The large-scale venture capital market has come to a new stage of "no company to invest in" .
Entrepreneurs want money, but new rounds of large-scale financing are becoming more and more difficult; investors find it difficult to find suitable targets, and those who are promising can't invest, and those who can afford it can't.
This phenomenon is both unexpected and reasonable.
Ten thousand arrows have been launched for less than half a year, and the frequency of ordinary people's use of AI and large models has declined. When the enthusiasm has calmed down, even the growth rate of the leader ChatGPT's visits has begun to slow down.
But the volume and congestion of this track is far from over :
As the first batch of AI start-up companies, a team that has been deeply involved in lightweight models has gone straight to unicorns; the Tsinghua Department of AIGC startups is constantly emerging, and another AI company focusing on dialogue systems has recently entered the market, as well as undergraduates. The AI assistant started by graduates has started beta testing soon.
On the one hand, the financing progress of popular players is not smooth; on the other hand, new companies and new products continue to squeeze into the track.
It's now that reality is finally starting to ring the alarm bells:
The frenzy is fading, and AI start-ups are undergoing a greater test in the venture capital market . Perhaps, the reshuffle has come quietly.
The track resources are limited and the time window is short
Driven and stimulated by OpenAI, domestic and large-scale model startups are scrambling to burst out.
But the further we go, the more tests and difficulties we need to face are becoming more and more obvious:
first--
limited track
Today, almost every large-scale model start-up company can keenly feel that although this track is hot, the track resources are limited and the time window is short .
Every update of ChatGPT is crowding out the living space of start-up companies. Even Jasper, the AIGC star unicorn with a valuation of 1.5 billion US dollars in 18 months, has also begun layoffs.
Looking at the open source field, represented by LLaMA-2, the overall strength is also increasing rapidly.
In vertical comparison, there are too many companies that have gone to large-scale entrepreneurship one after another. According to incomplete statistics, including general and vertical, open source and closed source, the number of domestic large-scale models has exceeded 100 .
It can be intuitively felt that in just half a year, among these hundreds of companies, celebrities and small giants have appeared at an almost gushing speed .
One of the brightest stars, MiniMax , is on its way to becoming a unicorn. According to market news, a team that was the first to end in China took the lightweight route and has already gone straight to a valuation of 1 billion US dollars.
It's exciting and mind-boggling:
Even in the era of the Internet with the most hot money, a 6-month valuation surge is almost like a fantasy.
Not to mention the special case of light years away - before being acquired by meituan, it took 4 months without demo outflow, and the valuation exceeded 1 billion US dollars.
The dazzling stars mean that they will occupy most of the positions on the large-scale model track, and the resource grabbing battle for the remaining 99% of players is destined to be fierce .
Venture capital support, start to get together
In such a general environment, after half a year, the investors who really sat at the poker table "clustered together" was particularly obvious.
According to the market valuations that have been exposed, Qubit has sorted out the companies and investors that have made moves in the large-scale venture capital circle since January 2023 .
△ Qubits are compiled according to public information, only in 2023
In addition to the table, the latest valuations of star companies such as 01Wanwu and Baichuan Smart are still dormant and unknown to outsiders.
However, the information in the table is enough for people to discover that in the era of large-scale entrepreneurship, VC’s game of gamblers is very regular:
Or, look for the head .
A typical example is the most frequent shots in the past six months, such as Dachang Venture Capital/Venture Capital represented by Tencent Investment, or Zhipu AI, which is speeding all the way on the track.
For them, eggs will neither be placed in the same basket, nor will they be placed in too many baskets, nor will they be placed in baskets other than the head echelon.
Or, firmly choose a certain invested company .
It's like Tom Cat's overweight on Xihu Xinchen, or Zhihu's continued optimism on Mianbi Smart, and his partner and CTO serve as the other party's director and CEO.
△ Li Dahai, partner and CTO of Zhihu, served as the CEO of Mianbi Smart
If you want a high valuation, you need to face the bottleneck period
Furthermore, even though everyone in the venture capital market talks about large-scale models, the amount of money that can be invested is limited, and with the retreat of US dollar funds, the amount of money is not hot, let alone large .
Qubit learned that a start-up team that announced its entry into the large-scale model a few months ago has not been smooth sailing in the negotiation with the employer. The current external quotation has been lowered to below 100 million US dollars, which is far lower than the initial expectation.
The discussion is hot, but financing is not easy, not only reflected in this.
Large-scale entrepreneurship is only in the rising stage, and a high valuation means that the invested company will need to bear higher growth expectations and profit pressure in the future.
