I invested in OpenAI, but I'm extremely confused: I still can't find the AI trillion-dollar unicorn.

Interview | Wang Fang, Founder of Pencil News

Written by | Zhao Songge, Editor of Pencil News

In 2022, Zhang Min from HeLi Capital made an interesting decision: he pooled a "small fund" with overseas friends using their own funds and participated in the investment of several large - model companies such as OpenAI and Anthropic.

However, watching the upsurge of large models around the world, this veteran who has experienced the Internet bubble and witnessed the rise of mobile Internet is now in unprecedented confusion. "I'm certain that the storm of AI changing the world is inevitable, but I can't find the key to sail through the storm."

This is very similar to the shock he felt at the Yahoo headquarters in 1998. At that time, he was working at the oil giant Shell. He couldn't understand why, without any assets or resources, a few young people with a few servers dared to claim to "conquer the world".

Now, facing large models, he is once again standing at a historical crossroads. "I have two confusions. One is what exactly will large models rely on to change the world in the end? Is it the large models themselves? The other is: where on earth are the trillion - dollar unicorns in AI applications?"

In response to these two "confusions", in the conversation with Pencil News, he candidly shared his thoughts in the field of AI investment.

1. Identifying trends and vague paths.

Almost all investors are convinced that AI will disrupt the world (including Zhang Min), but currently, most people can't judge which specific types of companies (including large - model companies) will become the ultimate winners.

2. The core motivation for investing in large models is "fear of missing out".

Even though he participated in investing a small amount in leading projects like OpenAI and Anthropic, Zhang Min believes that the essence of this investment is out of "fear of missing out", rather than a firm judgment on the technological end - game. Whether large models can form a monopoly ecosystem like iOS remains unknown.

3. Where is the trillion - dollar ecosystem of large models?

Back then, for the trillion - dollar companies in the Internet industry, there were several key points, such as hardware, operating systems, payment, and super apps. Perhaps in the future, there will be an AI super - platform that combines these elements, but what's the path?

4. The real AI super - native applications haven't emerged yet (large models themselves don't count).

Currently, 90% of AI projects are just efficiency tools with "AI added" (such as intelligent customer service). The real native applications at the level of "ride - hailing apps" haven't emerged yet. What's more cruel is that in the future, such applications may be crushed by traditional enterprises with capital and traffic.

5. AI entrepreneurs will face fierce encirclement and suppression from traditional enterprises such as large companies.

Different from the Internet era, traditional enterprises back then were "watching from the sidelines". Now, the so - called "traditional enterprises" are the Internet and mobile Internet companies from the past. They are still in their prime, with a strong acceptance and application ability of AI. They have money, technology, and traffic. Startup teams will face huge challenges.

But confusion is the normal state of early - stage investors. Innovative things often grow in places where people can't see, can't understand, look down upon, and don't have time to notice. Just like he once missed the investment opportunity in Pop Mart.

When Wang Ning, the founder of Pop Mart, was seeking financing for the first time, the first group of investors he met included Zhang Min. "Now every time my two daughters buy blind boxes, they ask: Dad, why didn't you invest in Pop Mart?" Whenever Zhang Min mentions this, he always forces a smile.

The AI era may be the best opportunity for investors of his generation. Zhang Min doesn't want to miss it this time.

Statement: The interviewee has confirmed that the information in the article is true and accurate. Pencil News is willing to endorse the content.

Confusion 1: Who exactly will AI rely on to change the world? Large models?

Pencil News: Hello, Mr. Zhang. Long time no see. You said you're quite confused now. May I ask what exactly you're confused about?

Zhang Min: Sure. I think it's not just me. Actually, every early - stage investor, especially angel investors, should be quite confused now.

I've been doing early - stage angel investment for many years. I started in 2002 and have experienced the entire Internet cycle.

Zhang Min, Managing Partner of HeLi Capital, has previously worked at Shell International, PwC, and Morningside Venture Capital. He founded HeLi Capital in 2012. Representative investment projects include Debon Logistics, Moji Weather, Xi'an Rongxin, and JD Industrial Products, etc.

