For a company, the success or failure of the company He Tuber mostly depends on the founder, but this is not the case for AI-type companies. In this article, the author talks about the situation after the rise of AI from the perspective of computing models, and looks at what determines the success or failure of AI companies.

I actually wrote several articles about products and technologies before this etc., but the underlying logic is not thorough enough. This article discusses the previous articles from the perspective of computing models. The underlying logic of some judgments may be a little obscure, but if you really understand it, it should be enlightening.

At today's time point, what is the distribution of the entire calculation like?

The operating system provides the most basic computing abstraction. Various applications use the computing power of the hardware through the operating system . Therefore, the operating system is the core node of the industry and has the largest application scope. Who currently has the dominance at this node? Apple, Google, Microsoft and the open source community (please compare the domestic situation yourself).

This is a situation in the end, and the cloud is a different situation. This is often mentioned and I will not elaborate on it.

Both operating systems and cloud services are abstract methods for basic hardware, but this abstraction alone is not enough. To provide this computing power to users, applications are also needed.

The closer it is to the users, the closer it is to the benefits, so almost all companies that have control over the underlying computing will have their own applications, and they will stick there where they are the most versatile and the most rewarding . Typical ones are Windows and Office. Even after Cloud First, Microsoft adopts an infrastructure + application model.

On top of this are various applications that are vertical in various fields, such as search, IM, etc. for individual users, such as SAP's ERP for enterprise users, and various industry software for different industries.

If you look at it from an endpoint of view, it is a general computing platform covered with large and small pillars. Each different pillar represents different benefits, and there will be fierce competition between the most common parts used by everyone.

From a cloud perspective, applications come first, but applications and infrastructure are integrated into one . Only by being dominant in one of the most common applications can the advantages of infrastructure be brought into play, such as search and e-commerce.

If the initial PC applications were focused on the terminal, and the intermediate Internet applications were focused on the cloud, such as search, then now it has basically reached a hybrid state, such as Douyin . The interesting thing here is not the ratio of cloud and computing, but that the toll of general computing is actually controlled on the terminal . The various "Apple taxes" during transactions in various app stores are actually controlled on the terminal. Control.

The basic feature is a subject that provides computing infrastructure and has a high probability of making the most valuable applications. The further back it goes, the more likely it is to be a subject.

Based on this computing model, we can imagine what will happen after the rise of AI.

After the rise of AI, the model of outputting computing power based on API (Application Programming Interface) will definitely be overturned. In the picture above, it is the base part that needs to be replaced.

That means there needs to be a new package that becomes a middle layer that universally provides computing power.

From a computational perspective, the positions of large model companies and operating systems are equivalent.

This kind of provision will obviously have to be on the cloud in the early stage, but it will gradually transition to the end in the later stage.

Combining the above mentioned, you will find that there is a very magical topic here:

Is ecological control really only in the end? Can we only build super applications on the cloud, but not have ecological control over computing?

The reason for the magic is: if the answer is yes, then the large model will be merged with the robot operating system; if the answer is no, then a real cloud operating system will be produced.

If you make an analogy, you will find that a large model company will inevitably follow a similar path to Microsoft in business. It cannot only make Windows but not Office, or Google can only make Android and not Google Play, etc. The most typical applications for the public must be It will be classified as a large model company. Otherwise, it will be like simply building roads by yourself, but there will be no toll stations, which is not commercially viable.

It is still difficult for us to predict what Office will be in the big model era. If it is similar to a smart speaker or Siri, it will be a new super application, where the provider of computing power and the provider of super applications are one.

Even when Microsoft was at its peak, it did not have a big impact on companies like SAP. Why?

Because there is domain knowledge.

SAP actually encapsulates computing capabilities with business attributes for the domain. On top of it are finance, HR, supply chain, CRM and other modules, and it also provides interfaces for secondary development.

Unless you are super intelligent, you can ignore this kind of domain knowledge, otherwise it is a natural boundary between different companies.

What kind of structure is this? General Big Model will provide new computing bases and will most likely take away one or more super applications. But at the same time, new applications vertical to a certain field will grow on this general computing base. And this large vertical application is most likely a fractal structure.

(The picture above is from an article by Chen Guo on ERP. If the operating system abstracts the von Neumann structure, then this re-platforming actually abstracts the domain model )

The difference is that because the intelligence density of many gadgets is too small, they may not exist independently. Only tools that make a huge difference with AIGC, calculation and other tools may exist independently. (It is difficult to have a separate application behind the calculator)

For these applications, human efforts can determine whether you are successful, but it cannot change the cycle of the application and the size of the corresponding pattern.

For example, Nvidia, what analogy can be made?

When the PC rose, the first ones to rise were actually various computer companies. At that time, they were IBM, HP, and Compaq abroad, and Lenovo, Great Wall, etc. domestically. Now it seems difficult to imagine that these companies were once the protagonists of an era, but they indeed were. At that time, a 286 with less than one-tenth of the computing power of today’s computers sold for 20,000 to 30,000 yuan, which corresponds to the housing price in Beijing. That's only three thousand per square meter.

In the process of determining this calculation pattern, from an enterprise's perspective, the intrinsic composition of business value is constantly changing.

Initially, technical value or application value is usually output, and only in some special occasions can the application value be transformed into ecological value. The sustainability of ecological value is higher than the first two.

I need technology, but I don’t know how to do it or don’t want to do it, so I use yours. This is the value of technology. When IBM used MS-DOS, it used this technical value to a certain extent. Algorithm authorization is even more so. The current large model is actually at this stage.

I need to solve something and you can provide it, then you create application value, such as PS. Search actually has application value, but this application is too big.

When iOS reaches the point where it can collect Apple tax, it actually changes from simply providing technology or functions to ecological value.

Once the ecological value is established, the business model can achieve stability far greater than the state of technology and application.

But this ecological value position is actually innate.

AI has the potential to give birth to new ecological value positions. GPTs is actually one, but it is obviously more than that.

Different people have different technologies and financial resources that they can deploy, and their starting points are different, so the models they can use to form their own cash flow are different.

When facing the new field of artificial intelligence, it is actually a bit like Qin losing its deer, and the whole world is chasing it. Who can finally establish the ecological value of computing is actually a topic worthy of imagination.

It is actually a bit problematic to focus on the involution of the model at this time. Of course, it needs to improve, otherwise nothing can be done.

But the idea of ​​​​involution is very similar to Microsoft saying that I will only work on Windows, and Office can work on whoever they like.

Ecological value is actually a process of spiral amplification, which is very challenging for traders. And this spiral amplification is actually the growth of ecology. The road must be full of both wants and needs.

Sam Altman seems to be thinking this way.

From this perspective, the recent speech of a well-known domestic large-scale model entrepreneur is very problematic.