In the second half of 2025, discussions about the AI bubble evolved from a fringe topic to a market focus. Among the top 10 companies by market capitalization in the U.S. stock market, 8 are AI-related tech companies, indicating a much higher concentration of AI than during the internet bubble.
Capital flows have also shown extreme characteristics. On one hand, tech giants have initiated an &#34;arms race,&#34; while on the other hand, institutional investors are cashing out. For a time, the &#34;bubble theory&#34; has been rampant.
On December 17, XPeng Chairman and CEO He Xiaopeng shared some of his thoughts after a business trip to the U.S. on social media, discussing his views on robots, the AI bubble, and more.
Regarding the AI bubble, He Xiaopeng believes that any technological era has phases or segments with bubbles, whether it is the internet era or the new energy vehicle era, but this is an inevitable competitive process from chaos to order in the market.
He Xiaopeng stated: &#34;I believe AI will inevitably drive tremendous changes in society. Today, we are still in the very early stage from 0 to 0.1. If we must talk about bubbles, it seems that valuations in China are relatively reasonable, while those in the U.S. are indeed a bit high; China focuses more on market applications, while the U.S. emphasizes cutting-edge research. Overall, I believe there is currently no AI bubble, and the future market for AI holds enormous opportunities.&#34;
Additionally, during his business trip, He Xiaopeng observed: &#34;The U.S. has a lot of entrepreneurship in AI, biotechnology, and finance. In terms of AI entrepreneurship, particularly from the perspective of Silicon Valley, there are many startups in the SaaS and physical AI robotics fields, with particularly high valuations. Half of my discussions this time were with friends who are starting robotics companies, which are very concentrated.&#34;
He also compared the differences in robotics entrepreneurship between China and the U.S. &#34;Many Chinese robotics companies start from joints and control, while many American robotics companies start from models. I believe humanoid robots will be a competition among giants, while different specialized robots will have many players in various fields, and there will be numerous opportunities for success.&#34;
In the field of physical AI, he believes, &#34;In the next three years, the most likely breakthroughs may not necessarily be in the digital world (it is evident that OpenAI, while focusing on AGI, is fully committed to business implementation. Of course, we look forward to the next Transformer or a significant drop in training costs), but rather there may be major changes in the field of physical AI, such as autonomous driving reaching near L4 or complete L4; humanoid robots achieving rapid transitions from L1 to L4 in semi-autonomous driving. These may lead to significant breakthroughs. The development speed of physical AI will be slower than that of digital AI, but its impact on our lives will be greater.&#34;
Regarding AGI, He Xiaopeng also shared his views. In his opinion, today's AI mainly involves imitation learning similar to humans, such as learning knowledge from textbooks and observing how others perform tasks. Adding reinforcement learning to this is akin to repeatedly practicing problems to improve the effectiveness and efficiency of a task as quickly as possible.
He also used autonomous driving as an example, noting that it can quickly learn from the driving habits of millions of people, which is why we see it can drive better and safer than most people in a short time, as well as the emergence of entirely new capabilities, but these do not represent true creativity
In He Xiaopeng's view, &#34;true AGI still requires many capabilities, such as multimodal understanding and world modeling (similar to our understanding of the reasons behind things, or an upgraded version of the left-right combat learning in autonomous driving world models), continuous learning, long-term insight, and planning abilities, which have not yet fully arrived and may still require several years and the enhancement of some underlying capabilities.&#34;
He Xiaopeng's views on the AI bubble are similar to those of many current industry analyses.
A report from UBS Securities shows that the capital expenditure-to-revenue ratio of China's internet giants (10%) is significantly lower than that of their American counterparts at 27%. On the application side, there are high service capability barriers in vertical fields, and leading companies are actively integrating AI functions into their own applications.
Most analysts believe that it is too early to say that the industry is fully in a bubble, and the next 2-3 years will be a critical verification period.
Wellington Management believes that compared to the internet bubble, while the current AI market has shown bubble characteristics, it is still in the early stages, and there is still room for market development. The liquidity tightening and profit realization failures that triggered the collapse of the internet bubble have not yet occurred.
Behind the bubble debate is a significant game of strategy regarding the future of technology and capital rationality.
The key to the future remains the deep integration of AI with the real economy. Only those companies that can penetrate market noise and focus on the essence of technology and commercial value are more likely to win in this technological revolution

OpenAI

Still in the very early stage of 0 to 0.1

He Xiaopeng: There is currently no AI bubble, and the future AI market has huge opportunities