Is AI not taking jobs but rather talent? Jensen Huang makes his debut at Davos: It has sparked the largest infrastructure wave in human history

On January 21st at the World Economic Forum in Davos, Jensen Huang, the CEO of NVIDIA, made his first appearance for a dialogue.

This time, Jensen didn't confine the topic to NVIDIA's own GPUs, computing power platforms, or network architectures. Instead, he rarely stepped out of NVIDIA and explained these things from a more macroscopic perspective:

What changes are taking place in AI technology itself

How the AI industry is actually built

How AI may impact society and employment

In Jensen's view, the AI industry system can be divided into five layers, namely energy, chips and computing infrastructure, cloud infrastructure and cloud services, the AI model layer, and the application layer above the models.

What truly determines whether AI can become productive forces and contribute to economic growth is the application layer. He said bluntly:

"Last year was an extremely astonishing year for AI. Frankly speaking, it's because the models advanced so rapidly that the layer above, the application layer, which is what all of us really need to succeed with, began to boom."

In the AI model layer, there were three "disruptive" events in 2025:

Agentic AI

Breakthroughs in open - source models

Great progress in physical AI

What's even more interesting is that in the face of the widespread anxiety about "whether AI will replace jobs", he gave an almost completely opposite judgment, believing that AI may not bring about a wave of unemployment; instead, it will lead to a "shortage of people":

"This is the largest - scale infrastructure construction in human history, and it will create a large number of jobs... We need plumbers, electricians, construction workers, steelworkers, network technicians, and people to install and deploy equipment.

In the United States, the salaries of these positions are already close to or exceed six figures. You don't need a computer doctorate to have a decent career."

The following is the content of this interview. InfoQ edited it without changing the original meaning.

This time is different: AI is completely reconstructing the entire computing logic

Host: Most of today's discussions around AI focus on how it will change the world and the global economy. But I'd like to approach it from another angle: Is AI just a way to change the world, or can it make the entire economic pie bigger? And can more people really use it and keep up, rather than just a few benefiting?... So, let's get straight to the point: Why do you think AI has the potential to be such an important growth engine? How is this technological wave different from past technological cycles?

Jensen Huang: First, when you think about AI and interact with it in various ways - for example, of course, you use ChatGPT, of course, you use Gemini, and of course, you use Claude from Anthropic. The "miraculous" things it can do are amazing.

But if we go back to first - principles to understand what's happening to the computing stack, things will become clearer: This is a platform shift. A platform is something on which applications are built.

Throughout history, every platform shift has brought about a new application ecosystem:

From mainframes to PCs

From PCs to the Internet

From the Internet to mobile and cloud computing

In each platform shift, the computing stack is reinvented, and new applications are created.

AI is also a platform shift. ChatGPT itself is just an application. More importantly, countless new applications are being built on it, and will also be built on Claude and other large models.

Traditional software is essentially like "pre - recorded": humans input, write algorithms or "recipes", and let the computer execute them.

It can handle structured information - that is, you must first organize information such as names, addresses, accounts, ages, and residences into a structured table, and then the software can retrieve it. We call it SQL, SQL queries. SQL may be the most important database engine in the world. Until now, almost everything has run on SQL.

Now, for the first time, we have a computer that is not "pre - recorded" but processes in real - time and can understand unstructured information. Specifically:

It can understand unstructured information (text, images, sounds)

It can understand semantics and context

It can infer your intentions in a real - time environment, and these intentions can be expressed in a very unstructured way

You can describe them however you like.

We call this kind of input a prompt. As long as it understands your intention, it can perform tasks for you. This is no longer "pre - written software" but a system that generates intelligence in real - time.

The AI industry system has "five layers", and the application layer is the most important

Since we are reinventing the entire computing stack, people will ask "What is AI?" Many people think of models directly when talking about AI.

But from an industry perspective, AI is actually more like a five - layer cake:

The bottom layer is energy. AI processes and generates intelligence in real - time, and it needs energy.

The second layer is where I'm involved: chips and computing infrastructure.

The third layer is cloud infrastructure and cloud services.

The fourth layer is AI models - most people think AI is just this layer. But don't forget that models can exist because all the layers below exist.

