's office building, product reviews are like trials.

From the wiring of circuit boards to the seams of the shell, the tolerance is close to zero. This pressure - cooker culture, although causing frictions in internal collaboration, has also refined the most scarce engineering practice ability in the industry.

Wang Tao once expressed a similar concern: Once competitors make money, they will compete with DJI for talents, which will be the biggest trouble.

Later, his words came true.

According to statistics from multiple media outlets, more than 20 core DJI employees have left to start their own businesses in recent years. Tao Ye took away the technical backbones of the Mavic department and founded Tofu Technology. Wang Lei founded Bluetti.

These DJI - affiliated companies have a common feature. They not only took away technology but also the gene of saturated R & D from DJI. In the red ocean of hardware entrepreneurship, they reshaped industry standards with a way of dimensionality - reduction strike.

Why do people from DJI have such a high success rate in starting businesses?

Behind it is an almost obsessive product worldview derived from DJI.

Engineers who have been pushed to the limit at DJI find any other track too slow after leaving. In areas where others think "good enough" is okay, they insist on using their ultimate engineering ability to drive their opponents into a corner. Tofu's 3D printing and Bluetti's mobile energy storage follow this path.

This is the violent aesthetic asset that DJI has left for the entire hardware circle: Not only the people but also a cruel methodology of making products so excellent that opponents have nothing to say.

Of course, DJI has also paid a price. This culture of extreme refinement, while achieving excellent products, has also generated a huge centrifugal force. The reverence for Wang Tao and the exhaustion from the system are intertwined. After leaving, these people often become the most headache - inducing challengers for DJI. (Extended reading: Arrogant Wang Tao Learned to Bow His Head in Eight Years)

Different from the gentle talent flow at the Microsoft Research Asia, this is a real tear.

But what's even more intense is yet to come.

3. Systematic Talent Migration

ByteDance's story is another form.

Since its inception, ByteDance's Seed team has had the meaning of a special forces within a large company. The team reports directly to the group management. In order to get the trump card in this AGI era, ByteDance is even willing to break its efficiency rules that it relies on for survival.

According to a report by LatePost, ByteDance has spared no expense for this team. It has a self - owned data team of nearly a thousand people, and the assessment period for the Seed Edge, which specializes in long - term projects, is as long as three years. This strategic patience is an outlier in ByteDance's culture, which advocates cutting off projects that don't show positive results in two months. However, it has become the most desirable haven for AI geniuses in the entire industry.

To build this organization, ByteDance poached Jiang Lu, an expert in video generation from Google, and Zhou Chang, the former head of Alibaba's Tongyi Qianwen. With such a high density of talents, even Yan Junjie, the founder of MiniMax, admitted: ByteDance has the highest density of AI talents in the industry.

But what really makes Seed different is ByteDance's industrialized gene.

In the era of large models, the most scarce thing is not one or two top - level papers but the ability to implement systems after devouring data and computing power. Since joining the company, ByteDance's engineers have been immersed in an environment where data speaks, results are oriented, and ROI is maximized. When they leave, they take away a set of large - scale actual combat experience verified by industrialization.

Then, talents began to leave one after another.

The industry once called the overflow of ByteDance's AI talents an unprecedented technological migration.

As soon as the window for large - model entrepreneurship opened, the Seed team members began to leave systematically. Not only startups but also established large companies like Tencent and Alibaba have seen the presence of ByteDance talents. Li Yukun, the first employee of DeepSeek, is from ByteDance. In every core department of Yuezhianmian, Zhipu, and Jieyuexingchen, there are almost traces of ByteDance.

According to a report by Jiemian News, in the past year, nearly 70 technical talents from the Seed team have left, and more than 30 ByteDance - affiliated AI startups have completed financing. The scale of this talent flow has gone beyond the scope of simple resignations and is more like a spread of the spark for the future of large models.

People left the Microsoft Research Asia one by one, those from DJI left in groups, while those from ByteDance left systematically. The implication behind this is that the competition in the AI era has shifted from individual heroism to systematic and organized positional warfare.

4. An Unsolvable Knot

Putting the three models together, you can see a clear trajectory.

The Microsoft Research Asia represents the natural flow of talents to the industry. DJI was pushed into a new battlefield under high pressure and turmoil, while ByteDance experiences a systematic overflow.

When the imagination of the external market exceeds the internal organizational boundaries, talents will overflow with industrialized power to fill the innovation gaps that large companies can't reach.

The three models have a common ending. Behind this lies a cruel structural law:

A company is qualified to be called a Whampoa Military Academy precisely because it is strong enough, the tasks are difficult enough, and the talent density is high enough.

Ironically, these three cornerstones of success are also the accelerators for talents to leave.

The more difficult the tasks, the more they help people build their reputations; the higher the talent density, the easier it is to incubate startup teams; the more intense the training, the more confident people are to leave.

Being extremely strong becomes a springboard. This seems to be an unsolvable curse for the talent Whampoa Military Academy.

ByteDance has seen and foreseen this. To deal with talent loss, ByteDance has not only launched an independent valuation and repurchase mechanism specifically for large - model talents. The Seed team has also started to try a strategy of investing heavily in young people, directly recruiting AI - native fresh graduates and having Wu Yonghui and Zhou Chang lead them personally.

This is a cruel replacement. Since it can't keep senior experts, it tries to train fresh - graduate geniuses into new combat forces through extreme process - standardization.

This solution may be the most realistic and cold - blooded move for large companies to combat talent overflow at present.

5. Each Generation of Whampoa Military Academy Answers the Same Question

The formation of each Whampoa Military Academy essentially shows: What kind of ability is the most scarce in this era?

In the era of the Microsoft Research Asia, the most scarce thing was the belief in basic research. DJI represents the most scarce extreme engineering ability in hard technology. Behind ByteDance's Seed team is the most scarce large - scale AI implementation system engineering ability.

The emergence of each Whampoa Military Academy is the era saying: There is a kind of ability that we are still in great shortage of. First, gather them, and then spread them out.

Who will be the next Whampoa Military Academy? Obviously, there is no answer yet. But one thing is certain: when it appears, it will definitely fall into that glorious and painful cycle again:

It will become the best place in that era, and yet, it still can't keep the best people.

This is probably the fate of the Whampoa Military Academy.

Words Beyond the Page:

The three different models of Whampoa Military Academies all overflow with technical abilities. But technical ability is just the surface. Deeper down, it is a gene of organizational culture that is spreading with these people.

This gene is like scattered dandelions. The Microsoft Research Asia brought out the belief in doing difficult but right things. DJI brought out the obsession of crushing opponents with products. ByteDance brought out the methodology of disassembling everything with data.

These genes are the real things that are changing the industry.

In a sense, the Whampoa Military Academy has never failed. It just melts into the soil of the entire era in the form of a clone and continues to grow in a different way.

This article is from the WeChat official account "Beyond the Page", author: Huahua. Republished by 36Kr with authorization.