It is sometimes difficult for coffee shops in Zhongguancun to make money by selling coffee.

Garage Coffee and 3W Coffee were born here. Ten years ago, the "Haidian Book City Commercial Pedestrian Street" was renamed "Zhongguancun Entrepreneurship Street". Since its establishment in 2011, it has been aiming to "grow China'sMicrosoftThe garage coffee shop where the founders of the “Google” group “went to” has become an important symbol of Internet technology entrepreneurship. More and more startup coffee shops and incubators have gathered nearby, attracting countless young people and entrepreneurs to chat and exchange ideas in the hope of gaining attention and start-up funds. 3W is the most prestigious one among them, with a number of well-known investors and founders and executives of star companies standing behind it.

Most young people and entrepreneurs are stingy. They often bring their own bottle of water and sit there all day. Some even spend the night in the cafe, but the cafe doesn't care about them. It's not just coffee that's sold here. Talent, passion and dreams are the main commodities, and there are always investors with a lot of money willing to pay for them. Zhang Yiming got the first investment in Toutiao with a simple model sketched on a napkin in a cafe.

Later, Internet venture capital declined, and Zhongguancun's entrepreneurial cafes also closed one after another. But in every wave of wealth, a similar place is needed - "Meta Space", which took over the mantle of Garage and 3W and became the busiest distribution center for Web3 and AI entrepreneurs in Beijing.

This cafe near Wudaokou was established in July 2021. It has witnessed the rise of Web3 from frenzy to silence, and the rise of generative AI from popularity to cold reception. In just three years, it has gone through two cycles.

The cafe has left behind many symbols that once excited people: the huge "Bored Ape" sculpture in the glass display cabinet at the door, which is the most famous NFT avatar in the currency circle. Zhu Xiaohu also paid 500,000 US dollars to buy it and once set it as his WeChat avatar; the decorative walls of the cafe are covered with well-known NFT avatars, including Mfers stickman, pixel owl, etc. Behind each avatar is a dream-making story.

● AI product list display screen. Photo: Zhao Yifan

Now, the most eye-catching thing in the cafe is an LED display screen, which lists the monthly rankings of domestic and foreign AI products based on the number of calls. ChatGPT ranks first, but none of the domestic large-model products are on the list.

This seems to be in line with the deserted cafe. On this weekday afternoon, there were only four tables of guests, but in the first year when Meta Space opened, along with the wealth-making wave of Web3, a large number of young people gathered here every day to discuss projects, give presentations, or use this place as an office, making it a lively scene. At present, most domestic AI projects are not well funded, and the market is experiencing a cold winter.

I ordered a cup of coffee and sat down. The name of the coffee was "ALL in AI", and the latte art was all the logo of ChatGPT. A middle-aged man behind me made many phone calls. He should be an investor of some institution. He asked the entrepreneur about the progress of the project and the user demand satisfaction rate of the product. At five o'clock in the evening, he got up and left. He chatted with the boss standing at the door for a few words about his recent situation. When he left, he smiled and said loudly: "I'm all in AI!"

In a startup-themed coffee shop, such scenes have long been commonplace. After experiencing two booms, this coffee shop has also become a window for observing the development from Web3 to AI: What did those who made money from Web3 do when they switched to AI? What are the differences between AI startups and Web3 startups? And what happened to those who went all in on AI?

We talked to the owner of the cafe, Liu Jialiang, who is a witness to the changes in the industry day after day. Through his observations, we can get a glimpse of the passion and ups and downs, as well as the collision of dreams and reality from Web3 to AI.

The following is Liu Jialiang’s account:

High-profile Web3, low-profile AI

The first batch of people who came into contact with Web3 were definitely people from Web2. I belong to this category. I started out as a product manager in a large company. In 2011, I first came into contact with Bitcoin at 3W Coffee. I thought it was interesting, so I bought some. At the time, it was only 1,300 RMB. After leaving the large company, I started working on Web3.

When AI came, the same group of people followed suit. I sometimes joke that it was because they had graphics cards, but in fact, most of the people who came from Web3 to do AI did not make money and wanted to continue to work hard. Even if what came out after Web3 was not AI but BI, it would still be this group of people.

