The most fantasy may not be the large models, but AIGC itself.
By Wang Ziwei @ Retail Wei Observation
AIGC, or "Generative AI," is an extremely popular investment field nowadays. Recently, after communicating with a large number of projects and investing in some early-stage projects, I think it is necessary to reflect on and review the journey of the past few months.
The content of this article is only a personal summary and personal opinion, which is inevitably biased and may contain errors. I hope it can stimulate more discussions. Please do not use it as investment advice. DYOR.
[One] Model#
The first investment dimension of AIGC is the model.
The outbreak of Generative AI was accompanied by the emergence of ChatGPT in November 2022. Since then, a large number of large language models (LLMs) have gradually entered the market. There is no doubt that logically, the model is an investment perspective.
However, the problem is obvious: the investment is too large. The cost of a single model training is extremely high, and the talent requirements are also very high. Therefore, in China, it is mainly done by large companies such as Baidu, Alibaba, and iFlytek. Although some non-large companies have provided excellent large models, their valuations have mostly reached billions of yuan, which is not acceptable to ordinary VCs.
In other words, from the current perspective, the investment window for models in China has basically been closed.
[Two] Data#
The second investment dimension of AIGC is data.
Data needs to focus on both quality and quantity. For example, the data sources of Chat GPT include Wikipedia, specific books and journals, selected content from Reddit (WebText), and specific web crawling content (Common Crawl, which is a large dataset collected from websites from 2008 to the present, including raw web pages, metadata, and text extraction. The text comes from different languages and domains). This is a typical case of having both quality and quantity. When combined with excellent large models, it can achieve good results.
For the domestic generative AI industry, companies with unique data sources are worth paying attention to. In other words, if you have data from a specific industry, your model is customized for that specific industry. This is the so-called "one meter wide and one hundred meters deep" logic, focusing on vertical and segmented industries.
Now the question is, where does your data come from?
Public data is good, but please pay attention to the compliance of the data. Otherwise, it will be like "using web crawlers well and eating prison food to the full."
Private data or internal data, where do they come from? This depends on the accumulation of the team. It can be said to be the realization ability of the team's "hidden assets" over the years:
For example, a team of "AIGC + design" comes from a platform that owns a large number of image and video copyrights. However, this design business may ultimately serve consumer product brands that sell products on e-commerce platforms.
Another example is a team of "AIGC + e-commerce" from top e-commerce giants such as Alibaba and JD.com. They may serve enterprise e-commerce customer service, product selection, sample testing, etc., with the ultimate goal of creating a small-scale fast-fashion brand like SHEIN.
[Three] Scale#
The third investment dimension of AIGC is scale.
Scale, or revenue, has always been the most important indicator for investors. Although scale does not necessarily mean a moat, it at least proves that you are a player worth paying attention to.
In fact, truly scalable projects in the field of generative AI are quite rare at the moment. Most projects are still in the early stages and are falling into two traps.
The first trap is project-based. Making money from one project to another is valuable for early-stage teams to create benchmark cases and understand customer needs. It also generates cash flow to support the team (we temporarily ignore accounts receivable). The trap here is how to prove that you are not project-based in the future. From the perspective of VCs, it becomes a troublesome issue once the word "project-based" comes to mind.
The more troublesome trap is the second one. Many entrepreneurs claim that they are not project-based and will move towards subscription + on-demand payment. However, this subscription is essentially mostly SaaS. No matter how you use AI to empower this SaaS, you are still fundamentally a SaaS.
There are two problems with SaaS:
First, SaaS for small and micro enterprises in China is relatively deceptive, which is fundamentally different from the US market. If you are really doing SaaS, please pay attention to various core indicators, such as CAC, LTV, ARPU, NDR, etc., and use these indicators to monitor your development in real-time.
Second, if your SaaS only reduces costs and improves efficiency, I'm sorry, but you will hardly make any money. Only by increasing revenue - yes, it must not be just "improving efficiency" - can you have the possibility of extracting commissions from the increment and doing so in the long term.
In fact, when you can increase revenue, you may not only be a SaaS, but also a BaaS that integrates into business and supply chains. This, in turn, imposes new requirements on your team.
In conclusion, in terms of investment in the AIGC field, the model focuses on technology, and the investment window has almost closed. Data focuses on the team's past accumulation and has opportunities. Scale focuses on the model and must avoid deviating from the right path.
"Retail Wei Observation" takes a global perspective and focuses on the latest strategies, tactics, and thoughts in the field of new retail and new consumption. It conducts in-depth research on super membership systems and domestic and international new retail cases. The founder of the platform, Wang Ziwei, is an independent retail analyst.