What exactly is being built in AI infrastructure? Jensen Huang provided an answer at Davos.

“Is AI just a conceptual bubble, or a real infrastructure investment?”

On January 21, 2026, at a panel discussion on the main stage of the World Economic Forum in Davos, Jensen Huang, the founder of NVIDIA, gave the following answer:

This is the largest infrastructure project in human history.

This “project” involves actual construction, power supply, and recruitment.

Jensen Huang said that AI can be divided into five layers: the bottom layer is energy, followed by chips, cloud services, models, and the top layer is applications. Each layer requires real factories, equipment, electricity, and human resources.

That's why he told all countries: AI is infrastructure. Every country should build it.

So, in this article, we won't discuss macro - trends. We'll just clarify one question: What exactly is being built in AI infrastructure?

Section 1 | The First Thing Is Electricity, Not Chips

In 2025, technology companies around the world were competing for computing power. However, for computing power to function, it must have electricity. Moreover, it requires a continuous, stable, and large - scale supply of electricity.

As a result, the first thing to increase in price was electricity contracts.

This is not a coincidence. Jensen Huang said in Davos:

“AI processes and generates intelligence in real - time, and it needs energy to achieve this.”

This means that even if you have the best models and the latest chips, without a stable and sufficient supply of electricity, AI won't work. It's not just slow; it won't run at all.

The electricity required by AI is different from ordinary electricity. For data centers to train models and perform real - time inferences, they need high - density, low - latency, and year - round uninterrupted power.

This means that building AI is not just about “connecting to power.” It requires building a whole energy supply system: when choosing a location, you need to consider whether the power grid can handle it; power generation must ensure stable output; energy storage should be able to handle peak loads; and the distribution network must withstand continuous high loads.

Jensen Huang didn't use terms like “energy crisis.” He said: We need more energy, land, and data centers. In his view, the largest infrastructure project in human history has begun.

From the United States to the United Arab Emirates, from Southeast Asia to Northern Europe, the first thing in places that want to build AI is to discuss whether there is enough electricity.

What exactly is being built in AI?

The first step is not writing code. It's connecting to power.

Section 2 | Chip Factories and AI Factories Are Already Under Construction

Jensen Huang said:

“We are building chip factories, computer factories, and AI factories around the world.”

He listed several figures: TSMC will build 20 new chip factories globally, and Quanta, Wistron, and Foxconn will build 30 AI computer factories. These factories are not for mobile phone assembly or consumer electronics production. They are specifically for providing production equipment for AI training and deployment.

Chips are for computing, but AI also needs storage. To train a large - scale model, you need to process a huge amount of data, and this data needs a place to be stored. Micron announced an investment of $200 billion in memory production, and Samsung and SK Hynix are also increasing their investment.

That is to say, building AI not only requires electricity. It also needs to produce a complete set of hardware: chips are for computing, memory is for storing data, and computer factories assemble these into AI servers, which are then delivered on a large scale.

This is not just the business of a few companies. It's a global construction wave.

Why is it called a wave?

Because, like the early days of steel, electricity, and railways, it starts with building factories and then develops industries. The concept of AI factories used to sound a bit abstract, but now you can see the complete process of construction, recruitment, completion, power connection, and delivery.

So, how large is the scale? Jensen Huang gave the figure: We have invested hundreds of billions of dollars so far, but this is just the beginning. There are still trillions of dollars worth of infrastructure to be built.

The popularity of AI is not because of “concept hype.” More and more countries and enterprises are investing real money in building factories, buying equipment, and recruiting employees.

From electricity to factories, the AI infrastructure project has moved from the drawing board to reality.

Section 3 | Models Are Just the Fourth Layer, Not the Whole of AI

In the past few years, when people talked about AI, they almost always mentioned models.

Which model is stronger, how many parameters it has, and which one ranks first in benchmarks.

However, in Jensen Huang's view, models are just one layer in the five - layer structure of AI. To be precise, it's the fourth layer. Below it, there are energy, chips, and cloud services to support it; above it, the application layer is where real value is generated.

How to understand this? He used an analogy: In the past, AI models were like an engine on a display stand. It looked beautiful, but you couldn't drive it directly. You had to build the frame, fuel tank, and electrical system first, then install the engine, and finally tune it. Only when it runs on the real road can it be called a product.

The biggest problem now is not that the models are not good enough. It's that many people only see the “engine.”

Jensen Huang sees it differently. He focuses on whether this engine can enter the workshop, go on the road, and be used in industrial scenarios.

Because a model itself is not equal to an application. It only has value when it is put into practice.

What really benefits industries and promotes economic growth is the application layer. That is, above the models, there are products and services that can be applied to specific fields such as healthcare, finance, and manufacturing.

So now, the focus of the industry is shifting: it's no longer about whose model has more parameters, but who can really use AI.

