Investors want to persuade AI entrepreneurs to take a rational view of their own valuations.

Three months ago, Deepseek completed its first round of external financing since its establishment, with the total financing amount exceeding 500 billion RMB. The closing of this financing successfully made the total financing amount of China's AI industry in the first half of 2026 exceed that of the whole year of 2025, which once again proves that the financing boom in the AI industry is still continuing and the market sentiment is still at a high point.

However, as funds in the primary market keep pouring into a few popular tracks such as large AI models and embodied intelligence, the decision-making logic of investment institutions has changed from the original "track selection" to "position grabbing". They no longer dwell on technical routes and business models, but care more about whether they can squeeze in to secure a place.

This shift has even reversed the status of capital and projects, and the pricing power of valuation has begun to shift to the hands of start-ups. The valuations of AI startups have been rising steadily.

Jiang Haonan, an investor at GGV Capital, confirmed to *Funtalk* that some teams now have a very high valuation in their first round of financing, which can reach 3 to 5 times that of two or three years ago.

This phenomenon inevitably raises concerns.

In 2021, both the fundraising amount and investment amount of the primary market hit new record highs, which is widely recognized as the "historical peak". Looking back at that period, many AI companies relied on benchmark ranking brushing and concept packaging to obtain extremely high valuations at the peak of market sentiment, but due to reasons such as lack of technical barriers and insufficient product-market fit, most of them no longer exist today.

Some leave while others come in. Before people have time to mourn for those that have "passed away", new projects are pouring in. This cruelty is not unique to the AI track, but a process that all tracks will experience when competition becomes white-hot.

Jeremy Grantham, a legendary Wall Street investor, once asserted that we are facing an AI bubble directly, which may lead to an economic collapse. Ray Dalio, founder of Bridgewater Associates, also agreed with this view in a recent interview, saying that he has seen typical bubble characteristics.

Ray Dalio also mentioned that this does not mean AI has no value, but that the market's expectations for it far exceed its current actual capabilities, and the bubble will burst when expectations return to reality.

Srinivas Rao, a Silicon Valley-based author, expressed a more pessimistic view. A few months ago, he wrote an article titled *99% of AI Startups Will Die Out in 2026*, which went viral on the Internet. It mentioned that the current AI boom is just a replica of the dot-com bubble.

How to face up to the bubble in popular tracks and find projects with the ability to cross cycles in the bubble is a common test for investors.

Of course, compared with two or three years ago, investors now have a clearer judgment on AI projects.

Although the current capital layout is still dominated by large models, world models, and more upstream energy-side solutions, it also pays great attention to the application layer.

Moreover, some very potential early-stage projects are stepping out of the blind spot of capital and starting to get opportunities to talk with investors through social media platforms and other channels.

At the just-concluded 2026 Inclusion·The Bund Summit, media including *Funtalk* communicated with investors from many venture capital institutions about the bubble and valuation, opportunities and risks of AI entrepreneurship.

The guests participating in the interview include: Wang Chao, head of the artificial intelligence team at CAS Star; Yan Qianhang, vice president of Frees Fund; Feng Tian from BAI Capital; Zhao Fuyan, investment director at Hejiu Venture; and Jiang Haonan, investor at GGV Capital.

*Funtalk* has edited the content of this communication without changing the original intention of the conversation. The following is the on-site communication record:

01

Q: It is now widely recognized that there are some bubbles in the US stock market, and the A-share and Hong Kong stock markets have also experienced extremely sharp fluctuations. Will this cause certain difficulties to the valuation and pricing of our primary market? Including how to manage founders' expectations?

Jiang Haonan from GGV Capital: This is actually something that should be communicated more with entrepreneurs. Recently, the valuations of primary market projects have all been rising, and some teams may have a very high valuation in their first round, which could be 3 to 5 times what we saw two or three years ago.

From the perspective of entrepreneurs, this is understandable. They can see so many secondary market counterparts, and in the primary market, "people worse than me have received so much money and such a high valuation", so they will feel that they are also worth that price.

Actually, I still want to advise everyone. Anthropic and Open AI overseas are going to go public one after another, including several leading large model companies in China that are preparing to go public one after another, and SpaceX has already gone public. A large amount of capital will be absorbed by these companies, and I think the secondary market prices of other companies will face great challenges.

