Jensen Huang predicts again: AI combined with the real economy, a market worth $90 trillion

Jensen Huang said, “Knowledge is a very large industry, but the truly massive industry is the integration of information and the real world, valued at $90 trillion.”

On March 10, 2026, local time, NVIDIA founder Jensen Huang published a signed article, mentioning the “five-layer cake” framework of AI (artificial intelligence). The article reviewed the past year, summarizing that AI has crossed an important threshold—model performance has significantly improved, allowing for large-scale deployment; reasoning capabilities have strengthened, with fewer hallucinations, and the ability to apply AI in practice has greatly enhanced; applications built on AI have first created real economic value.

Reviewing Huang’s statements in various settings over the past three months, this creation of “real value” primarily stems from the practical applications of intelligent agents. In February, during NVIDIA’s fourth-quarter earnings call, Huang stated, “Intelligent agent AI has reached a turning point in development, with its practicality widely validated across global enterprises, leading to an explosive increase in demand for computing power.”

Since February, the rapid deployment of OpenClaw (open-source AI agents) across multiple industries worldwide has validated this turning point. The popularization and maturation of intelligent agents have allowed AI to move beyond a simple dialog window, beginning to handle real tasks for people in the virtual world through computers and networks.

However, beyond the wave of intelligent agents, a significant technological wave with immense potential, whose impact has not yet been fully realized, has begun to enter the task lists of global AI giants—world models or physical models aimed at the real world. This model is believed to be able to understand the spatial and physical logic of the real world, allowing AI to truly integrate into real-life and production scenarios.

From his January CES speech to the aforementioned earnings call, Huang has repeatedly emphasized the potential of this approach. In February, NVIDIA also announced a long-term strategic partnership with French industrial software giant Dassault Systèmes to jointly build a shared AI architecture for key business scenarios across various industries. Dassault Systèmes’ SOLIDWORKS and CATIA are mainstream design software in the industrial sector.

During the fourth-quarter earnings call, Huang said, “We are welcoming the wave of intelligent agent AI development, and the next wave will be physical AI—applying AI and intelligent agent systems in physical fields like manufacturing and robotics, which will bring us tremendous development opportunities.”

AI Moving into the Real World

What capabilities does an AI that understands the real world need?

Huang provided the answer during his keynote speech at CES 2026. The real world has some basic characteristics, such as constancy: “If I put something here and look away, when I look back, it is still there”; and causality: “If I push it, it will fall.”

However, for an AI to understand the real world, it also needs to grasp physical laws like friction, gravity, and inertia, knowing that a heavy truck requires a longer braking distance. While these are common sense for humans, they are completely foreign to AI.

The past AI revolution was essentially a breakthrough in the “symbolic space.” From BERT to ChatGPT, large models learned to understand the grammar, semantics, and context of language, even completing complex reasoning tasks, but they knew almost nothing about the physical laws of the real world—gravity, friction, inertia, causality. A large model that can write beautiful prose does not understand what happens when a rock rolls down a hill. “Physical AI” was born to fill this gap. Huang defines it as “AI that can understand natural laws,” with the core goal of enabling AI to not only process linguistic symbols but to truly understand and interact with the physical world.

He believes this will be the next stage of AI development and will have a profound impact on the world. From an industry perspective, AI that understands the real world will reshape the vast automotive, transportation, and manufacturing sectors through autonomous driving and robotics. In the first three months of this year, NVIDIA has launched several architectures and products focused on the physical world.

Not only NVIDIA, but in the first three months of 2026, while the agent wave swept the globe, Silicon Valley has frequently made moves regarding the world AI layout.

Google DeepMind’s Genie3 was officially opened to the public at the end of January, allowing users to input text and have Genie3 generate interactive 3D environments in real time. Waymo then transformed it into a simulation tool specifically for autonomous driving in February, used to generate extreme scenarios that a fleet would almost never encounter on real roads, such as tornadoes, flooded roads, or elephants suddenly appearing at intersections.

More symbolically, is the departure of Turing Award winner Yann LeCun. He left Meta at the end of 2025 to found AMI Labs and announced on March 10 the completion of $1.03 billion in seed funding, with a pre-funding valuation of $3.5 billion, marking the largest seed round financing in European history, with NVIDIA and Samsung among the investors.

LeCun’s judgment is that large language models are a dead end because they cannot truly understand how the physical world operates. His new company bets on the JEPA architecture—a framework that does not predict every pixel detail but learns to understand the structure of the world, with target applications including healthcare, robotics, and industrial automation.

Meanwhile, “AI mother” Fei-Fei Li’s World Labs completed a new round of $1 billion financing in February, nearing a valuation of $5 billion, with its first product, Marble, already launched, focusing on generating 3D worlds that accurately follow physical laws. The CEO of AMI Labs predicted after the funding: within six months, every company will claim to be a “world model company” to secure funding. This prediction itself already indicates the temperature of this track.

Making AI better understand the real world has become quite clear in 2026.

