Image source: generated by Unbounded AI

The AI ​​unicorn company "Dark Side of the Moon" has once again been pushed to the forefront.

On December 5, Zhu Xiaohu, the managing partner of Jinshajiang Venture Capital, posted two messages on WeChat Moments, revealing more that "Jinshajiang Venture Capital has teamed up with several circular intelligent investment institutions to invest in the founder of Dark Side of the Moon." Yang Zhilin and co-founder and CTO Zhang Yutao filed for arbitration. Many details of the matter have attracted industry attention.

The tense relationship between investors and entrepreneurs is just a microcosm of the harsh entrepreneurial environment of large-scale models.

Two years ago, ChatGPT led to the AIGC craze. Baidu, Alibaba, Tencent, and Byte successively invested tens of billions of yuan into the large-scale model industry, and a large number of startup companies and venture capital teams also rushed in. enter. Both platform giants and entrepreneurial teams believe that this will be a disruptive technology wave in history.

The illusion is over, the development of large AI models has slowed down, and platform giants and entrepreneurial teams have flocked to application development. The entire industry has experienced drastic changes from the blue ocean to the red ocean, and the competition for entrepreneurship has shifted from technology to user growth, product launch, and commercialization.

An AI large-scale model entrepreneur told Tech Planet that the effects of B-end AI products launched by many startups are far from meeting customer expectations, and it is difficult to get money back and renew contracts. The C-side lacks blockbuster-level applications, making it difficult for users to gain mass.

An investor in the AI ​​​​large model track said that many companies that advocate technology supremacy have reached the application stage, with no room for growth or realization. If investors’ emotions turn from trust to doubt, companies will Subsequent financing will become increasingly difficult.

As if the "fast forward button" has been pressed, the AI ​​large-model entrepreneurship boom has accelerated from hot financing to fierce competition, and then to the highly questionable technology entrepreneurship cycle.

As 2024 draws to a close, AI large model entrepreneurship has entered a period of adjustment. At this stage, companies are either seeking acquisitions, laying off employees, or even choosing to leave.

ChatGPT was born two years ago, and the entrepreneurial elimination competition is accelerating

AI large model of the knockout rounds, pressing the accelerator button in May this year.

In May of this year, in order to attract users, cloud vendors launched a price war for inference computing power. ByteDance's Volcano Engine, Alibaba Cloud, Baidu Smart Cloud, Tencent Cloud, and iFlytek have successively joined the ranks of price reductions, reducing the price of large model inference computing power by 90%.

The price war launched by big manufacturers and giants will first impact the start-up companies in the industry. They have difficulty in realizing cash and have meager profits. As the gross profit of the industry declines, the pace of self-development has also been disrupted. In the face of chaos, the problem of rapid commercialization has been put on the table.

An industry practitioner told Tech Planet that large manufacturers inherently have more resources and greater scale advantages, and can operate at lower margins.To provide services at an international cost, the current low-price strategy offered by large manufacturers is already far below the cost line of start-up companies, making it impossible to follow up.

Previously, Wang Xiaochuan, founder of Sogou and founder and CEO of Baichuan Intelligence, also judged that with the price war of domestic large models reduced to free in May this year, the To B model market will prosper faster. But tail companies will withdraw from this track.

Being acquired is the ideal outcome for many large AI model entrepreneurs. An entrepreneur told Tech Planet that the growth of the AI ​​industry has not brought corresponding increases in market demand and innovation in profit models. Many companies have been unable to maintain operations. “Compared with canceling or suspending business, being acquired at least means survival. ."

At the end of October this year, Waveform Intelligence, a large model company in the AI ​​vertical field, was acquired by smartphone manufacturer OPPO. From its establishment in 2023 to its acquisition, Waveform Intelligence has received two rounds of financing from investors including Lanchi Venture Capital and West Lake Venture Capital. After being acquired, Waveform Intelligence stated in a follow-up statement that "the products and the company are still operating normally and are not affected."

Waveform Intelligence is not an exception. Looking at the global market, American technology giants are also operating Acquire core talent and technology from AI startups. In August this year, the artificial intelligence startup Character.AI was acquired by Google at a valuation of US$2.5 billion. In June, Amazon announced a technology licensing agreement with Adept, and its co-founders and some team members joined the e-commerce giant. Back in March, Microsoft also bought Inflection AI for $650 million.

As AI entrepreneurship enters a turbulent period, a group of mid- to high-tech and product talents are also flowing. In the second half of this year, the founding team of Waveform Intelligence was packaged into OPPO to be responsible for interactive products, while Qin Yujia, the founder of Xuzhi Technology, and Huang Wenhao, the person in charge of pre-training of Zero-One Everything Model, successively joined ByteDance...

A senior headhunter Tell Tech Planet, "Currently, some large companies will go out of their way to find a senior executive of a startup company's corresponding business and directly recruit people with large sums of money."

These actions all show that after the two-year boom from the end of 2022 to the present, , the large model market has entered a period of consolidation. Just like the entrepreneurial trend in other industries in the past, the resources that once surrounded startup companies eventually returned to large manufacturers.

Star startups and investors turn against each other

What makes the AI ​​​​big model startups even worse is the slowdown in financing.