Section 4 | The Explosion of AI - Native Companies: Infrastructure Building Has Just Begun

Jensen Huang presented a set of data in Davos: 2025 was one of the years with the highest investment in the history of venture capital, and a large amount of capital flowed into AI - native companies.

These companies don't build models or design chips. They directly use existing models to do things, such as drug research and development, financial analysis, and re - engineering manufacturing processes.

He used the term “AI - native companies.” It means that these companies design their product processes and business models around AI from the very beginning.

For example, Eli Lilly, a pharmaceutical giant, used to mainly invest its R & D budget in wet laboratories, buying equipment and conducting chemical experiments. But now, they have invested in a large - scale AI laboratory and supercomputer, and they have handed over part of their new drug development process to AI.

Similar changes are happening in many industries.

Robot manufacturing, medical diagnosis, automated trading, customer service systems, compliance review... In these fields that used to require a large amount of human labor, AI - native companies' approach is: take existing models, train them with industry - specific data, make them learn to handle specific tasks, and then turn this ability into directly usable products.

For example, in the past, a customer service system might require hundreds of employees. Now, with an AI - based customer service system that is available 24/7, the cost has been reduced to one - tenth. This is not just a proof of concept; it's a product that is already in large - scale commercial use.

So, what will happen when there are more and more such AI - native companies?

Jensen Huang's answer is: When the upper - layer AI applications explode, the lower - layer infrastructure must keep up.

These companies need to use AI not just for a one - time demonstration. They need it to be stable, cost - effective, and usable on a large scale. This forces the underlying infrastructure to expand:

There must be more electricity supply,

Chips must be available,

Factories must be able to produce on a large scale,

Cloud services must be able to handle the load.

Jensen Huang said: We have just started building the foundation of AI. It's not that the industry is not hot. It's that companies using AI are just emerging. The role of infrastructure is to enable these companies to really use AI.

Section 5 | Who Is Involved: The Roles of Labor and Countries

After discussing what is being built, let's look at who is building it.

The answer may surprise many people: First of all, it's plumbers, electricians, and steelworkers.

In the United States, these types of workers involved in the construction of chip factories, computer factories, and AI factories are in short supply. Jensen Huang gave a figure on - site: Their salaries have almost doubled in a short period of time, and some people's annual salaries have exceeded six figures. Most importantly, these jobs don't require a computer science doctorate. They require people who can do hands - on work, on - site construction, and operate equipment.

For many countries, this is an opportunity for blue - collar workers to become part of the middle class again.

So, when AI is really put into use, how will it affect those who are already in the workforce?

Many people's first reaction is: They will be replaced. But the reality is the opposite.

For example, radiologists. In the past, they were predicted to be one of the first occupations to be replaced by AI. But after 10 years, AI has deeply penetrated radiology departments, and the number of doctors has increased.

Why?

Because after AI takes over the repetitive and mechanical task of reading images, doctors can spend more time talking to patients and making comprehensive judgments. The hospital's patient reception volume has increased, its revenue has risen, and the number of doctor positions has also increased.

Another example is nurses. Many AI tools are helping them with complicated tasks such as document recording and medical transcription. As a result, nurses spend more time taking care of patients instead of filling out forms. The patient turnover rate has increased, and the hospital has hired more people.

Jensen Huang summarized: AI replaces tasks, not jobs.

As long as your job is not purely mechanical and repetitive, but requires judgment, interaction with people, and creativity, AI is a helper, not a competitor.

Besides individuals, what about countries?

Jensen Huang emphasized the right to participate. In the past, many developing countries thought AI was too far away. But he provided a local - based starting plan:

Open - source models are already very powerful;

Many countries can fine - tune these models with local languages and local knowledge;

You don't have to start from scratch, but you must participate in the construction.

AI should be like electricity and roads, the infrastructure of every country. This is for all developing countries.

If in the previous technological revolutions, they started in Silicon Valley and the West and then spread to other countries, then in this round of AI, the right to partial participation has been open from the beginning.

You don't have to build models first, but you can use them first; you don't have to understand chips, but you can build AI applications first.

Conclusion | Shortage, Not a Bubble

It's very difficult to rent a GPU.

The spot price is still rising, not just for the latest ones, but also for those two generations old.

Jensen Huang said:

“Bubbles don't cause price increases. Shortages do.”

Pharmaceutical companies are starting to invest in AI laboratories. Countries are competing for electricity and land. Investment institutions are looking for AI infrastructure projects.

What exactly is being built in AI?

Energy, chip factories, data centers, model layers, and application layers.

This is the answer Jensen Huang gave in Davos.

📮Original Article Links:

https://www.youtube.com/watch?v=hoDYYCyxMuE

https://www.weforum.org/stories/2026/01/live - from - davos - 2026 - what - to - know - on - day - 3/

https://blogs.nvidia.com/blog/davos - wef - blackrock - ceo - larry - fink - jensen - huang/

This article is from the WeChat official account “AI Deep Researcher”. Author: AI Deep Researcher. Republished by 36Kr with permission.