The “Turning Point Moment” for AI in Manufacturing

On February 3, 2026, local time, after finishing his tour in Asia, Huang traveled to Houston, Texas, to appear at Dassault Systèmes’ 3DEXPERIENCE World conference.

There, he jointly announced a strategic partnership described as “the largest in 25 years” with Dassault Systèmes CEO Pascal Daloz—deeply integrating NVIDIA’s accelerated computing and AI capabilities with Dassault Systèmes’ virtual twin platform. Dassault Systèmes has been established for over forty years, and its 3DEXPERIENCE platform serves more than 45 million users and 400,000 clients. This French company is one of the deepest software suppliers in global manufacturing—its shadow can be seen behind everything from aircraft engines to consumer product housings.

This collaboration points to an important field: industry. In the past, although large language models played important roles in some areas, their applications in industrial fields remained quite limited due to reliability and a lack of understanding of the real world. The core goal of the collaboration has been defined as building an “industrial world model”—a scientifically validated AI system rooted in physics that can serve as a key task platform in biology, materials science, engineering, and manufacturing. Dassault Systèmes is also deploying “AI factories” based on NVIDIA’s latest AI infrastructure across three continents through its cloud brand OUTSCALE, aiming to provide enhanced capabilities for AI models on the 3DEXPERIENCE platform while ensuring customer data privacy and sovereignty.

Gian Paolo Bassi, global senior vice president of Dassault Systèmes’ professional customer division, stated in an interview with The Economic Observer and other media that many large model companies cannot create new atomic structures, nor can they develop new alloys, aircraft, or aerospace equipment, because they primarily focus on language models and lack the specialized knowledge to develop drugs or new devices. Dassault Systèmes’ core advantage lies precisely in the “hard knowledge” that has already been embedded in its software. Bassi said, “Our knowledge is in the software, with some industrial-related expertise, and in this area, Dassault Systèmes has a unique advantage.”

This means that artificial intelligence must rebuild its system starting from specialized knowledge. Taking the medical device industry, with its strict regulatory requirements, as an example, traditional validation cycles are long and costly, but with AI’s assistance, engineers can simulate many different scenarios, completing products more effectively, quickly, and with higher quality.

The bigger ambition is that, with AI’s assistance, tests that previously required repeated physical prototypes can now be completed with numerous parallel iterations in the digital world at very low costs, meaning that the entire supply chain process from raw materials to assembly, and the production processes within, can all be reconstructed in the virtual world.

Gian Paolo Bassi stated in an interview with The Economic Observer and other media, “Dassault Systèmes has built highly realistic virtual twin models over the years, and our collaboration with NVIDIA allows these models to operate on a large scale, with high precision and near real-time conditions, and to be directly utilized by AI, thus upgrading virtual twins from engineering tools to sustainable system-level capabilities.”

The concept of virtual twin is not new; it describes the precise mapping of physical systems using digital models, allowing engineers to test and iterate in the virtual world before applying conclusions to the real world. Dassault Systèmes has been working in this direction for many years, and its technical system is quite mature. However, for a long time, the large-scale deployment of this technology faced a fundamental bottleneck: computing power. Sufficiently realistic and complex physical simulations require computing capabilities far beyond what has been available. Now, this bottleneck is being broken.

Huang said that in the past, industrial enterprises spent one-third of their time on design and digitalization, with more time spent on building physical forms. In the future, they can spend 100% of their time on digitalization. From producing a pair of tennis shoes to manufacturing a car, whether it is designing, sketching, simulating, or operating, “everything is defined by software.”

Once this precise virtual reconstruction is completed, the combination of AI and robotics could almost reshape the processes and efficiencies of manufacturing.

A factory is not a whole but a collection made up of millions of objects. AI can help simulate all parts of this collection in the digital world and organize the robots within the factory to operate more rationally on the production line.

In large manufacturing enterprises, such scenarios have already emerged: maximum virtual twins and widely deployed robots. But the cost of achieving all this is high, which has also meant that robots can only survive in industries where tasks are highly repetitive and in large volumes, such as the automotive industry, where a robot is specifically programmed to do one thing.

In Huang’s view, this is also the value of AI entering the industry. Through the enhancement of simulation modeling efficiency and the improvement of robot intelligence levels with AI technology, a vast majority of small and medium-sized enterprises occupying the global supply chain can also utilize these cutting-edge technologies, which will undoubtedly reshape the efficiency of the entire industry.

Huang said, “Knowledge is a very large industry, but the truly massive industry is the integration of information and the real world, valued at $90 trillion.”

For technology believers, the arrival of physical AI may be as significant as the internet lowering the cost of information flow to nearly zero. If that revolution reshaped the production and distribution of information, this revolution reshapes the design and operation of the physical world itself.

Dassault Systèmes CEO Pascal Daloz stated, “We are entering a new era where AI is no longer limited to predicting or generating content, but is beginning to truly understand the physical world. When AI is rooted in science, physics, and validated industrial knowledge, it will become an amplifier of human intelligence.”

In Daloz’s view, when AI enters the physical world, its achievements will not lie in replacing designers and engineers but in that “success is not about automation; engineers do not want to automate past achievements; they want to create the future.”