What do you think of the yellow Ai five story cake?

English | Chinese Original

English translation of the Chinese original. This version is generated for international readers and may be refined over time.

Date: 2026-03-11

Yesterday, March 10th, Wong published an article called "The Five Layers of Cake ", which says that AI is not just a single application, it is not a single model, but an infrastructure for the Internet, as essential as electricity.

Original Linkhttps://blogs.nvidia.cn/blog/ai-5-layer-cake/

Yellow has analyzed the structure of AI from five levels, driven the core chip from the bottom of the energy source, built the infrastructure of the AI factory, supported models that understand the world and eventually generated applications at the top.

Huang said that in computer development history, software is pre-engineered in computers, and if you run software computers, you'll be able to open a space where your software can run. Data will exist in a database, and you have to write SQL statements to store them, and these operations are structured.

And AI broke this pattern. AI has built a system that identifies images, reads text, listens to voices and understands their meaning, makes decisions based on context and intent, and develops real-time intelligence.

There is no standard answer for each decision.

Each response depends on your context, which is not a command for software retrieval storage, but for software to reason on demand and generate intelligence.

Therefore, the entire computing architecture must be redesigned.

And there are five levels in which to build this architecture.

** Energy chips, infrastructure, models, applications**

The framework proposed by Wong can be described as the most recent of the clear values of AI industry:

The five layers of the model, which essentially put AI from the "technology concept" to the "new infrastructure", were inspired by three judgements:

• Bottom-based binding shifts: A core bottlenecks in AI ' s calculus have shifted from chip production to energy supply and infrastructure landing capacity, and the value of energy, cold stations and IDC industrial chains will be revalued over the next few years.

• Value inverse transfer: the growth of the upper open-source model will boost demand at the bottom, which is why Yveda will be able to sustain the open-source ecology - - Whoever made the top model needed token, needed to figure it out.

• Aberrant knowledge of employment: AI, on the other hand, expands the size of industries, such as radiology. AI, on the other hand, increases the volume of services provided by video sections, and doctors are able to handle more high-value cases.

And I think that maybe in the next three to five years, the AI industry's opportunities are going to be "back-to-back" driven, first by a bottom-up energy, calculator, infrastructure, and then by a scale-down of top-level applications.

We used to squeeze our heads for models and applications, and now we find that the most stable are shovels or even electrics? The recent installation and unloading of OpenClaw is good proof of this.

The middle layer connecting the top and bottom is the most profitable.

For practitioners, rather than crowding into model rolls, it would be better to move up and down in two directions: either to go to the bottom or to go to the vertical industry for application.