The trend of "quick withdrawal" is prevailing, and AI investment has entered an era of "fast in, fast out".

In the current AI race, the strategy of "exiting before the next round" seems to be a "product of the times" shaped by the interests of multiple stakeholders. It represents a rational evolution of the market under high uncertainty.

If you've been following the capital market recently, you must have noticed a record - breaking Series A financing deal in the embodied AI field. We won't mention which company it is for now. Large - scale capital transactions in the embodied AI or AI - related fields are nothing new because capital has been pouring in aggressively.

However, a detail about this company caught my attention.

As a star project, it has completed multiple rounds of financing in the past two years. In the recent round, there was a spectacular scene where several first - tier institutions participated in a group investment. But just a year ago, we saw that some of the project's early investors withdrew less than a year after their investment. The fastest one only held the shares for three months.

Later, I talked to several investors about this phenomenon. They told me that this is not an isolated case but a common trend.

As far as I know, even among the six most popular large - model startups in the past two years, some projects couldn't avoid the fate of early investors' rapid exit. Whether it's the large - model field or the embodied AI field, in the current environment of strong consensus, this seems to be the go - to solution for investors.

So why are investors in such a hurry in the AI field, which is hailed as the "golden track for the next decade"?

Exiting before the next round has become the new normal in AI investment

This phenomenon of investment institutions "exiting before the next round" is not only common in China but also overseas.

The difference is that mergers and acquisitions are more aggressive in the overseas market, and the "big fish eating small fish" phenomenon is more prominent. So some projects are more passive and exit through mergers and acquisitions, such as well - known projects like Character.AI and Inflection AI.

Take Character.AI as an example. After it completed a $150 million Series A financing led by a16z in 2023, early seed - round investors such as Sequoia Capital did not participate in this round of financing. Instead, they exited when Google acquired the company for $2.7 billion in 2024, ultimately achieving a return of over 8 times.

Let's clarify first. "Exiting before the next round" means that after a project completes a new round of financing, the investment institution chooses to exit before the next round, usually through methods such as secondary share transfers, mergers and acquisitions, or transactions on the S - fund (secondary market).

This investment strategy is fundamentally different from the traditional VC model. In the traditional VC logic, the core of early - stage investment is "investing in people + investing in the track", and the weight of the business model is relatively low. VCs hope to accompany the founding team and the company's long - term growth and obtain high returns through the company's value leap from the early stage to the mature stage.

For example, during the Internet era, after Sequoia Capital invested in ByteDance, it participated in multiple consecutive rounds of financing, accompanying it from a startup to a global giant and ultimately achieving a return of over a thousand times. When IDG Capital invested in Tencent, it also accompanied the company for a decade and finally reaped huge profits.

In the current AI race, the strategy of "exiting before the next round" seems to be a "product of the times" shaped by the interests of multiple stakeholders. It represents a rational evolution of the market under high uncertainty.

In the past year, we've heard more than one investor say that the consensus in the primary market is too strong and lacks innovation. Everyone knows that AI is the future, but no one can clearly define the specific technological path, clear business model, or profit cycle.

Although people are eagerly looking forward to a flourishing era of Internet applications with various innovations, they all know that such a moment for AI is still far away. Currently, most AI startups are still stuck in the shallow - level innovation of "fine - tuning large models + scenario adaptation", and there are few projects that can break through technological barriers and reshape industry logic.

Regardless of the market situation, work still needs to be done, and money still needs to be invested. The "exit before the next round" strategy can solve two core problems: one is to maintain capital liquidity and avoid long - term capital lock - in in a single project; the other is to quickly recover the principal and obtain short - term returns, locking in profits in an uncertain market.

Moreover, it's not easy to implement the "exit before the next round" strategy in the current situation where the AI technological path is unclear and the business model is still being explored. It requires the project to be hot enough to attract subsequent investors or for industrial capital to be willing to acquire it.

We conducted an incomplete statistics of the financing situation in the domestic AI race from September to mid - October this year. In less than two months, there were more than 200 financing deals in the AI - related fields. More than half of them were early - stage projects before Series B, and more than a quarter were angel - round transactions. These data clearly show that more and more institutions are shifting their funds to early - stage projects. "Investing in early - stage and small - scale projects" is no longer just a slogan but a concrete action.

This also explains why some institutions exited after only three months at the beginning of this article. On the one hand, they may not have accurately predicted the project's development. On the other hand, in the current trend of investing earlier, it's not easy to wait for the next - round investors to take over.

From an investment strategy perspective, using a small amount of capital to invest in multiple early - stage AI projects as early as possible is a cost - effective choice: If the project grows well, it can become a star investment; otherwise, exiting early won't cause significant losses. This "scatter - wide - and - exit - fast" logic is replacing the traditional VC approach of heavy - betting and long - term companionship.

An inevitable choice under the valuation bubble

Of course, the deeper reason lies in the current valuation bubble in the track and the unclear exit path.

The most obvious is the irrational inflation of the valuation bubble.

A large amount of capital pouring into the AI race has directly intensified the Matthew effect in the industry. Especially in the large - model and embodied - intelligence fields, the valuations of leading star projects can soar dozens of times within half a year or a year.

For example, a domestic embodied - intelligence robot company had a valuation of only 200 million yuan in the angel round in 2023, but its valuation soared to 3 billion yuan in the Series A round in 2024, a 15 - fold increase in one year. An overseas AI data annotation company had a valuation of 800 million US dollars in the Series A round in 2023, and its valuation directly exceeded 5 billion US dollars in the Series B round in 2024, with the valuation growth rate far exceeding the business growth rate.

