Jensen Huang sees a $200 billion processor market for AI agents opening up for Nvidia

Nvidia CEO Jensen Huang has identified a $200 billion market for the company: processors built specifically for AI agents. This is a new generation of artificial intelligence that performs complex tasks autonomously, without direct commands from the user.

What are AI agents

AI agents are not just chatbots that answer questions. They are programs that see the screen the way a person does, click buttons, fill in forms and make decisions on their own. They work in the browser, in the operating system, in apps: anywhere a user can work.

OpenAI recently showed a demo of Operator, an agent based on GPT-5.5 that orders food through DoorDash, buys plane tickets and compares prices on marketplaces. Google and Anthropic are developing similar systems. This is no longer theory: agents are arriving in production versions of popular platforms, available for testing.

Agents need computing power to work. They must see the screen, analyze information, talk to web services and make decisions in real time, all at once. That is nothing like simply printing a line of text in reply to a question.

Why Nvidia sees this market

So far, Nvidia's GPUs have dominated one area: training large language models. Millions of graphics cards went into training ChatGPT, Claude and Gemini. But training is a one-time event. Afterward, the finished models have to run in production, day after day, millions of times.

Agents will run constantly. They perform many small operations, make decisions based on new data and interact with their environment. That demands a very different architecture from the one that suits training. It calls for central processing units (CPUs), which handle logic and decision-making well. GPUs specialize in matrix computation, which is not quite what agents need.

Huang expects AI agents to blow up the market within the next 3-5 years. Large companies will run agents on their own servers to serve customers, process data and manage logistics. People will install them on home computers to automate routine chores. That means demand for an entirely new class of computing: not for training models, but for running them day to day in production.

Market size and speed

$200 billion is an impressive figure. For comparison, it roughly equals the current size of the entire server processor market (Intel, AMD and their competitors). If Huang's forecast is right, the CPU market will simply double. And that is only the beginning.

But the more interesting thing is not the number but the pace of development. A year ago, serious AI agents did not exist. Six months ago, people talked about them as science fiction. Today they are already in beta tests with millions of users. Tomorrow they may become an everyday tool in every company. It shows how fast the artificial intelligence industry is moving.

What this means

Nvidia is preparing not for a single new feature but for a shift in computing paradigm. When a new class of computing tasks appears (as GPUs once did for graphics), a new market is born. If agents become as popular as search engines or social networks, demand for computing power will grow exponentially.

Companies will begin competing not only on the model (whether its results are better) but also on speed: whose agent is faster, more reliable, cheaper. For business, this means routine tasks such as online transactions, document processing and customer interaction will move to agents ever faster. Companies will have to rethink what their people do and which skills are truly valuable in the AI era. This is happening not over ten years, but over two or three.