We always use past historical experience to view things. It's human nature. What people are best at and most often use is historical experience. But we also know that there are two sayings to remember: First, history always rhymes. Second, history doesn't repeat itself simply.

Let's think about the Internet back then and this wave of artificial intelligence. There was also anxiety and confusion in the Internet era. We've all been through it. But the anxiety then was different from what it is now.

In 1999 and 2000, the Internet first went crazy, then crashed, and then restarted and gradually became one of the greatest revolutions in our memory. For investors, it was a revolution, right?

What were its characteristics? Now when I look back, I started to get in touch with the Internet in 1997. At that time, I was studying and working abroad and was sent to Yahoo to learn. I went on behalf of the Shell headquarters. You know that Shell is still a very great and giant company today, relying on assets and resources.

But when I saw Yahoo, I was stunned. Because it had almost no assets, just a few people. The total assets of the company were not even as valuable as one of our oil fields. They wanted to change the world. For someone like me who grew up in the traditional economic system, it was a huge shock.

After I came back, the company organized a group of the youngest people to study the Internet. Everyone was shocked. Everyone was thinking: This Internet is so amazing. It might change the world. But how? Actually, we were quite confused at that time.

We read a lot of books and had countless internal discussions. At that time, the first thing we could think of was to add a ".com" after the company name. Around 1999, what giant traditional companies around the world could do was to build a website first.

But what to do next, we didn't know. We even had internal arguments: Is it useful to put a product introduction? This is not the Internet thinking at all. We also read KK's books, which had some great judgments, but after reading them at that time, we were still confused. You couldn't even imagine things like e - commerce, mobile payment, WeChat, and Douyin.

We had a very smart colleague. Although he was also confused, he was one of the most far - sighted among us. One day, he came to me and said that he had figured it out. The greatest thing about the Internet in the future would be mobile phones. Just think, it was 1998 and 1999, and mobile phones weren't even popular yet. He had already realized that mobile phones might be the greatest product of the future Internet.

At that time, there were no iPhones, no apps, and no mobile Internet, but he could see this.

I always tell this story to show that history always rhymes but doesn't repeat itself simply. Now, artificial intelligence is very similar to the Internet back then. Almost everyone with a bit of thinking knows that it will change the world.

What I'm confused about is: Before, I didn't know that the Internet would change the world. Now, I clearly know that AI will definitely change the world, but I don't know exactly who will bring about the change and how.

"Who" is very important. Let's imagine: Can large models change every industry and the world? Maybe. But large models can't do everything. What will they become? Currently, I'm really not clear.

I remember that the year before last, I pooled some money with a few people to form a small fund and invested in OpenAI, Anthropic, Cohere, SpaceX, etc. If I had foresight and knew that these companies would definitely be the future Google, I would have invested all my money in them. But I only invested a little because I really don't know which large - model company will finally become the greatest enterprise. Even more, I'm not sure if large - model companies will become great enterprises instead of just simple tools.

Even though large models are so powerful now, as an investor, I'm still confused. Among the first - wave companies that emerged during the early Internet boom, apart from portal websites, many were technology companies doing web pages. Now, we don't even know where they've gone.

How will large - model companies form their architectures in the end, and can they become high - valuation, monopolistic companies? I'm really not sure. I know that great things will grow on top of them in the future, but I'm not confident whether this underlying platform can become a great enterprise.

Confusion 2: Where on earth are the unicorns in AI applications?

Pencil News: Do you have any other confusions?

Zhang Min: Yes, I have a second confusion.

Actually, apart from large models, I'm also confused about "applications".

I really support what Mr. Ming Shun (founder of Global AI) said. I even borrowed one of his PPT slides, which said that the future will definitely be an era of applications, similar to the Internet era.

But the problem is here: How do you know which application or which type of application will succeed? Actually, we're not clear.

Applications can be roughly divided into two categories: one is native applications, and the other is applications with "AI added" - just like adding "Internet" back then.

