When Nathan Lambert finished his visit to several top AI laboratories in China and took the high-speed train from Hangzhou to Shanghai, the fields, high-rise buildings, and wind turbines on the ridges outside the window were like a microcosm of China's AI development - both solid foundations and vibrant vitality. The researcher returned with humility and a new understanding, finally realizing that the rapid development of AI in China does not lie in distant industry debates, but in the daily operations of every laboratory.

During the visit, Lambert's biggest discovery was that the AI ecosystem in China and the United States are not simply "catching up and being caught up", but have formed two completely different development paths. The development of AI in the United States relies more on capital investment, breakthroughs in original paradigms, and the personal influence of top scientists, like a cutting-edge competition led by capital and celebrity laboratories; Chinese AI, on the other hand, relies on its strong engineering optimization capabilities, thriving open source ecosystem, technological self-control awareness, and the investment of a large number of young researchers to quickly catch up in existing directions, pushing its model capabilities to the forefront of the world. It is more like an execution competition focused on industrial landing.

The core difference of this competition lies not in surface factors such as model size, data, or computing power - the seemingly similar investment in top laboratories between China and the United States, but the real gap lies in the organization and cultural genes of these factors.

###1、 Laboratory culture: Humility and pragmatism, replacing "individual sharpness" with "collective excellence"

The core competitiveness of Chinese AI laboratories first stems from a unique organizational culture. Building a top-notch big language model is essentially a meticulous engineering process that runs through the entire technology stack - from data annotation, architecture optimization to reinforcement learning algorithm implementation. Every small improvement in each link requires complex integration to achieve overall optimization. And the culture of Chinese laboratories perfectly fits the needs of this' systematic engineering '.

Unlike American laboratories, Chinese researchers are more willing to put aside their personal strengths and devote themselves to less glamorous but crucial foundational work. Lambert found that the trait of "speaking up for oneself" in American culture can easily influence the self-awareness and promotion aspirations of top scientists, affecting the overall optimization of the model; In Chinese laboratories, researchers tend to set aside personal ideas and maximize multi-objective optimization in order to achieve the final model effect.

This cultural difference is also reflected in the personnel composition of the laboratory. A large proportion of the core contributors to China's top AI laboratories are students who are treated as peers and directly integrated into the core team, without the barrier of "interns being isolated from core work" in top American laboratories. These young researchers bring fresh perspectives, are not bound by past AI hype cycles, can quickly absorb cutting-edge technologies, and are willing to invest in basic work with a humble attitude, just to obtain opportunities to improve models.

More noteworthy is that Chinese researchers have a highly targeted focus. When asked about philosophical issues such as the economic impact and long-term social risks of AI, most of them express confusion - for them, the core role is to "build the best models", and those complex off field discussions are more like a "category error". As a researcher quoted Dan Wang's judgment: compared to the lawyer led United States, China is governed by engineers, and "construction" itself is their core demand.

###2、 Talent ecology: sufficient quantity, open source collaboration rather than "tribal competition"

The rapid development of AI in China cannot be separated from sufficient talent supply and unique ecological atmosphere. Similar to the United States, China has also seen a trend of AI talent flowing from academia to industry - a researcher who originally planned to become a professor admitted that the popularization of big language models made him feel that "education has been solved" and ultimately chose to enter the industry.

These young talents who have flooded into the industry have formed a powerful engineering force. They are accustomed to quickly absorbing massive literature and internal technology stacks, and are skilled at solving problems that have already been validated, which perfectly fits the current development needs of AI for "rapid iteration and continuous optimization". And the sufficient talent base also enables Chinese laboratories to quickly promote technology implementation and form scale advantages in engineering optimization.

What's even more unique is the collaborative atmosphere within the Chinese AI community. In Beijing, the top laboratories are close to each other, resembling the Bay Area in the United States, but the ecological atmosphere is completely different: Chinese laboratories are more like a symbiotic ecosystem rather than warring tribes. In the private communication, researchers mostly expressed their respect for their peers - everyone feared ByteDance (the only cutting-edge closed source laboratory), and generally respected DeepSeek's research taste; In the United States, competition between laboratories is often more intense.

This sense of collaboration is also reflected in the development of the open source ecosystem. Chinese laboratories generally adhere to a pragmatic attitude of "openness first": open models not only receive strong feedback from the developer community and give back to the ecosystem, but also empower their own products - many companies will fine tune based on open source models, retain customized versions for internal products, and form a virtuous cycle of "open source feedback internal optimization".

###3、 Industry logic: Technology self-control as the core, pragmatic response to needs and constraints

During the visit, Lambert also found that the development of AI laboratories in China has always revolved around two core areas: "technological self-control" and "industrial landing", forming a completely different industrial logic from the West.

On the one hand, Chinese enterprises generally have a strong "technology ownership mentality". Not only ByteDance, Alibaba and other technology giants are building general big language models, but also Meituan, Xiaomi, Ant Group and other platform and consumer enterprises have joined in one after another - they are not trying to "rub hot spots", but they regard big language models as the core foundation of future products, hoping to grasp the initiative of development by controlling the technology stack. This mentality is highly compatible with China's extensive "construction culture".

On the other hand, Chinese laboratories have demonstrated strong practicality in demand assessment and resource constraints. Lambert found that the question of "Chinese companies unwilling to pay for software and unable to support the AI inference market" is not valid - Chinese AI spending is closer to the fundamental cloud market rather than the smaller SaaS market, and laboratories generally do not worry about the growth potential of the AI tool market. Although Claude is nominally banned in China, most Chinese developers are still deeply influenced by it. Practical technicians are more willing to invest in valuable tools, and this motivation far exceeds the historical habit of "not spending money to buy software".

In terms of resource constraints, Chinese laboratories have also taken their own path: although the data industry is not as developed as that in the West, laboratories have chosen to build their own internal data and training environment, and large enterprises such as ByteDance and Alibaba even have internal data annotation teams; Although there is a strong desire for Nvidia's computing power, domestic accelerators such as Huawei have received positive reviews in the field of inference and have become important supplements.

In addition, government support does exist, but it is mostly a partial assistance of decentralization - for example, different neighborhoods in Beijing simplify bureaucratic processes and remove licensing barriers to attract technology companies, and there is no sign of high-level government intervention in model technology decision-making. The development of laboratories is still mainly based on market and technological logic.

###4、 Conclusion: It's not about copying Silicon Valley, but about forging our own path in AI

Lambert admitted that this trip to China has completely broken the stereotype of Chinese AI - the rapid development of Chinese AI is not simply copying the Silicon Valley model, but based on its own culture, talent, and industry needs, it has embarked on a unique path. It does not pursue a breakthrough in the original paradigm of "from 0 to 1", but can quickly catch up with the forefront with the advantages of "from 1 to 100" engineering optimization, open source collaboration, and talent; It does not advocate the personal influence of celebrity scientists, but can achieve continuous improvement in model capabilities through collective collaboration.

The future competition of AI is no longer just a competition on model rankings, but also a competition of organizational capabilities, developer ecology, and industry execution. The true progress of Chinese AI does not lie in whether it surpasses the United States, but in whether it no longer passively follows, but participates in the global forefront in its own way - this kind of transformation based on the culture, talent, and industrial logic within the laboratory is the core password for the rapid rise of Chinese AI.

And when Lambert looks back on this visit, the most profound feeling is not only the vigorous vitality of Chinese AI, but also the humanity and sincerity of the researchers - they are far away from the noise of geopolitics and devoted to "building better models". This pure enthusiasm for construction may be another important answer for the sustained and rapid development of AI in China.