Nvidia's Jensen Huang latest article: The "Five Layers of the AI Cake"

Author: Huang Renxun

Compiled by: Peggy, BlockBeats

Editor’s note: Artificial intelligence is gradually evolving from a cutting-edge technology into a fundamental infrastructure supporting the operation of the modern economy. In its first long-form article published on its official account, NVIDIA attempts to systematically analyze the AI industry structure from first principles: from energy and chips, to data center infrastructure, to models and applications, forming a complete five-layer technology stack.

The article points out that AI is not just a competition of software or models, but a global industrial development involving energy, computing power, manufacturing, and applications. Its scale could become one of the largest infrastructure expansions in human history. Through this “five-layer cake” perspective, NVIDIA aims to illustrate that the true significance of AI is not just smarter software, but an infrastructure revolution comparable in scale to electricity and the internet.

Below is the original text:

Artificial intelligence is one of the most powerful forces shaping the world today. It is not just a clever application or a single model, but an infrastructure as important as electricity and the internet.

AI operates on real hardware, real energy, and real economic systems. It transforms raw materials into scalable “intelligence.” Every company will use it, and every country will build it.

To understand why AI unfolds in this way, starting from first principles and examining the fundamental changes in the computing field can be very helpful.

From “Pre-made Software” to “Real-time Generated Intelligence”

For most of computing history, software has been “pre-made.” Humans describe an algorithm first, then the computer executes instructions accordingly. Data must be carefully structured, stored in tables, and retrieved through precise queries. SQL is indispensable because it enables this entire system to operate.

But AI breaks this pattern.

For the first time, we have a computer capable of understanding unstructured information. It can interpret images, read text, listen to sounds, and understand their meanings; it can infer context and intent. More importantly, it can generate intelligence in real time.

Each response is a new generation. Every answer depends on the context you provide. This is no longer software retrieving existing instructions from a database; it is software reasoning in real time and generating intelligence on demand.

Because intelligence is generated in real time, the entire computing technology stack supporting it must be reinvented.

AI as Infrastructure

From an industry perspective, AI can be broken down into a five-layer structure.

Energy

The bottom layer is energy.

Real-time generated intelligence requires real-time electricity. Every token produced involves electrons moving, heat being managed, and energy being converted into computational capacity.

Below this layer, there is no abstraction. Energy is the first principle of AI infrastructure and the fundamental constraint on how much intelligence the system can produce.

Chips

Above energy are chips. These processors are designed to convert energy into computing power with extremely high efficiency under large-scale conditions.

AI workloads demand massive parallel computing, high-bandwidth memory, and high-speed interconnects. Advances in chip technology determine the speed of AI expansion and how cheap “intelligence” ultimately becomes.

Infrastructure

Above chips is infrastructure. This includes land, power transmission, cooling systems, construction, networking, and the scheduling systems that organize tens of thousands of processors into a single machine.

These systems are essentially AI factories. They are not designed for storing information but for manufacturing intelligence.

Models

Above infrastructure are models. AI models can understand various types of information: language, biology, chemistry, physics, finance, medicine, and the real world itself.

Language models are just one category. Some of the most transformative work is happening in fields like protein AI, chemical AI, physics simulation, robotics, and autonomous systems.

Applications

At the top is the application layer, where real economic value is created. Examples include drug discovery platforms, industrial robots, legal copilots, and autonomous vehicles.

An autonomous vehicle is essentially an “AI application carried by a machine”; a humanoid robot is an “AI application carried by a body.” The underlying technology stack is the same; only the final form differs.

Thus, the five-layer structure of AI is: Energy → Chips → Infrastructure → Models → Applications. Every successful application influences all layers downward, until the power plant at the bottom supplies energy.

An Infrastructure Construction Still in Its Early Stages

We are just beginning this construction. Currently, investments amount to only hundreds of billions of dollars, but in the future, trillions of dollars of infrastructure will need to be built.

Globally, we are seeing the emergence of chip factories, computer assembly plants, and AI factories.

Unprecedented scale is being built. This is becoming one of the largest infrastructure projects in human history.

Labor Demands in the AI Era

The scale of workforce needed to support this construction is enormous.

AI factories require: electricians, plumbers, pipefitters, steelworkers, network technicians, equipment installers, operations and maintenance personnel.

These are highly skilled, well-paid positions, and there is currently a severe shortage. Participating in this transformation does not necessarily require a PhD in computer science.

Meanwhile, AI is driving productivity improvements in the knowledge economy. Take radiology as an example. AI has begun assisting in medical image interpretation, yet the demand for radiologists continues to grow.

This is not contradictory.

The true role of radiologists is patient care, and image reading is just one task. As AI takes over more repetitive tasks, doctors can spend more time on diagnosis, communication, and treatment.

Hospital efficiency improvements can serve more patients, which in turn requires more staff. Productivity creates capacity, and capacity drives growth.

What Has Changed in the Past Year?

In the past year, AI has crossed a critical threshold.

Models are now sufficiently advanced to truly perform in large-scale scenarios.

• Significant improvements in reasoning ability
• Marked reduction in hallucinations
• Substantial enhancement in “grounding” with the real world

For the first time, AI-based applications are beginning to generate real economic value.

Clear product-market fit has emerged in fields such as drug discovery, logistics, customer service, software development, and manufacturing.

These applications are strongly driving the entire underlying technology stack.

The Role of Open-Source Models

Open-source models play a key role. The vast majority of AI models worldwide are free. Researchers, startups, enterprises, and even entire nations rely on open-source models to participate in advanced AI competition.

When open-source models reach the cutting edge of technology, they not only change software but also activate demand across the entire tech stack.

DeepSeek‑R1 is a typical example. By making a powerful reasoning model widely available, it accelerates application layer growth and increases demand for training compute, infrastructure, chips, and energy.

What Does This Mean?

Viewing AI as infrastructure clarifies everything. AI may have started with Transformers and large language models, but it is much more than that.

It is an industrial-scale revolution that will reshape:

• The way energy is produced and consumed
• The way factories are built
• The organization of work
• The pattern of economic growth

AI factories are built because intelligence can now be generated in real time. Chips are redesigned because efficiency determines the speed of AI expansion. Energy becomes central because it determines how much intelligence the system can produce. Applications explode because models have finally crossed the “scalability” threshold.

Each layer reinforces the others.

This is why the scale of this construction is so enormous, why it impacts so many industries simultaneously, and why it will not be confined to any one country or field.

Every company will use AI.

Every country will build AI.

We are still in the early stages.

Much infrastructure remains to be built, many workers need training, and many opportunities are yet to be realized.

But the direction is very clear.

Artificial intelligence is becoming the foundational infrastructure of the modern world.

And the choices we make today—speed of construction, breadth of participation, and responsibility in deployment—will determine what this era ultimately becomes.