After more than 200 days of AIGC boiling, investors reached three major consensus

Source: First New Voice

Image source: Generated by Unbounded AI

On July 19, Meta, the parent company of Facebook, launched the Llama 2 large model, which will be used free of charge for research and commercial purposes. It is known as the strongest GPT-4 replacement open source. This will change the situation that many large models around the world were developed based on Llama, but are limited by the fact that they cannot be used for free commercial use.

New changes have taken place in the AI market structure, and the focus of the venture capital circle has been locked again. When people are discussing that the singularity of human artificial intelligence is approaching and the era of AIGC is coming, things are slowly changing as AIGC runs wildly.

First of all, in terms of how long AIGC’s popularity can last, the investment circle has gradually divided. Some people said that investment pursues return and certainty, while the liquidity of the general-purpose large model is still unclear. The market is cooling down, so you need to be cautious when buying.

Some people hold the opposite view, thinking that the development of AIGC has just begun, and it will become even more popular next year. Today's AIGC is only in the field of text, and the multi-modal large model has not yet come out. By the end of this year, some breakthroughs in images by Open AI may further stimulate everyone's imagination.

It is not only the attitude that differentiates, but also the indifferent data that can match the high attention of the market. According to relevant data, from the beginning of this year to May, the growth rate of ChatGPT's visits dropped from 131.6% to 2.8%. From the perspective of actual actions, there is a huge contrast between the low number of investments by investors and the enthusiasm for swiping the screen in the circle of friends.

It seems to have become a law of nature that the emergence of new things is always accompanied by "polarized" attitudes. During the more than 200 days of continuous fermentation of AIGC, what consensus have investors reached? Where are the opportunities for entrepreneurs?

First New Voice contacted a number of investors, trying to figure out what has settled down in the AIGC's turbulent process based on the current situation? What happened to boot? It is hoped that this will positively promote and contribute value to the development of the industry.

AIGC wave opens opportunities for the next era‍‍‍‍‍‍‍‍‍‍‍‍

The explosion of AIGC has excited the investment circle.

According to the estimation of Qubit Think Tank, it is estimated that by 2030, the AIGC market size will exceed one trillion RMB.

According to public data, in 2022, there will be more than 500 investment events in my country's AIGC industry, with an investment amount exceeding 90 billion yuan. According to incomplete statistics from Tianyancha and First Voice, from January to June 2023 (as of June 27), the total financing of the domestic AIGC industry reached 4.959 billion yuan, and the number of financings totaled 46 times.

Time goes back to the end of 2022. Fang Zhenghao, managing partner of Xiaomiao Langcheng, noticed that AIGC has attracted a small amount of attention in the technology and investment circles. In March 2023, ChatGPT became popular in the circle. "The circle of friends is swiped by relevant information almost every day. People's excitement and potential concerns about artificial intelligence have reached an unprecedented peak." Fang Zhenghao said.

As the generation of some synthetic video images has made breakthroughs in C-end applications, the feeling that "a new productivity paradigm change has arrived" became stronger in Fang Zhenghao's heart. He believes that artificial intelligence will play a more important role in various vertical fields and industry applications in the future.

Bai Zeren, vice president of Linear Capital Investment, has similar feelings, "AIGC is a very long-term opportunity, analogous to the Internet, and the future development trend must be that AI penetrates into various scenarios like capillaries." He is optimistic about this wave of AIGC, believing that there must be a large number of investment opportunities following it.

"We are very much looking forward to seeing more innovations and changes." Wang Xiao, founder of Jiuhe Venture Capital, said that when a new wave of AI comes, behind the rapid popularization of applications such as ChatGPT is the emergence of a new generation of AI capabilities represented by the emergence of intelligence.

"From now on, whether you are working or starting a business, please make sure that you are related to AI." Lu Qi, former global vice president of Microsoft, Baidu COO, and founder of Miracle Forum, has a firmer attitude. "AIGC is not a current trend, and a trend means opportunism. It is too much to underestimate the impact of AI on the development of the world."

Opportunities for the next ten years, or even the next era, are slowly unfolding.

Only look at the big model and don't cast it, the real money is spread to the vertical model and application layer

During the more than 200 days of continuous fermentation of AIGC, investors have reached some consensus, which is mainly reflected in three aspects:

Consensus 1: There are deterministic opportunities for computing power infrastructure, and the big model is a "game" for the rich‍

‍In the AI new wave ecological architecture composed of computing power infrastructure layer, model layer (base model, open source model, self-built large model), and application layer, some deterministic opportunities have emerged.

**First of all, with the development of artificial intelligence, the demand for computing power has shown explosive expansion. **There are definite opportunities in the computing power infrastructure layer, which has become the consensus of the capital markets in China and the United States.

The performance of the secondary market supports this view. From the end of October 2022 to July 17, Nvidia's stock price soared from $123 per share to $464 per share. Since the beginning of 2023, the stock prices of artificial intelligence companies in the domestic computing power infrastructure layer such as Cambrian and Sugon have been strong.