And the most important layer, the one that is currently evolving, is the application layer above the models.

Last year was an extremely astonishing year for AI. Frankly speaking, it's because the models advanced so rapidly that the layer above, the application layer, which is what all of us really need to succeed with, began to boom.

This application layer can be in fields such as financial services, healthcare, and manufacturing. Ultimately, it's this layer that generates economic benefits.

But the key is that this computing platform needs the support of all the layers below. Because these five layers must be built simultaneously, we are experiencing the largest - scale infrastructure construction wave in human history.

So far, we've only invested tens of billions of dollars, while the real investment needed is in the trillions of dollars range.

Chip factories, computer factories, and AI factories are being built globally in sync:

TSMC announced the construction of 20 new wafer fabs

Foxconn, Wistron, and Quanta are building dozens of computer factories

Storage manufacturers (Micron, SK Hynix, Samsung) are expanding comprehensively

You can see that the entire "chip layer" is growing very fast today.

Of course, we also pay close attention to the model layer.

But what's even more exciting is that the application layer above the models is performing very well.

There's an intuitive indicator: Where is the venture capital money flowing? In 2025, it was one of the years with the largest venture capital investment in history, and most of the funds were invested in so - called AI native companies.

These companies come from all major global industries such as healthcare, robotics, manufacturing, and financial services. You're seeing a large amount of investment flowing into these AI native companies because, for the first time, the models are good enough to build products and businesses on them.

There were "three major events" in the model layer last year.

Host: Let's dig deeper. Obviously, I believe everyone here has their own chatbots to get information. You just mentioned that the "dispersion of AI" will be the key. Let's talk about its dispersion in the physical world - you mentioned that healthcare is a good example. In fields such as transportation and science, what opportunities do you think will bring about disruptive changes?

Jensen Huang: Last year, I think three major events took place in the AI technology layer, especially in the model layer.

First: Agentic AI. At first, the models were just "curious and interesting" but had a lot of hallucinations. Last year, we could relatively reasonably accept that these models became more "solid" and "well - founded". They can conduct research, reason about situations they haven't been trained on, break down problems into step - by - step reasoning steps, form plans, and then answer your questions, conduct research, or perform tasks. So last year, we saw language models evolve into what we call agentic systems, that is, Agentic AI.

Second: Breakthroughs in open - source models. A few years ago - or was it a year ago? - When DeepSeek came out, many people were very worried. Frankly speaking, DeepSeek was a big event for most industries and companies around the world because it was the world's first open reasoning model. After that, a large number of open - source reasoning models emerged one after another. Open - source models allow companies, industries, researchers, educators, universities, and startups to use these open - source models as a starting point to create something - to develop models and systems tailored to their own fields.

Third: Great progress in physical intelligence, or physical AI. That is, AI can not only understand language but also understand "nature" and the physical world: understand proteins, chemical substances; understand nature and physics - such as fluid mechanics, particle physics, and quantum physics. These AIs are learning various structures and "languages". In a sense, proteins themselves are a kind of language. The progress of these AIs is so rapid that industrial companies in manufacturing, drug discovery, etc. are making significant breakthroughs.

One strong indicator is our cooperation with Eli Lilly. They realized that AI has made great progress in understanding protein structures and chemical structures - in a sense, we will be able to "talk to proteins" just like we talk to ChatGPT. We will see very significant breakthroughs.

AI doesn't take away jobs; instead, it will lead to a "shortage of people"

Host: These breakthroughs have also raised concerns about "people". You've talked to me many times, but today you must tell everyone here: People are generally very worried that AI will replace jobs. And you've always been saying the opposite. Do you really think we'll face a labor shortage? How do you view AI and robotics changing the nature of work rather than eliminating jobs?

Jensen Huang: There are several ways to analyze this.

First, this is the largest - scale infrastructure construction in human history, and it will create a large number of jobs. Even better, these positions are closely related to technical skills: we need plumbers, electricians, construction workers, steelworkers, network technicians, and people to install and deploy equipment.

In the United States, the salaries of these positions are already close to or exceed six figures. You don't need a computer doctorate to have a decent career.

Let me give you two real - life examples.

