AI bubble is about to burst？ Allen Zhu said that the big model business model is very poor, and the founder of Sea said that it is difficult for enterprises to transform AI ｜ Titanium Media AGI
Although ChatGPT has been arrogant for more than a year and triggered a new wave of AI catching up, domestic doubts about AI losses remain unabated.
On the afternoon of May 8th, at the 18th China Investment Annual Conference, Allen Zhu, the managing partner of Jinshajiang Venture Capital, said that the business model of the big model was very poor, and there was no difference in technology. Moreover, every generation of models had to spend money again, and the realization cycle of each generation was only two or three years.
"I think altman (co-founder and CEO of OpenAI) is bragging, and GPT-5 is definitely not so amazing." Allen Zhu believes that GPT-4 has met most of the commercial needs, and he predicts that a new wave of AI applications will come next year.
At the same time, Forrest Li (Li Xiaodong), the founder of Sea Donghai Group, a technology leader in Southeast Asia, and the former richest man in Singapore, recently issued an internal letter to 60,000 employees, warning that the transformation of enterprises to AI technology in the next few years will bring chaos and the transition will become very difficult.
Questioning twice a day, is the AI "bubble" really broken?
At present, the scale and quantity of investment in the global AI field have declined across the board.
In a recent report, the Stanford Institute for Human-Centered AI (HAI), co-led by Li Feifei, indicated that in 2023, the global investment in the AI field has declined for the second consecutive year. Considering minority equity transactions and public offerings, the total investment in AI reached $189.2 billion last year, a decrease of 20% compared with 2022.
Diana Iovanel, a macro economist at Kaitou, predicted that the bull market of US stocks driven by AI will continue. By the end of 2025, investors’ enthusiasm for AI will not be reduced, and the stock market index of AI will rise by more than 27% in 2025 (compared with the current level).But the AI capital bubble may burst in 2026.
So, is the "AI bubble" really about to burst?
At the 18th China Investment Conference held on May 8th, Allen Zhu andLiebao mobileFu Sheng, CEO and founder of Orion Star, made a round table discussion on AI investment and financing.
A year ago, Allen Zhu said that ChatGPT was very unfriendly to startups, so please give up your fantasy of financing in the next two or three years. At that time, Fu Sheng forwarded Allen Zhu’s article and criticized that "half of the startups in Silicon Valley started around ChatGPT, and our investors can still be so ignorant and fearless."
A year later, the two sides reviewed the dispute, and Allen Zhu thought that the title party had caused the misunderstanding. "Later, we communicated, and we actually have the same point of view. In AI, I am personally very optimistic about the outbreak of AI applications next year." He believes that the two small models of Llama3 are very powerful, and the iPad Pro has recently launched the M4 chip, which means that a small model with tens of billions of parameters can be directly run on the end side in the future.
Allen Zhu said that after GPT-4, he didn’t expect much from GPT-5. "We all feel that altman is bragging. GPT-5 is definitely not so amazing, or it doesn’t matter whether it is amazing or not. GPT-4 has met most of the business needs, and it will be very similar to the x86 architecture. The upgrade in the next 10 years will not change the business model much."
Allen Zhu pointed out that,When the business model of the big model is too bad and the technology is not too bad, every generation of technology needs to be invested. Now it may cost tens of millions of dollars for version 3.5, hundreds of millions of dollars for iteration to version 4, and billions of dollars for version 5. Every generation of models needs to be re-invested, and the realization period may be two or three years. "This is worse than a power plant."
"Power plants basically don’t need to invest a lot of money after you invest in infrastructure, and the big model is to spend more money to upgrade every two or three years, and it may be two or three years in the later stage of realization. To be honest, the business model is very poor. And we all feel that Sam Altman is bragging, right? " Allen Zhu said.
"I don’t think the’ Four Little Dragons’ (Despise, Yitu, Shang Tang and Yuncong) are the lower limit but the upper limit", while Allen Zhu is a little more radical. He thinks that the’ Four Little Dragons’ are very lucky and have met with a huge bonus from the security construction. Although the cash flow is not good due to the nature of the project, there are also overdue funds that constantly cushion the cash flow. "Why didn’t we vote for the’ Four Little Dragons’ in those years? We felt that technology had no threshold, and it was half a year ahead of others at most. Why is the competition in the market still so fierce today? (Because) there is no difference in technology, and the big model is the same. I think it is even worse."
Fu Sheng, who spoke at the same meeting, agreed.
Fu Sheng believes that, unlike the AI 1.0 era where the AI Four Little Dragons are located, in the AI 2.0 era starting from the release of ChatGPT, the whole society quickly reached a consensus that this technology will bring about a productivity revolution. This means that in the AI 1.0 era, the time difference left for start-ups is enough before the technology giants leave the stage, but in the AI 2.0 era, the technology giants in China leave the stage earlier than the start-ups.
Fu Sheng also said that he doesn’t think the "Four Little Dragons" is a good model today. "After reading some listed companies’ financial reports, they have a revenue of 3 billion yuan and a loss of 7 billion yuan. I think good companies should certainly invest in the early stage, but they should not be like this for 10 years." He said that the cheetah also suffered some losses, but all of them lived within their means, leading to a certain extent, but they did not invest at all costs. All technologies must finally form a closed loop.
Fu Sheng pointed out that the business model of large model enterprises, including OpenAI, may have challenges. He believes that the emergence of intelligence with large models is like Columbus’s voyage. After Columbus first arrived in the American continent, everyone will follow up quickly, and it is difficult for large model enterprises to stay ahead of their competitors for more than one year in technology, leaving the time window for commercialization. Too short.
