From "Lobster Party" to "Five-Layer Cake": How does Jensen Huang plan to mine the AI gold mine?

Reporter Ni Yuqing from 21st Century Business Herald

In the past two years, the wave of generative AI has swept the globe, leading to a surge in computing power demand, with GPUs becoming the hottest resource in the entire industry chain.

As a core driver of this round of AI infrastructure revolution, NVIDIA’s founder and CEO Jensen Huang and his company are almost at the most critical position in the AI industry.

However, as the supply of computing power continues to expand, a new question has gradually emerged: with more and more “shovels,” where is the real “gold mine”?

Recently, Jensen Huang rarely published an article, using the metaphor of the “AI five-layer cake,” breaking down the AI system into five levels: energy, chips, infrastructure, models, and applications.

(Image source: NVIDIA)

This seemingly simple metaphor actually reflects his rethinking of the structure of the AI industry. AI is not just a competition of models; it is also a system engineering that covers the entire technology stack.

It is also noteworthy that, in addition to emphasizing computing power and infrastructure, Jensen Huang has begun to frequently discuss the importance of the application layer. Recent reports indicate that NVIDIA is preparing to launch an open-source AI Agent platform called NemoClaw, primarily aimed at the enterprise software ecosystem. Currently, NVIDIA has not responded to inquiries about NemoClaw, and Jensen Huang will unveil the latest chips and everything related to AI at NVIDIA’s annual conference GTC next week.

We can see that after the explosive popularity of OpenClaw (nicknamed “Lobster”), NVIDIA also wants to join the “Lobster party,” targeting AI Agents. As the supply capacity of AI continues to improve, only by continuously emerging new application scenarios can the entire industrial ecosystem continue to expand.

From the AI “Lobster” to the “five-layer cake,” and then to the focus on the application ecosystem, Jensen Huang is telling a new story for the AI industry: how to continue discovering new veins above the already exploited gold mine.

Five-layer Cake: Redefining the Structure of the AI Industry

As a central figure in this wave of computing power, Jensen Huang is continually updating his description of the structure of the AI industry. From last year’s three developmental stages of AI to the recently proposed “five-layer cake” theory, he is essentially trying to answer the same question: where does the value of the AI industry actually come from, and how will it continue to expand?

The so-called “AI five-layer cake” is a vivid metaphor used by Jensen Huang to describe the technology stack of the AI industry. According to this framework, the AI system can be roughly divided into five levels: the bottom level is energy, followed by chip computing power, then data centers and cloud infrastructure, followed by AI models, and at the top level is applications. The five-layer structure together constitutes the complete technological system of the AI industry.

This theory may seem like a summary of technological architecture, but it means that AI competition is not just a struggle over single technologies; it is a competition at the ecosystem level.

In the current AI industry, NVIDIA undoubtedly occupies the core position at the bottom level. GPU computing power has become the infrastructure for AI training and inference, and the CUDA software ecosystem has further solidified this advantage. However, as the AI market rapidly expands, NVIDIA also has its anxieties.

A senior software expert told reporters from 21st Century Business Herald that in the past internet wave, infrastructure providers only needed to provide tools, and application innovations would quickly emerge; however, in the AI era, the application ecosystem has not appeared in sync.

This is precisely why Jensen Huang has been emphasizing the importance of the application layer recently. In other words, while the supply of AI computing power is becoming increasingly strong, the actual application scenarios that can consume this computing power are still limited.

In the gold rush era, those selling shovels did not need to prove that gold mines existed, as miners would seek them out themselves. But when the production capacity for “shovels” is strong and creates a larger industry scale, those selling shovels need to stand up and tell the market—there is still a gold mine here.

In a sense, the “AI five-layer cake” is completing such a narrative reconstruction. By breaking down the AI industry into multiple levels, Jensen Huang is actually emphasizing that AI does not solely belong to model companies or application developers; it is a vast ecosystem that requires participation from multiple parties.

More importantly, this layered design is also redistributing the imaginative space of the AI industry. In the traditional software era, value was mainly concentrated at the application layer; whereas in the AI era, energy, computing power, and infrastructure have also become new centers of value. Through this narrative, the “cake” of the AI industry is being enlarged, and not only model companies and application vendors have the opportunity to profit, but infrastructure providers are also key beneficiaries.

In other words, the proposal of the “five-layer cake” is essentially searching for a new growth narrative for the AI industry.

As AI computing power has become a reality, the next critical question is no longer whether technology can be realized, but how the entire ecosystem can continue to expand. Only when more enterprises and developers believe that there is still a huge “gold mine” in the AI industry can the entire computing power economy continue to operate.

NVIDIA Constructs a “Three-layer AI Matrix”

If the “AI five-layer cake” describes the industrial structure, then NVIDIA’s own strategy resembles establishing a solid “power structure” within this cake. In recent years, NVIDIA has been gradually forming a clear three-layer layout: computing power, infrastructure, and application ecosystem.

The first layer is computing power dominance. As the main supplier of GPUs, NVIDIA has become the core infrastructure for AI training and inference globally. AI chips continuously enhance performance, while the CUDA software platform has formed a very high technical barrier. This layer corresponds to the chip layer in the “AI five-layer cake” and is also the most profitable part for NVIDIA.

The second layer is AI infrastructure. As the scale of AI models continues to expand, a single GPU can no longer meet the demand, and data center-level computing clusters have become the new form of infrastructure. NVIDIA proposed the concept of “AI Factory,” launching NVLink, NVSwitch, and super-node architecture, integrating GPUs, networks, and software into a complete computing system. This means that NVIDIA is no longer just selling chips; it is providing a complete AI production line.

Jensen Huang emphasized: “AI is an industrial-level transformation; it will change the way energy is produced, the way factories are built, the way work is organized, and the mode of economic growth.” He believes that AI is still in its early stages, with much infrastructure yet to be built and many talents still untrained; AI is becoming the infrastructure of the modern world.

Looking to the future, the third layer is particularly noteworthy: applications and ecosystems. In Jensen Huang’s five-layer cake diagram, the area occupied by applications is the largest.

For a long time, NVIDIA’s ecosystem has mainly focused on developers and large enterprise clients, constituting a typical B-end ecosystem. In contrast, technology companies like Apple have built a broader industrial network through their application ecosystems. Apple not only sells hardware but, more importantly, creates a developer economy through the App Store, ensuring the continuous prosperity of the entire platform.

This is precisely the key issue NVIDIA currently faces. The supply of computing power is extremely strong, but application scenarios are still insufficient. If AI technology cannot be transformed into broader business opportunities, the demand for computing power will ultimately encounter a ceiling.

Therefore, NVIDIA has begun to take a more proactive role in building the application ecosystem. The recent market reports about the AI Agent platform NemoClaw plan are seen as an important signal of NVIDIA extending into the application layer. By providing development frameworks and platform tools, and supporting external chip platforms, NVIDIA hopes to enable more enterprises to build their own business systems based on AI.

In recent discussions about the future of the application layer, he further predicted that with the development of AI Agents, traditional software and APP forms may be reshaped, and users may directly call various digital services through AI assistants.

NVIDIA’s role is not just that of a “shovel seller” in this era. By constructing a complete system from computing power to applications, it aims to become the foundational platform for the entire AI ecosystem.

From the AI “five-layer cake” to the exploration of the application ecosystem, Jensen Huang is constantly reshaping the narrative of the AI industry. The underlying goal is quite clear: to ensure that the AI industry not only has strong supply capabilities but can also continue to create new demand.

Only when more enterprises and developers can truly dig up “gold mines” in this land will the AI computing power economy continue to expand, and NVIDIA will continue to occupy a core position within it.