According to Orange IT data, since the beginning of this year (January 1st to December 5th), 439 financing cases have occurred in the domestic AI field, with the total financing amount exceeding 56.4 billion yuan, which is about 80% of last year. According to rough estimates, the average monthly financing amount in the AI ​​field this year is less than 5 billion yuan.

On the one hand, it is common for institutions to reduce investment. According to incomplete statistics from Zero2IPO Research Center, in the first half of this year, domestic foreign investment activities included early investment, VC, and PE.The number of institutions decreased by 23.9%, 19.2%, and 25.2% respectively year-on-year.

On the other hand, the patience of many investors is gradually being exhausted, and cracks have arisen in their relationships with companies. In November this year, the big model star startup "Dark Side of the Moon" turned against its former investors and went to arbitration, which attracted widespread attention in the industry.

A primary market practitioner said that there are not many cases like Dark Side of the Moon that have reached the arbitration stage. He said that in industries with high concentration of venture capital, it is not uncommon for disputes between investors and founding teams, and the conventional solution is to maximize the interests of all parties through negotiation. The direction of Dark Side of the Moon can be called a special case in the industry.

However, the above-mentioned practitioners also believe that with the deterioration of the AI ​​large-model startup market environment and the extension of the exit cycle, the originally harmonious relationship is gradually becoming tense, which reflects the relationship between investment institutions and enterprises. The trend of trust collapse.

A member of the AI ​​entrepreneurship team told Tech Planet that they did feel pressure from investors this year. “We had several disagreements with investors and the founding team, but in the end we managed to reach an agreement. "

A private equity investment industry insider told Tech Planet that the core reason is that it is becoming more and more difficult to raise funds and exit, and the proportion of previously invested projects that cannot exit normally is increasing. High, investment institutions have taken on greater risks than before. Looking specifically at the current investment environment, he said that some high-quality projects are difficult to promote the investment process due to their high prices, and cheap projects will cause internal hesitation. Whether to invest or not is a dilemma.

AI investors during the period of the Four AI Tigers were patiently waiting for the product to mature and be implemented. Most companies were able to receive a steady stream of funds, and this round of AI large-model venture capital investment has ended. It seems that it is faster, and it is difficult for entrepreneurs to find the venture capital institutions that spent huge sums of money in the past.

An investor who has participated in the financing of large-scale enterprises revealed to Tech Planet that compared with the past investment logic of investing based on the technology, prospects and team, the internal decision-making cycle of many institutions is getting longer and longer. We will also repeatedly calculate and evaluate the return rate of new money investment. "Based on this year's trend, we will most likely further reduce the amount of investment projects next year."

More and more difficult games

After more than two years of excitement in the AI ​​large model industry, many people have found that "burning money" with large models seems to be a bottomless pit. Whether they are companies or investors, it is difficult for them to make money in a limited time. Receive return on investment within 10 days.

Thanks to the popularity of ChatGPT, OpenAI’s annual revenue has climbed by about 4 billion US dollars, but AI model development is like a bottomless pit. OpenAI, which has 180 million individual users and 1 million enterprise subscribers worldwide, is the same Earn in the quagmire of lossestie. According to the "2024 Artificial Intelligence Index Report" released by Stanford University's Li Feifei team, OpenAI is expected to lose about US$5 billion this year.

From a commercial perspective, the hope of the basic large model has been shattered. It is indeed not a good business that can bring profits. However, huge capital needs and extreme dependence on financing have made large-scale AI model companies increasingly eager for commercialization.

In the second half of this year, everyone from giants to start-ups have turned their attention to the field of AI applications. Among them, competition in tracks such as AI search, AI companionship, AI video, and AI productivity tools is particularly fierce.

A practitioner in the large model industry believes that the so-called AI application is to lower the threshold for use, and the logic behind commercialization is actually the user. Fu Sheng, chairman and CEO of Cheetah Mobile and chairman of Orion Star, also publicly stated, "Large model casing is okay, as long as there are users, you can make money."

Whether it is B-side users or C-side users Users and large AI model companies are exploring better implementation methods. For example, the domestic large model "Six Little Tigers", Zhipu AI and Baichuan Intelligence are mainly oriented to the To B market and focus on enterprise-level applications; Zero One Wanwan applies large model technology to many fields such as AI writing, programming, and medical care. The scenes are relatively diversified; Dark Side of the Moon is mainly engaged in the To C market, focusing on the entertainment and social fields.

Although the AI ​​application market is very large, one of the above-mentioned entrepreneurs said that at present, entrepreneurship at the application layer cannot make clear differentiated competition with big companies. The tentacles of giants are extending to every business. The scene makes it more difficult to start a business.

For example, Zhu Jun, Vice President of ByteDance, mentioned the meaning of the name "Doubao" in a public speech this year. He said that Doubao means "everything is included", which means that the needs of work, life and study are all included. It’s included, which means exactly the product vision of Universal Assistant. According to data from QuestMobile, from its launch in August 2023 to May 2024, Doubao’s monthly active users exceeded 26 million. By September, its monthly active users were close to 42 million.

If measured by the standards of the mobile Internet era, large model products such as Doubao backed by giants cannot yet be regarded as super applications of the era, but for startups, every battle to find opportunities in the future , may have to start a battle with the giants.