Ten years ago, people thought AI would eliminate radiology departments. As a result, AI has fully entered image analysis, and doctors can view images much faster. Then they have more time to communicate with patients and make diagnoses, and the number of radiologists has actually increased.

The logic is: The hospital's reception capacity increases → revenue increases → more radiologists are hired

The same goes for nurses. There is currently a shortage of about 5 million nurses in the United States.

If AI is used for medical record - keeping and visit record transcription, nurses actually spend about half of their time on these "record - keeping" tasks.

Now they can use AI technology. There's a company called Abridge, our partner, which is doing very well. So nurses can spend more time on real "patient - reaching" visits and care.

Moreover, since more patients can be received now, we're no longer limited by the number of nurses, and more patients can enter the hospital earlier. As a result, the hospital operates better, and they hire more nurses.

The logic is: The hospital's operating efficiency increases → more nurses are hired.

So, the simplest way to judge the impact of AI on a certain job is to figure out what the purpose of this job is and what the specific tasks are.

For example, if you just set up a camera to film us, you might think we're typists because I spend most of my time typing. If AI automates a large amount of text prediction and helps us type, we'll be unemployed.

But obviously, that's not the purpose of our work.

So the question is: What's the purpose of your work? The purpose of radiologists and nurses is to care for patients. When the tasks are automated, this purpose is actually strengthened and magnified.

Using the "purpose vs. task" framework to analyze each type of occupation will be very helpful.

AI is an opportunity to "narrow the gap" for developing countries

Host: Let's shift the topic outside of developed economies and discuss how AI can spread globally and help the world. How can we ensure that AI becomes a truly transformative technology, just like Wi - Fi and 5G have been for the emerging world? What does it mean when it intersects with emerging markets, and how can we "grow" the global economy? Turning to the developing world, how do you think it will evolve?

Jensen Huang: I'm very optimistic about the impact of AI on emerging economies. AI itself is a kind of infrastructure . Every country has roads and electricity, and it should also have AI.

There are now a large number of open - source models, and training AI is no longer out of reach. Language and culture are a country's "natural resources".

Second, remember: AI is the easiest - to - use software in human history, which is also why it's growing and being adopted the fastest. In just two or three years, the user base is approaching one billion.

For many people without a computer science degree, you can now become programmers. In the past, we had to learn how to program computers; now, the way you "program" is to directly ask the computer: "How should I program you?"

If you don't know how to use AI, you just go to AI and say: "I don't know how to use AI. How should I use it?" It will explain it to you.

You can also say: "I want to write a program to build my own website. How do I do it?" It will ask you a bunch of questions: What kind of website do you want, and then write the code for you.

In the past, humans learned how to write code for computers; now, humans only need to tell computers what to do in language.

This will greatly lower the technological threshold and help more countries and people participate in the digital economy.

The key opportunity for Europe: Industry × AI × Robotics

Host: We've talked about many companies, mentioning many American and Asian companies. Please talk about how AI intersects with Europe's success and future. What role will NVIDIA play in Europe?

Jensen Huang: For Europe, this is an extremely critical window period.

The United States dominated the software era; but AI is a kind of "software that doesn't require writing code".

Europe has the world's strongest industrial manufacturing and deep - tech foundation.

If AI is deeply integrated into manufacturing, engineering, robotics, and physical world modeling, Europe has every chance to make a leap in the Physical AI / Robotics era.

But the prerequisites are: increasing energy supply, increasing infrastructure investment, and fully engaging in the AI ecosystem construction as early as possible.

So it's not a bubble; the key is "whether the investment is enough"

Host: So what I'm hearing is that we're still far from an AI bubble. The real question is: Are we investing enough? Because many people are talking about a bubble, but what you're saying is: Are we investing enough to achieve the goal of "growing" the global economy?

Jensen Huang: This is not a bubble but a long - term construction cycle with insufficient investment so far.

The spot rental prices of GPUs are rising. Not only for the latest generation but even for GPUs two generations ago, the spot rental prices are increasing. The reason is that the number of AI companies is growing, and more and more companies are adjusting their R & D budgets.

Eli Lilly is a good example: Three years ago, almost all of their R & D budget was in wet labs; but notice that they've now invested in a large - scale AI supercomputer and a large - scale