If the founder of a primary market project now asks for a very high valuation with a small equity stake, the corresponding funds they can get will be relatively limited. Since primary market investment is not a short-term matter of one or two years, our investment is at least considered on a five-year long-term basis.

Then we have to consider whether the project can cross the cycle — when the market sentiment is relatively high, after raising funds at a very high valuation, can the team deliver real results.

In the 2021 cycle, many companies raised a lot of money at very high valuations and did a lot of PR, but many of them no longer exist now.

So I hope everyone can be more rational, set a reasonable valuation corresponding to each milestone of the company, and raise corresponding funds to support the business development.

When you reach the next milestone, you can get further support.

For the development of the company, this can get more investors to support it, and for internal employees, it can bring more incentives, because they can see the company's growth track, which is very helpful.

Feng Tian from BAI Capital: First of all, about the bubble, I may be relatively optimistic, because I have experienced the live streaming cycle. My own judgment is that bubbles are positively correlated with innovation. Without bubbles, a large amount of capital and talents will not pour in. Every innovative industry is bound to be in this state.

I think the challenge for investors is to judge which companies can still cross the cycle after the bubble bursts.

From the perspective of valuation, I believe the market is still effective.

In fact, at every development stage of enterprises like Open AI, Anthropic, Kimi and Zhipu, everyone will raise a lot of doubts no matter in terms of products or revenue. But when calculating Anthropic's 1000 billion USD ARR, it is only 10 to 20 times of PS, which is relatively reasonable.

Although there are doubts at every stage, it is evolving towards AGI, and its intelligence is improving.

So I think just leave it to the market. If the company can cross the cycle, its valuation will inevitably go up. In the process of innovation, it is unavoidable for investors to pay a certain premium.

Because now both investors and entrepreneurs are on the same starting line, facing something completely unprecedented, but investors need to make decisions quickly, which is quite difficult.

If entrepreneurs believe that their future upper limit is infinite, they can ask for a very high valuation, but they need to judge whether their team, product and project can cross the cycle.

02

Q: There are many OPC (one-person companies) now, and there are many innovation opportunities. How do investment institutions find seed projects?

Yan Qianhang from Frees Fund: We mainly focus on the first and second round of early-stage investment. The trend in the past year is that we spend a lot of time looking for university teachers and professors to persuade them to start their own businesses.

Our style is to first build connections with these university teachers, and then observe them. For example, we will jointly hold some science and innovation activities with universities.

In this process, we will interact with these teachers. After learning that the teachers have the idea of starting a business, we will observe some good seed opportunities and get in touch.

There is also a new change that we will reach out to hackathons on social media. On some platforms that are more suitable for young people to show themselves in the AI field, we communicate with entrepreneurs who have relatively early but not yet perfect ideas, and we will find some good seed opportunities in this process.

I think this is a must-use tool in this era. For example, we will observe and follow up both Xiaohongshu in China and X.com overseas.

Because some founders will post their ideas on these social media to communicate with everyone first, but at this stage they may not necessarily figure out how to start a business. Talking to them at this time may lead to a very good seed opportunity.

Feng Tian from BAI Capital: First, we have a small internal AI Lab that tests all AI products, including models and AI applications. For example, we pull the lists of AI applications every week to see which companies are rising in the rankings, discover interesting products through product testing, and then build connections with their teams.

Second, we read a lot of papers, some papers related to our current focus directions, such as world models and intelligence. We will reach out and talk to the first authors of those interesting papers.

Third, now is an era of personal entrepreneurship, so we can see many interesting products and articles on many social media, and we also find seed projects through this way.

03

Q: Now the investment circle prefers to invest in the cutting-edge upstream, or in the application layer?

Wang Chao from CAS Star: In the ToC field, we invested in an AI fitting mirror company in 2024 and 2025, and also invested in some content platforms. We also invested in a desktop-level CNC company. Now more funds are placed in AI infra, including embodied intelligence, which are some hardware-heavy opportunities.

Yan Qianhang from Frees Fund: Cutting-edge technology is one of our directions, very hard-core technology is another direction, and we also have some consumer-related layouts. From the perspective of investment, technology and consumption are inseparable, because consumption determines demand, and technology determines supply.

Therefore, we will make layouts on cutting-edge technologies earlier. For example, we made very early layouts in solid-state batteries and humanoid robots. In the past two years, we also made layouts in quantum computing, all of which were explored and laid out before the demand picked up.