For investment institutions, this short - term skyrocketing of valuation is the only certain opportunity in the current AI race. So instead of waiting for a long and uncertain IPO, it's better to take the profits and run.

After all, the development histories of multiple hot tracks have shown us that IPO no longer guarantees high returns. The "exit before the next round" strategy can directly give a good account to LPs. Most VC funds have a duration of 7 - 10 years, and they need to regularly show returns to LPs during this period.

Zhu Xiaohu, who is good at "calculating" in the investment circle, provided a typical case for the industry: At the peak of the embodied - intelligence track in early 2025, the GSR Ventures he managed exited three robot projects in batches. These projects were all invested at a low level in 2023 and cashed out through secondary share transfers at the peak of the track's popularity, achieving low - level investment and high - level cashing out.

This operation not only allowed the fund to present a good report card to LPs but also laid the foundation for subsequent fundraising.

In addition, the pressure of valuation correction risk also makes early exit an effective means. Currently, most AI companies have high valuations, but most projects don't have stable revenues, let alone profits. Their valuations are completely supported by technological imagination. However, as the market cools down, this over - valued logic that overdraws the future will inevitably face correction pressure.

Not to mention the long - term situation, this can be clearly felt from the capital situation of domestic large - model startups. In the second half of 2024, due to the slow progress of commercialization, the valuations of some companies corrected by 30% - 50%, and there were even cases of "valuation inversion in the next round of financing". This over - valued situation will only become a stumbling block for the company's further development, and there are few investors willing to take over now. Early - stage institutions can only choose to wait.

Speaking of startups, the "exit before the next round" strategy is not just the one - sided choice of institutions. For startup teams, the introduction of capital is not just about the money, especially in such a highly competitive track.

From the startup projects I've contacted this year, many companies are not raising funds for cash flow but simply want to "take sides" and obtain industrial resources.

I remember a startup founder who just completed an angel round at the beginning of the year said that although she contacted many investment institutions, she preferred to find capital that could help with business growth. This is because after the equalization of AI technology, the startup cost has decreased, and the key to this competition is who can quickly occupy the market and users' minds.

Therefore, the incremental value that pure financial institutions can bring is quite limited. Of course, to balance the company's equity distribution, the founder still chose to allocate some shares to financial institutions in that round. But these shares are more like "transitional equity", and early - stage financial institutions usually choose to exit when subsequent industrial capital enters.

A double - edged sword

Undoubtedly, this choice is a kind of tacit two - way pursuit for entrepreneurs and investment institutions. However, on the other side of the coin, it is accelerating the impetuosity in the primary market.

We don't deny that due to multiple factors such as capital, policies, and trade relations, venture capital is deviating further and further from the traditional path. But "patient capital" has always been an irreplaceable and necessary part of the market.

Many people like to look back at the glory days of the Internet era. At that time, our hard - tech industries were in a difficult situation. Similarly, it was the persistence of those "patient capitals" that made us less passive in the subsequent wave of domestic substitution.

Now, the rise of Deepseek has fired the first shot of domestic independent AI. Then, many people analyzed why they didn't invest in it. Later, a fact emerged - the uncertainty was too high, and the investment required was too large. Even if it opened up early - stage financing, the result might not have been good, except for its own team.

The rise of the "exit before the next round" strategy and the absence of patient capital are undermining the entire startup ecosystem.

First, it changes the growth rhythm of startups. To successfully complete round - after - round financing, many companies may be rapidly "forced to mature", which is particularly obvious in the large - model field. Some startups pursue superficial indicators such as the number of published papers and the scale of model parameters to make their financing data look good, while ignoring the actual application scenarios and business model verification.

When the main focus is not on the business but on finding financing, the capital pressure may force the company to pursue short - term data and ignore the construction of long - term technological barriers.

Second, the early exit of early - stage investment institutions often sends a negative signal to the market, causing self - doubt among the team. Even if the company has a good fundamental situation, this exit behavior itself will be interpreted by subsequent investors as a "signal of lack of confidence", and this psychological impact is often more profound than the actual capital impact.

What's more worrying is that the "exit before the next round" phenomenon will continue before there is a disruptive technological breakthrough and the business model is fully established.

Some analysts point out that if this trend continues for a long time, projects that ride on the wave and raise funds just for the sake of it will quickly occupy the market because their goals are the same as those of institutions, which is to achieve quick success. These projects are often good at packaging concepts and creating a stir but have obvious shortcomings in core technology accumulation and product implementation. Naturally, it will be difficult for truly valuable and problem - solving great companies to survive.

From a more macro perspective, the prevalence of the "exit before the next round" strategy reflects the short - termism tendency of the venture - capital industry in the strategic emerging AI industry.

Different from the Internet era, AI technology has a longer R & D cycle, requires more capital, and has a higher technological threshold. This exactly requires more strategically patient capital support.

However, the current investment environment shows the opposite trend - capital is more eager for quick turnover and short - term returns.

Specifically within investment institutions, this trend has also triggered new changes. Many VC institutions are adjusting their internal incentive mechanisms, including the investment - return cycle in the assessment indicators. This makes investment managers more inclined to recommend projects that can be exited quickly.

At the same time, the fund structure is also changing, with more special - purpose funds with shorter durations emerging, specifically targeting these "quick - in - and - quick - out" investment opportunities. These changes further strengthen the investment culture of "exiting before the next round".

So an age - old question is now in front of investment institutions: How to balance the long - term companionship with truly great companies while pursuing financial returns? The answer to this question may be redefining the value of venture capital in the AI era.

This article is from the WeChat official account "Intelligent Reconstruction". Author: Lijuan. Republished by 36Kr with permission.