**Secondly, large models will bring about huge changes in R&D and application paradigms. ** Some investors believe that the prospect of AI technology dominated by large models can effectively reduce costs and increase efficiency has aroused the excitement of entrepreneurs. This is also the reason why the Hundred Models War can take place.

Baidu launched "Wen Xin Yi Yan", Ali released Tongyi Qianwen, Xunfei's Xinghuo large model, Meituan, Baichuan Smart, Yunzhisheng, etc. have also joined the large-scale model track. According to statistics, as of July, my country has more than 80 large models with parameters above 1 billion.

Today, the Hundred Models War has intensified and gradually stabilized. People have gradually reached a consensus on the model layer: the model layer is a game for big "players". This "player" refers to both entrepreneurs and investors.

The inference and training of large models directly require the computing power of the chip and the graphics card, and the model layer requires a very strong technical team support, which makes the capital investment very huge.

Taking Open AI as an example, according to relevant statistics, the cost of GPT-4 training once is about 63 million US dollars, requiring 1.8 trillion huge parameters. This does not include the cost of data collection, RLHF, etc.

In the final analysis, high-end technical talents and chips need to burn money. "The most important thing at this stage is to see who has strong financing ability and strong capital strength, and whoever has a greater probability of success." Fang Zhenghao said that startups in the bureau are faced with the problem of being stuck in the computing power end. Compared with overseas companies, there is still a gap in the large model. On the basis of computing power and high-tech talents, it is a competition to see whose R&D investment is more effective and who can do better in technology.

The same is true for institutions.

"Investment opportunities at the model level can only continue to be played among some players with strong capital." Fang Zhenghao observed that for investment institutions whose management scale is not particularly large, if they do not make an early deployment before the industry is booming, they are unlikely to participate in large-scale model investment at this point in time.

It would be a good time to lay out large-scale models last year, but this year is not a good time window for most early and mid-term investment institutions.

In addition to the huge cost, there are many factors that lead investors to sell cautiously, such as the best time window, commercial liquidity and so on.

"If it were me, I would not choose to invest in large-scale model-related projects this year." Shi Mao, the founding managing partner of Changlei Capital, said that after missing the best time window to invest in the underlying large-scale model track, he observed that there is currently a big gap in the combination of the model layer and the application layer, and the realization of large-scale model technology is still unclear.

It is worth noting that vertical models with clear needs and landing scenarios have attracted capital attention. **

At the beginning of this year, Xiaomiao Langcheng reached a consensus internally not to invest in large-scale models, but to keep an eye on large-scale models in the tens of billions of industries. "Compared with large-scale companies that have already floated on the water, start-up companies will have more opportunities in subdividing vertical fields. Because after landing in a specific industry, it is easier for start-up companies to accumulate higher-quality data sets in it."

This is consistent with the opinion of Zhu Xiaohu, the managing partner of GSR. GSR Ventures is one of the early institutions that made the most vertical AIGC investments in China. Zhu Xiaohu once publicly stated that for most entrepreneurs, they should "scenario first, data is king" and train their own vertical models instead of being superstitious about general large models.

Consensus 2: At the application layer, "old forces" in certain vertical fields have better opportunities‍

A fact is that investors are more interested in the big model and don't invest in it, and invest more real money in the application layer.

"We are also very concerned about the progress and changes of the big model itself. Considering the current market competition pattern and capital threshold, we will tend to invest in opportunities such as the application layer and new infra when making a sale." Bai Zeren said that Linear Capital is more concerned about how new technologies can be implemented in the industry to solve industrial problems more effectively and bring huge commercial value to the industry. This is the constant investment logic of Linear Capital.

"We encourage all invested companies to think about the possibility of combining their businesses with AIGC in the future, at least from the perspective of corporate management, they must also think about how to improve internal human efficiency through AI." Bai Zeren said.

There is currently no consensus in the investment circle on the deterministic opportunities at the application layer. However, many investors said that from the perspective of enterprises, veteran players in various vertical fields of To B have obvious advantages.

Fang Zhenghao compared the companies that have successively entered the field of artificial intelligence to "old forces" and "new forces". The "old forces" started with the deep neural network in 2016. At that time, the first batch of artificial intelligence companies were born, including the AI Four Tigers and some start-up companies. Artificial intelligence companies that have emerged in recent years are regarded as new forces.

"The 'old forces' in some vertical fields have mastered the needs and scenarios of customers, and at the same time can take the lead in meeting the iteration of AI technology." In Fang Zhenghao's view, such companies are companies with relatively certain development opportunities at the application layer.

Some investors expressed similar views. In addition to new start-ups, there are actually a number of AI companies in the application layer in various vertical fields. They have been developed for six to seven years, and they may become the protagonists of AI companies on the application side in the next period of time. "Because they have customers and scenarios in their hands, they will have a more competitive advantage."

Wang Xiao, founder of Jiuhe Venture Capital, said that AI will change all walks of life, including SaaS, tool software, and the previous generation of AI companies are expected to use this technology iteration to carry out structural upgrades. "Xiaoduo Technology, which we invested in in 2015, is based on the big language model technology, and recently launched the Xiao model XPT in the vertical field of e-commerce. With the help of large models and industry data accumulated in the past, it will empower more e-commerce business scenarios and provide better solutions."