Because we believe that after the birth of such new technologies, they will bring huge changes to the entire industry or the entire market, which is the cutting-edge direction we focus on.

From the perspective of consumption, the birth of AI only brings a new technology, which ultimately needs to meet various demands, so many new application products will be born.

In the past one or two years, I have discussed the implementation of AI + consumption with my colleagues in the consumer investment team. We have now invested in some companies making AI recording products, as well as AI cameras, which are specially designed for shooting night starry sky, sunrise, sunset, high dynamic range total solar eclipse and lunar eclipse, which requires a lot of AI intervention.

Such products not only need to understand the demand, but also require us to judge the technology. We will follow up both directions synchronously, which does not mean that the institution only prefers one.

If you only focus on cutting-edge technology without understanding the demand, it is easy to have deviations in the prediction of the future application landing of cutting-edge technology, so it is more about the combination and collaboration of the two.

Feng Tian from BAI Capital: My own judgment now is that the model itself is the product, and the capability of the model still represents the capability of the product. The current productization capability is not enough for large-scale application.

We have invested in some model-related companies, such as companies doing expert data labeling or physical materials. In addition, we will also invest in some models and frameworks.

We are also paying attention to the application layer. But in essence, the large language model is an efficiency tool, and it does not have strong network effects yet. When one day it really has network effects and can spread on a large scale, we believe there will be new interaction methods and new innovations, but they have not appeared yet.

Zhao Fuyan from Hejiu Venture: Our core judgment at the current point in time is that the main line on this track is the improvement of intelligence, or consumers' perception of intelligence improvement is weakening.

If the improvement of intelligence is the theme of this track, we will continue to invest in embodied intelligence. We invested in the first embodied project Zibianliang Robotics in December 2023, and then there are world models. Up to now, we see that the improvement of intelligence in the physical world is still the most important theme in the physical world field.

If consumers' perception of intelligence improvement is weakening, theoretically, there should be a product that fits the model capability. If such a product exists, we will invest in it.

Jiang Haonan from GGV Capital: Personally, I think future economic and social resources will tilt to two ends. What we focus on is on the one hand what AI needs, and on the other hand what people need.

On the one hand, centered on AI, sufficient resources are needed to support the development of AI and break through the cutting edge. For example, AI training requires more chips, chips require more interconnection, and a large amount of heat will be generated in the interconnection process, which requires heat dissipation and liquid cooling. We are continuously paying attention to this side, and we have also invested in quantum computing and energy-side solutions.

On the other hand, after AI liberates productivity on a large scale and reconstructs production relations, we focus on what else people need.

We think further, not only focusing on how AI improves efficiency for people, such as GPT-6 Astra solving mathematical problems that have not been proven in recent years. We are thinking that after the emergence of intelligence, people no longer need to participate, and which links people can enjoy.

For example, through AI hardware, people can deepen their self-understanding, understand their own health status by wearing smart hardware, understand what they express more importantly through voice recorders based on past conversations, and there are also scenarios such as socializing and entertainment.

When AI solves 70% to 80% of the problems in society for people, where will people spend their time and money, these trends will have greater value in applications.

04

Q: More and more post-2000s entrepreneurs are emerging. For this group, are the risks greater than opportunities? What kind of young entrepreneurs will young investors pay for?

Zhao Fuyan from Hejiu Venture: The AI entrepreneurship of post-2000s or post-2005s has formed widespread investment popularity and phenomenon, which actually comes from the fact that Transformer has brought large models.

The paper was published in 2017, and GPT 3.0 was released in 2020. Counting back from this time point, the people who truly followed this technical paradigm and researched it from 0 to 1 theoretically entered the AI industry between 2017 and 2021.

People who started their undergraduate studies at this time are probably born between 1997 and 2002. They have absolutely no path dependence, because they started learning what Transformer is from the very beginning of their academic career. That's why we are willing to invest in such post-2000s and post-2005s in some fields.

Theoretically, investment is to find the people in this world who are most familiar with this track, have done in-depth research, and have good capabilities of training models.

And these people are exactly trained in schools before AI became popular, and schools have no requirements for their commercial output, which is the core reason for investing in young people.

This article is from the WeChat official account "Funtalk" (ID: iFuntalker), written by the Funtalk team, and authorized for release by 36Kr.