He tookappleFor example, the company said that although Apple has money, it only needs to focus on developing small and medium-sized models and chips at the end in the short term, and the large models can be connected to other manufacturers.
Fu Sheng judged that in the future, every device, including computers and mobile phones, will have a local model built in, and the entrepreneurial opportunity lies in how to package this technology into products for users to use. For example, when he made a private model, he found that the small and medium-sized model with 10 billion parameters was enough in most scenarios. If the scenes specified for it were clear enough, the performance would be better than that of the large model with more parameters and more generality.
On the path of the big model, Allen Zhu believes that the open source small model is bound to be the future direction, and there are many business opportunities. "Most of the jobs don’t need Tsinghua undergraduate or Harvard doctor, just a junior college student is enough."
Fu Sheng recognizes this. "In the end, the intelligence level of the 10 billion parameter model may be enough to meet most applications. Even if the scene specified for it is clear, it will be better than GPT-4. I will recruit a secondary school student with you and answer questions every day. You don’t need to write ancient poems or participate in the Olympic Games."
Allen Zhu also mentioned that an internal recovery is that applications make the most money from PC Internet to mobile Internet. Although at the beginning of each cycle, hardware and infrastructure earn a little more money, but the latter application makes more than 10 times as much money as the former.
He believes that this big model is faster than the iteration speed of the PC Internet period. The iteration speed of the PC Internet is 10 years, and this time it is annual."So why am I optimistic that the application will definitely explode next year, because the open source model has reached a sufficient level of intelligence."
Why next year? "Many times it is not a bad choice to let the bullets fly for a while." Allen Zhu said.
According to reports, recently, Forrest Li (Li Xiaodong), the founder and billionaire of Sea, warned in an internal letter that AI may be a bigger challenge for Sea now, and the company’s transition to AI will be difficult in the next few years. But he stressed that the company’s financial situation is better than in the past decade.
"the AI wave has not yet fully arrived, but we can see that it is coming. This technological transformation may be more difficult for us than the last one. At that time, as novices of e-commerce and mobile games, we completely focused on the new platform from the beginning. This time, we will have to adapt to new technologies. " Li Xiaodong said.
Li Xiaodong
As the owner of Shopee, an e-commerce shopping platform in Southeast Asia, Sea faces fierce competition in the Southeast Asian market with a population of more than 650 million, although the market value of the company exceeds 200 billion US dollars. ByteDance’s TikTok,AlibabaLazada, as well as new entrants such as Temu and Shein, are competing for markets in Southeast Asian countries such as Indonesia and Thailand.
In May 2023, Li Xiaodong announced that Sea had laid off about 7,500 employees, accounting for about 10% of the total number of employees, and frozen the salary of the senior management team. At the same time, in August last year, in addition to AI, Sea also stepped up efforts to build a live broadcast service, which may erode profits and trigger a price war with TikTok and Alibaba.
Now, with the help of cost reduction, Sea has improved its profitability and achieved some success. The company said in March this year that it expects to achieve profit for the second consecutive year in fiscal year 2024. Since the beginning of this year, Sea’s share price has risen by 61%, which is close to the highest point in 12 months, but they are still far below the historical level.
"We care about every dollar and every penny," Li Xiaodong said. "You can have big dreams and big ambitions, but what if you can’t survive? There will always be such a voice in your mind that we may have no money. "
Li Xiaodong hopes that Sea can break even, and the company has followed in many aspects during its development.AmazonAnd its founder, Jeff Bezos. "Innovators in Silicon Valley may have a hard time thinking about these issues."
To sum up, no matter Allen Zhu or Li Xiaodong, everyone is very skeptical about the profitability of AI and the difficulty of the whole industry, especially the "hematopoietic capacity" of the commercialization of AI model is not as good as expected.
In fact, AI is a long-term and innovative industry, rather than a typical "winner-takes-all" industry with near-zero marginal cost growth and high user stickiness in the Internet era. There is no "monopoly" in AI, cloud computing and other fields, so the barriers are not so high, and continuous technological innovation is needed. Once the R&D investment is reduced, it will be like IBM and IBM.MicrosoftLike the events of that year, it directly withdrew from the head position of the industry.Therefore, all AI companies cannot avoid high input costs.
iFLYTEKLiu Qingfeng, Chairman of the Board of Directors, once said that AI technology, just like Gartner curve (a professional chart describing the development cycle of technology), always has a dream period of concept explosion, and then the bubble bursts. However, it is not really shattered. Many people see that it is not so good from the perspective of over-optimism and over-deification, and start to be rational again. After being rational, they will continue to grow and enter the stage of promoting social progress.
Therefore, from a cyclical point of view, despite the obvious head effect in the AI field, this track has no end, and enterprises will not be able to achieve large-scale and monopolistic commercialization by relying on only one AI technology.
In the long run, the large-scale commercialization of AI technology still takes a long time and people need to wait patiently. As for capital, they certainly can’t wait for this moment.
Allen Zhu hit the nail on the head in a paragraph: "I want to give LP a sustained and stable cash flow, which is the most basic role. China wants to talk about’ patient capital’, which I said yesterday. I am patient, but I have no capital, right? This is a joke. How can we have patient capital? Only when there is a steady cash flow every year will people believe you. If I have invested 100 million yuan and haven’t returned a penny for 10 years, I don’t know whether you are a liar or not. But with a stable cash flow every year, everyone will feel more at ease, which will build a patient and long-term investment. "
(This article first appeared in Titanium Media App, author | Lin Zhijia, editor | Hu Runfeng)