Except for large models, there are actually very few native AI applications now. Most are efficiency - enhancing applications with "AI added". For example, helping you buy tickets, making travel plans, analyzing legal documents, summarizing chronic diseases - these are actually not native applications.

Then the question arises: Where are the native applications? They will definitely emerge, but we don't know where. By the time we know, as early - stage angels, we basically can't afford to invest. This is the most painful part for us.

Just like the ride - hailing apps back then. They were completely native Internet applications. Without the Internet, they would never have emerged. But by the time we knew about them, it was basically too late to invest. Those who followed blindly all failed.

The second confusion is also very important. When an AI application is an "AI - added" one, who can develop it? Can startup teams defeat current large enterprises?

This brings us back to the first question: traditional enterprises. This wave is different from the past. Back then, traditional enterprises were repellent and watchful towards the Internet. But now, the people in these so - called "traditional industries" are more aggressive than anyone else. Just look at Tencent, ByteDance, etc.

Because they themselves came from the Internet. They are still in their prime, with awareness and innovation ability. They know how to use AI. Moreover, they have money, technology, and traffic.

How can you, an entrepreneur, compete with them?

So, when I talk about these two confusions (who will AI rely on to change the world and where are the opportunities in AI applications) - it's not pessimism but a realistic view.

Pencil News: Regarding the first confusion, at least you've invested in OpenAI, etc. Even if you were confused at that time, at least you had some clear understanding, right?

Zhang Min: Actually, when I invested in OpenAI, Anthropic, etc., it wasn't early (from 2022 - 2023). By that time, it was already very popular in Silicon Valley. It was just a decision that followed the trend. I don't think it was a great decision. In the end, it was because of FOMO (fear of missing out) - I was afraid of missing the opportunity.

If I'm convinced that artificial intelligence is the greatest thing in the future and I don't invest, I'll regret it when I retire.

I don't want to have regrets. Even if I don't make money in the end, at least I won't regret it.

So, the decision to invest in OpenAI was essentially not because of my great foresight but because I was afraid of missing out.

Pencil News: There shouldn't have been many domestic investors who had the determination to invest in OpenAI at that time, right?

Zhang Min: There weren't many at that time. Since last year, many people have come to ask. Some domestic investors even came in groups to ask if I would sell my shares to them.

It was quite easy for me to make this decision at that time, but for many people, it wasn't a "consensus". When I proposed the idea of investing in OpenAI, I shared it with many people and even organized a lecture, but in the end, few people followed me.

Pencil News: Did you invest in domestic large - model companies at that time? Were you also making arrangements?

Zhang Min: I didn't make any arrangements at all. It's not that I think they don't have opportunities. Absolutely not. First, I'd like to state that China will definitely have its own large models. DeepSeek is an example. Although I'm not sure if it will be the final winner, it is indeed a proof. But why didn't we invest in domestic large - model companies?

There are two reasons.

The first reason: It requires a huge amount of money. To put it bluntly, it's "lack of money".

This is the biggest difference between early - stage investment in China and the US. In China, if you want to invest in large models, you need a lot of money because there is no system that allows small - scale capital to participate. But in the US, you can participate with just a few thousand or tens of thousands of dollars.

If you tell a domestic large - model company that you want to invest 500,000 - 600,000 RMB, they simply won't give you a chance.

The second reason is a basic judgment of ours - so far, it's still correct. Maybe we'll be proven wrong in the future, but it's common for early - stage investors to be proven wrong. We judge that domestic large - model companies may ultimately be backed by large companies or state - owned enterprises. For angel or individual investors, this opportunity is too far away.

So far, the ones that have emerged are basically of this background.

Pencil News: Actually, the domestic large - model ecosystem has been developing for 2 - 3 years, and there have been relatively large changes. Are you less confused now?

Zhang Min: On the contrary, I feel more confused. Let's look at the world's greatest models. If we rank the top ten, everyone should know who they are. Maybe there's no consensus on the eighth, ninth, and tenth, but the top five are just in a different order.

Why am I more confused? I really can't be sure if OpenAI, Gemini, and Anthropic can reach a market