Huang Renxun uses the "five-layer cake" analogy to explain the evolution of artificial intelligence

NVIDIA founder Jensen Huang introduced the concept of the “Five-Layer Cake” to explain what artificial intelligence really is. He used a creative storytelling approach to help the public understand that AI is not just a chip, model, hardware device, or application, but a complete intelligence system that develops from the energy foundation up through chips, infrastructure, models, and applications.

What does Jensen Huang mean by the AI Five-Layer Cake architecture?

He breaks down AI into five stacked layers, starting with Energy at the bottom. The essence of real-time generated intelligence is energy conversion; power supply is the fundamental constraint on AI output. Above energy are AI Chips, which efficiently convert electricity into computational power. Their bandwidth and interconnect technology determine the cost and scale of AI production. The third layer is Infrastructure, including data centers, cooling systems, and network engineering—these are called AI Factories. Their purpose is not just data storage but digesting and understanding vast amounts of data to enable AI learning. The fourth layer is Models, which handle language and data, and delve deeper into fields like biology, chemistry, and physics. At the top are Applications, such as autonomous driving, robotics, and biotech/medical platforms, representing the highest value realization of AI in the economy.

How is AI born?

Huang explains that in the history of computing, software was essentially pre-recorded—humans wrote algorithms, structured data carefully, stored it in tables, and retrieved it precisely with SQL. It was an era of “instructions first, data constrained.”

But AI has completely broken this framework.

Humans now have computers capable of understanding unstructured information. They can recognize images, read text, listen to sounds, and understand their meanings. They can also reason about contexts and intentions. More importantly, they can generate intelligence in real time. Every response is newly generated, and each answer depends on the human-provided context. This is no longer software retrieving from pre-stored instructions; it is software reasoning in real time, generating AI tailored to human needs.

The first computers humans possessed were no longer just rigid machines but “brains” capable of understanding unstructured information. They can read images and text, listen to music, chat with people, and even infer answers to human needs.

Energy drives AI computing power

The foundation of the AI architecture is Energy. Real-time AI generation requires electricity. Each token produced is the result of converting electrical energy into computational capability. Energy is the first principle of AI infrastructure and the fundamental limit on how much AI a system can produce.

AI Models help humans digest vast amounts of information

AI models can understand many types of information, including language, biology, chemistry, physics, finance, medicine, and the real world itself. Large language models are one example. The most transformative developments now include protein AI, chemistry AI, physical simulations, robotics, and autonomous systems. AI helps humans understand complex information faster. For example, in radiology, AI-assisted image interpretation allows doctors to focus more on patient care and clinical judgment, reducing workload and improving healthcare quality.

AI factories initiate the largest global infrastructure transformation

AI is sparking an unprecedented wave of infrastructure development. While its core competitiveness lies in complex algorithms, the backbone of this digital revolution is massive physical construction—from data centers to AI factories. Building these hardware facilities has created a surge in demand for skilled technicians, making advanced technical labor an essential pillar of this technological wave.

Top-tier AI applications bring intelligence into the physical world

At the top of the AI Five-Layer Cake, AI translates technology into tangible economic value through diverse applications. Development has expanded from simple text conversations to physical applications in the real world. For example, in biotech, AI-driven drug discovery platforms can precisely simulate protein structures; in transportation, autonomous driving technology embodies AI in vehicles; humanoid robots can perform tasks directly in factories or homes. These applications are driving a global industrial transformation.

This article, Jensen Huang’s “Five-Layer Cake” analogy explaining the evolution of AI, originally appeared on Chain News ABMedia.