Market news is that the later the financing round, the higher the valuation, and the more real money is needed, the more stringent the test for real capabilities and technology commercialization will be .
Until today, the valuation of 1 billion US dollars has become a ceiling that ordinary large-scale startups can hardly match; before reaching this amount, 1 billion yuan is also not easy to pass.
Gradually usher in the first wave of cooling-off period
However, the funding challenges faced by these start-ups have been particularly daunting from the start.
The cost of model training is high, data acquisition is difficult, and top technicians can live in rare goods. In a word, if you want to build a large model, you must spend money like water.
Now we still have to face problems such as the difficulty of transforming into actual business, which impedes the commercialization and further rapid growth of many large-scale companies in all aspects.
Ask for mergers and acquisitions into parallel options
Therefore, when the new round of large-scale financing became more and more difficult, mergers and acquisitions began to become a parallel option.
Qubit recently learned that a Tsinghua-based AI large-scale model company was exposed to be looking for a new round of financing at a valuation of 1 billion yuan.
At the same time, the company is seeking mergers and acquisitions in the market at a price of 100 million US dollars. This price is at par with the valuation when it completed the last round of financing.
News from the capital market also spread that Zhipu AI, which is also from Tsinghua University, is contacting the team about mergers and acquisitions.
This is not the only way of life. There are always similarities that can be mirrored in history:
In April of this year, Light Years Beyond completed the acquisition of First-class Technology, an AI deep learning framework company, holding about 46.52% of its shares.
At that time, the first-class technology that had completed 4 rounds of financing had repeatedly failed in financing, and almost fell before the rise of the AI 2.0 wave; and Wang Huiwen stood on the top of the mountain, and urgently needed to introduce industry resources to enhance competitiveness and reserve strength for "building China's OpenAI".
Everyone knows the story later, Wang Huiwen retired due to illness, and first-class technology was brought under the banner of Meituan light years away.
Investors stick to the "conservative" line
The accident of Wang Huiwen's torrent exit, to some extent, strengthened the common attitude of most investment institutions, which is conservative .
Many first-line investors confided that they have seen and talked about a lot of projects in the market, but in the end there are very few anchored targets.
As mentioned above, the capital concentrates on betting , the hunting target is concentrated on 3-5 players, and the chances of success are extremely compressed.
The intensification of the Matthew effect has further reduced the number of start-up companies that are expected to break through, but looking at the companies with names in the market, the asking prices have skyrocketed.
Even if you firmly believe that "AI 2.0 era has opportunities 10 times greater than the Internet", whether it is worth such a high price is a question mark .
After all, after pulling up the team, the business model of the large model is not clear, the profit point is not clear, and many entrepreneurial projects are still in the proof-of-concept stage.
There is still a long way for domestic players on the track to train large models, produce real products, and even catch up with GPT-3.5 .
Some investors simply gave up looking at large-scale model projects and turned to the old way of looking at chips and hardware at the infra level.
Wait for them to release the model at the end of the year, and try the depth.
Still on the sidelines
Investors hold chess pieces and dare not try it lightly. Those who have not yet bought tickets, some are waiting to see whether there will be a model more powerful than LlaMA-2 in the open source field, so that everyone has no chance to play?
There are still opportunities waiting for the application layer .
It’s just going back to the battle between Zhu Xiaohu and Fu Sheng in the circle of friends a month ago. For the slightly thin application layer of Know How, “Everyone knows whether there is value created, just don’t have too high expectations. .”
And, in the actual situation, it is basically built on the API provided by others. Is it not very friendly to startups?
Amid such loud voices, the doubts in people's hearts intensified:
Can it really be done in a short time? After making it, how far is the effect from GPT-4? Have a certain strength, can you find a suitable commercial landing scene? Is the investment really proportional to the value?
Investors who didn't make a move seem to be even less anxious.
Eight months ago, ChatGPT sparked a thousand waves; six months ago, the domestic large-scale model track began to attract attention, and the popularity gradually became more and more popular.
Up to now, there is still an endless stream of people who want to take the large-scale model route to AGI .
The top players still have to meet countless investors every week, the valuations of the top echelon companies continue to rise, and the emerging rising stars are still rushing to the field.
At the same time, we are faced with more calm and harsh evaluation eyes from investors, from the industry, and from ourselves—and more or less with increasing doubts.
After half a year of hot speculation, the fire burning under the large model began to turn to a small fire mode. In the gap between the "announcement of entry" at the beginning of the year and the "release of the model/product" at the end of the year, there are various signs that the venture capital circle of large models is gradually welcoming . The first cooling-off period came .