Why? Because they are very sensitive to trends. At that time, there was a word in the Web3 industry called "cognition", which means how you understand the market and why you can make money that others cannot. It requires a higher level of cognition, so most people have a richer understanding of overseas information. They check Twitter every day and see some overseas trends first, which may be a trend in the future.

This is why we opened a store in Wudaokou, a place dominated by academics and technology, where there are many young people who are curious about new things and accept them quickly. When GPT first came out, we all knew about it the next day and realized that this might be the next technological outlet. At that time, everyone was shocked by the terrifying effect of GPT and rushed to register GPT accounts in various ways, chatting with it for at least 2 or 3 hours every day.

At that time, everyone wanted to know what was going on, so we immediately invited Mr. Ma Zhankai, one of the creators of Sogou Input Method, to give a speech. He is very familiar with human language habits and everyone's search needs (Editor's note: Sogou was the first to propose the use of input method combined with search engine to realize associative input, quick search and other functions).

The event was held in the afternoon and lasted for three hours. The entire cafe was packed with people. We only had 40 seats, but more than 60 people came in the end. The distance between each seat was less than one meter, and even coffee could not be delivered. Those who came late had to stand near the door to listen.

At that time, in fact, no one knew what AI could do, but on the day of the event, we could already see everyone’s enthusiasm for AI. This was also reflected in the number of registrations for the mini program. All activities of the cafe will be posted on the mini program, but the number of registrations for this event was higher than previous Web3 events.

All of our PPTs and posters for that event were made by Midjourney, and all the copywriting was also written by AI. Since then, almost all of our own posters have been made by AI.

At that stage, everyone was still enthusiastic and excited, thinking about what AI could achieve and what they could do with it. Later, Microsoft held an event here to tell everyone how to start a business through AI, such as providing free servers and computing power for everyone to use, as well as Microsoft's support for AI startups, which tended to guide everyone to start their own businesses.

Soon, the first wave of AI startups began. But what’s interesting is that this wave of AI startups is very different from Web3. Web3 is more aggressive, while AI startups are very cautious.

Web3 practitioners hope that the whole world knows what they are doing, so that they can achieve the purpose of publicity. People in the AI ​​industry are afraid that others will know what they are doing. Unless my project reaches the financing stage, I will tell others about my project logic. But the problem is that most projects are actually shells, so many people dare not speak openly.

This can also be seen when holding events in coffee shops. Compared with Web3, guests from the AI ​​industry are particularly difficult to arrange.

● After entering the door, the display rack is filled with photos of web3 activities. Photo: Zhao Yifan

When we held Web3 events before, it was easy to invite founders and builders from some institutions or companies. It took only 20 minutes for Yi Nengjing to come to hold an event. But now, when we hold AI-related events, it is difficult to invite guests from other technical departments, except for those in charge of the developer platform in the company. For example, Mr. Ma Zhankai, who I just mentioned, disappeared after going to "Light Years Away". It is true that AI has higher confidentiality requirements for technical content.

Most AI activities are also shared behind closed doors, usually with invitations from the company hosting the event, or from internal partners, or developers of their platform.

The flow of customers in the coffee shop has also decreased. In the golden age of the Web3 industry, our place was the first choice for people in the industry. Some people even came here every day just to have the opportunity to meet people in the industry, introduce their products, and find cooperation opportunities. People who work in AI, except for competitions, training, sharing and other activities, rarely take the initiative to talk about their products. Some people who entered the AI ​​circle from Web3 have gradually disappeared and it is difficult to find them.

The change of my friends is also very obvious. When they were working on Web3, they introduced themselves as the builder or founder of a project. Even though the domestic control of this industry is so strict, they still promoted themselves in a high-profile manner. But after working on AI, everyone is very cautious and keeps their ideas confidential.

This is because AI has a particularly large information gap. Even if you have a particularly good idea, the cost of implementing the idea in the field of AI is actually very low, so if you tell others about your idea, they can also do it.