Consensus 3: Investors pay attention to the team and commercialization‍

In this wave of AIGC, investors mainly invest in people and directions, and the invisible background has become an important consideration.

"What does the founder see in the future, and what role does he hope to play in the future? We use the future described by the entrepreneur and the future he sees to find a resonance point in the middle. Then we judge whether the entrepreneur can really do this from the perspective of his growth background and technical level." This is the core logic of Zhang Jinjian, founding partner of Oasis Capital, in the investment process.

**In addition to investment, in terms of specific considerations, various institutions focus more on the commercialization ability of the project. **

From the perspective of Linear Capital, commercialization capabilities are mainly reflected in three aspects: barriers to entry, such as sufficiently differentiated technology that can be well adapted to a specific scenario, or the scenario itself requires domain knowledge; rapid productization, which can quickly integrate LLM capabilities to make products on pain points; form an effective closed loop of data and feedback.

Jiuhe Venture Capital, which has invested in star projects such as Eagle Eye Technology and Tanji Technology, has a similar view to Linear Capital. Wang Xiao said that in addition to considering the founder factor in investment, it is also necessary to consider clear landing scenarios and needs, which can truly reduce costs and increase efficiency for customers.

Xiaomiao Langcheng's investment strategy focuses on selection and heavy positions. For 5% of the projects that cannot be missed, we will actively shoot, and for 95% of the projects, we will start from the hunter's perspective and do a good job of research patiently. "The core of investment is to express your understanding of a matter with your heart. If you understand it, you should invest for a long time and continue to invest."

Different concerns: speed of technology development, valuation, business model

The first two waves of AI were in 2012 and 2016 respectively.

In 2012, the deep learning-based AlexNet developed by Professor Geoffrey Hinton and two students won the championship in the ImageNet Large-Scale Visual Recognition Challenge. Since then, deep learning has laid the underlying technical foundation of artificial intelligence.

In 2016, in the human-computer Go competition, AlphaGo defeated the world Go champion Lee Sedol, which quickly ignited the fire of global artificial intelligence venture capital.

**A fact that has to be faced is that in the first two waves of artificial intelligence boom, 90% of the start-up companies lost money. And investors only made money if they entered early. **

A CVC investor said that before the emergence of generative artificial intelligence, people's investment enthusiasm for AI was extremely low, because the commercial performance of these companies was far less than investors' confidence. "A lot of customization, data cleaning and preparation work, a lot of model tuning, almost every business scenario is a non-standard project system, and the industry's talent cost structure is unreasonable, resulting in less than 1% of profitable companies in the last two waves of artificial intelligence."

The current third wave of AI is described by Kai-fu Lee as the progress of "from the isolated island to the mainland". Compared with the previous two waves, this wave of AI makes it possible for GM to build a new world with cross-field capabilities. Once a powerful model is backed by enough data, in suitable scenarios, AI will create productivity that surpasses that of humans.

Of course, in this process, concerns will also exist.

Linear Capital’s concern for the company is that the team is not strong enough to flexibly try and make mistakes in the rapidly changing technological and business environment; the cut-in scenarios are too shallow to form an effective closed loop, and they will fall into red sea competition in the future.

Xiao Miao Langcheng has two concerns: First, the rapid development of open source models and algorithms will lower the threshold for technology acquisition, which will cause the technology investment of top artificial intelligence companies in the past to become invalid investment. Under the fierce competition of homogenization, the business model expected by investors will eventually fail to work.

Secondly, although artificial intelligence has a certain generalization ability at present, to achieve high accuracy, it is necessary to learn and adjust parameters for each scene. Customers have a large demand for customized services. That is to say, even if the main module is common at the model level, there are still a large number of functional plug-ins to be customized, which will eventually lead to startups having to provide customized services, thus falling into the dilemma of being difficult to scale.

In this wave, how do entrepreneurs seize opportunities? Several investors gave advice.

"If there is a creator in the world, he has already issued the order." Zhang Jinjian, founding partner of Oasis Capital, believes that in the big wave, entrepreneurs should actively embrace, not to draw a map, but to wait for the starting gun to sprint and enter the industry as soon as possible.

"Before the industrial revolution, people did not have redundant productivity, so there were no commodities, and there was no commodity circulation. After the industrial revolution, with the circulation of commodities, there was the development of the transportation industry and the retail industry. Now in the era of AI, in theory, the world's top 500 companies can redo it." Zhang Jinjian said.

Fang Zhenghao advises entrepreneurs to make good use of the window of the capital market to complete financing, and at the same time, do not burn money desperately after completing financing, and must make a decision before making a move. "Because the moment for the explosion of artificial intelligence applications has not yet arrived, entrepreneurs need to identify which of them are real opportunities for start-up companies, and then polish their products and businesses to take advantage of this wave of opportunities to go further."

From the perspective of investors, any industry in the world has a development cycle, and real entrepreneurs are the group of people with strong resilience who can withstand the temptation and test in the ups and downs of the industry cycle, and never give up. **