Foreign media: At a time when AI ' s business start-up finance continues to warm, it is becoming increasingly difficult for investors to price projects using old methods. TechCrunch recently interviewed M13 co-founder Carter Reum and Basis Set Ventures partner Chang Xu at an event in Los Angeles. Both agree that the AI market is still at an early stage, but that valuation, competition and product overlay have been significantly faster than in the previous technology cycle.

It's growing fast. It's harder to judge.

Chang Xu believes that the current AI market is like a bubble and cannot simply be classified as a bubble. This is due to the unprecedented growth in income of some companies. She cited, for example, the fact that some AI companies were able to raise recurrent income from millions of dollars to tens of millions of dollars in a very short time, making the traditional valuation framework inadequate.

She also stated, however, that if all projects were priced on the most optimistic growth curve, the portfolio would be difficult to establish in the long term. In other words, markets do have high valuations, but there is real growth behind them.

Carter Reum is more cautious. In his view, the AI cycle is similar to historical technological waves, such as cloud computing, smartphones, but this time the difference is that start-up companies not only compete with each other, but also face large technology companies with more capital, data, and talent, such as OpenAI, Google.

Increased focus on infrastructure and regulatory industries

In the selection of projects, the common denominator is not to focus on short-term income growth, but to focus more on the ability of enterprises to create long-term barriers.

Xu indicates that Basis Set Ventures currently divides opportunities into “under-AI” and “over-AI”. The former are mainly infrastructure, including databases, version control and deployment tools. In her view, the tools were originally designed for human developers, but as AI agents began to participate in the development and implementation processes, the bottom-up tools could also be reworked.

She mentioned that a number of teams had emerged in the market over the past year in an attempt to create new development collaboration tools for AI agents. This means that infrastructure levels are still evolving rapidly.

Reum put more emphasis on “the friction itself is a moat”. He indicated that the team would give priority attention to regulated industries, which were at a higher threshold, and that large model companies might not invest sufficient resources in the short term, even if they entered in the future. This makes it easier to form medium-to-large exit opportunities for medical and public services.

Second wave of opportunity or two or four years later

The answer to how start-up companies avoid being squeezed by OpenAI, Anthropic or Google points to the same: not only pursue the most crowded common scenes.

Reum believed that, in the past, entrepreneurs had also seen early signs of large companies entering a particular track, but that change could occur in a very short time. He suggested that the founders should keep an eye on what was going to be done and what was going to be done by big companies.

Xu distinguishes opportunities by “speed market” and “deep market”. In her view, the pace of implementation determined the winning track and later replicating would be faster; in areas requiring long-term research and development and complex processes, such as manufacturing, biotechnology and so forth, the challenges remained difficult, and such markets were better suited to building long-term advantages.

She also mentioned that truly interesting projects often did not look like mature businesses at an early stage. Many of the later AI companies were initially experimenting around generating images, videos or creative tools until business models were found.

According to both investors, the second and third waves of two to four years after the current first wave of the first wave of AI, which is the most crowded, may be of greater interest. As model capacity continues to spill, new industrial applications and infrastructure needs will continue to emerge.