During the college entrance examination period, a friend I met in a big company used "Button" (an AI application development platform developed by ByteDance) to create an AI application program. It took only two and a half days from getting familiar with the platform to programming the school's enrollment brochure and programming the program. Because common application services on the market cost thousands of yuan per time, his program only charged 50 yuan per time. In the week after the college entrance examination, he earned a total of 1.7 million yuan.

Of course, he only talked to me about this matter after he made money.

After July, investors have the ability to prevent fraud

After turning to AI, we have been thinking about how to implement the elements of AI and reflect them in physical stores, so we came up with the form of real-time call volume rankings. Some people even argued with me why their ranking was so low.

It can indeed be felt that the customer flow of the cafe is not as high as before when it focused on Web3, especially after August and September last year, the AI-oriented activities have gradually decreased. Before that, there were 8-10 events a month, and later it became 3-5 events a month, which are still mainly training, such as communication, literacy and the like.

We don't make money from coffee. Coffee doesn't make money. For us, coffee shops are a cost unit and they lose money. We hope to meet more young people and entrepreneurs through events, and help them incubate and complete investment through this kind of companionship. Here, we can link the relationship between people, the relationship between people and projects, and the relationship between projects and capital. In this process, we have completed the launch of about 14 projects.

● It is a cafe during the day and a bar at night. Photo: Zhao Yifan

For example, the investor you just met has invested in many projects here. There are many projects that I went to talk to first. After I finished talking, I felt that the project was good in terms of personnel structure or market direction. If the investor happened to be there, I would ask you to have a chat. Many projects were promoted in this way.

He is a senior LP who manages a fund of funds with a scale of over 10 billion. He is still on the front line looking at projects and talking to many young people. Normally, in large investment institutions, interns are the ones who connect with them. Why does he go to talk in person? I discussed this with him, and he said that if the interns go to talk, and wait for the group discussion to report to the company meeting, and then he goes to talk to the entrepreneurs, three to five months will pass, and the project may die.

When GPT came out, the most advanced investors would not miss this opportunity. At that time, investors were all FOMO-conscious, thinking that they would make money if they invested, so everyone was frantically looking for projects outside.

At the beginning, investors were actually very confused. They didn’t know “what is Transformer”, “what is the underlying logic behind Transformer”, or “why entrepreneurs need to buy graphics cards”. They also need a learning process for this whole set of new content.

But this wave of AI craze actually only lasted until July 2023. After that, the entire industry entered a cooling-off period, and people around me became much more cautious. Although Wenshengtu became popular, the phenomenal product "Miaoya" appeared, and there are some other medical models such as AI decoration, AI education, and AI-assisted diagnosis, many people found that the auxiliary role of AI in creation did not meet the initial high expectations. Professional practitioners felt that they could not use it, and amateurs generally could not use it.

The bubbles were burst one by one.

At this time, investors also began to have a certain ability to prevent fraud. They realized that entrepreneurs did not need so much money, and many products were actually shell products.

At the beginning, as long as entrepreneurs said they were working on AI, "I need money to buy cards and computing power, and I invited scientists to start a business together", they could get investment. Later, after sorting out, it was found that many projects, for example, invested 20 million US dollars, of which 10 million was used to buy cards, so investors became more cautious when investing in projects.

After July, investors will also compare the quality of different projects and future risks, such as "what impact will the future iteration of the big model have on the product" and "what is the current position of the product in the market". There was no ranking list of AI products before, but now everyone suddenly finds that every track is full of people, and investors will compare how much your product is different from others.

When everyone was scrambling for projects, it only took three months from seeing the project to understanding it and finally investing in it. But now, people simply don’t invest because some funds have run out of money and the latest round of fundraising is not going smoothly. Many funds are very anxious to remit funds, and they will urge founders to just give them some money from the funds that are about to expire.

In this case, investors have also changed their minds and feel that projects that can quickly recover funds will have a longer life cycle.

This is also the biggest change in the AI ​​industry. When GPT first appeared, investors regarded investing in AI as a career, for example, they invested money to support entrepreneurs to create the most powerful large model in China. Under this investment logic, investors will participate in the first, second, and third rounds of investment in order to support startups.

At that time, everyone wanted to know which direction would lead to success and make AI a reality. However, after a round of exploration, we found that AI developers did not understand the know-how of the industry and did not know how to implement it. Later, everyone realized that +AI might be a way for enterprises to use AI to improve efficiency, but this is completely a To B capability, not a project like a large Internet platform. This kind of investment does not require a great investor, and you cannot accomplish a great thing.

Therefore, investors now expect fixed returns. They no longer pursue incubating a great company or projects with hundreds or thousands of times returns, but hope to invest only a small amount of money and earn it back by the end of the year.

The ideal is too far away, it is better to make quick money

It’s not easy to make money from B2B.

Since this year, you will find that there are fewer and fewer entrepreneurs who talk about the relatively empty direction. More of them are combining scenarios to make the big model truly realized. This is also the most discussed thing in the entire industry during this period. Among these directions, there are not many C-end ones. Most of them are To B, and most of them are SaaS or AI tools. But later, it was gradually discovered that it is not easy to do such a track that involves too much internal workflow and know-how in the industry. On the contrary, it is easier for traditional enterprises to do it themselves, so a large number of entrepreneurs fell on this road.

Now investors will ask you how long it will take to make a profit, what is the scale, what is the cost, and how is the cost control. If you can't answer, you will hardly get any money. Only the five AI tigers (Editor's note: referring to the five big models of "Baichuan Intelligence", "Dark Side of the Moon", "Zhipu AI", "Zero One Everything" and "MiniMax")Unicorn Companies, in no particular order) is still in full swing, and investors are rushing to invest. In fact, if investors still have money, they would like to continue to invest, because they can see the potential of these companies and think it is worthwhile.

Especially since ChatGPT cannot be used in China, people are more certain about these five companies. Some people think that these companies are similar to Google after it withdrew from China.Baidu, there is a clear market space.

With money coming in continuously, these star companies are not concerned about commercialization.

For example, Dark Side of the Moon, their company and founders focused on R&D, not on monetization. Last month, I wanted to introduce a project to them. My friend wanted to cooperate with them on a company's private service deployment. But the person they contacted said, "Sorry, we don't take this." What it actually means is that we have too much money now, and we don't need customers or make money.

Compared to Web3, AI’s story is more difficult to tell. In the Web3 industry, there is a saying called narrative. You need to clearly explain your technology or product and what the next scenario or trend is. When this story cannot be told, it is basically broken. For example, Bitcoin was about digital gold, and the mines were limited. Once they were mined, it was over. Many people believed in this narrative and participated in it.

● The wall is filled with famous NFT portraits. Photo: Zhao Yifan

In addition, every industry needs a myth of wealth creation. From oil exploration and the San Francisco gold rush to the initial Internet, China's first generation of Internet financing has made many people rich. In Web3, this situation will be more obvious, but there are not many such stories in the AI ​​industry at present.

Currently, AI only has some small businesses that can make a little money. Previously, many bots were used for transactions in the Web3 industry. In fact, they are just intelligent entities.robotYou can give it some instructions, and it will give you feedback and help you automate tasks. But it was usually used for quantitative trading before. After AI became popular, many robots with good training results became pornographic products because this method makes money very quickly.

The proportion of pornographic content produced by AI is actually increasing. This is because the technology is very mature and profitable. Some products can earn more than one million US dollars a day.

There is another model that people who work on Web3 are very familiar with, which is to make money through training or selling courses. Previously, there was a KOL in the Web3 field. When ChatGPT was just launched and many people didn’t know how to access it, she made a small domestic portal and attracted new users through WeChat groups. She charged 90 yuan per user and had about 20,000 people at the peak.

After July last year, as the market cooled, this type of training obviously increased. But the Web3 group has also become accustomed to this way of making quick money. I think everyone has a dependence on the path to success. He used to do this in Web3 to eliminate the cognitive gap between people, and some people are willing to pay.

If the criterion for defining whether a Web3 project is successful is whether it makes money, even using this criterion, it is too early to talk about the success of AI.