Nvidia GTC keynote highlights AI factories and AI agents

In short: Nvidia used its March 16 GTC 2026 opening keynote to outline how it plans to build and sell more of the computing building blocks behind today’s AI.

What happened

Nvidia CEO Jensen Huang gave a keynote that ran for more than two hours at the company’s GTC conference in San Jose, California. The talk kicked off the event, which runs March 16 to March 19.

Huang framed Nvidia’s business around three main “platforms.” First was CUDA-X, which is Nvidia’s software tools that help developers use Nvidia chips more easily. Second was Nvidia’s full computer systems for data centers (large rooms of computers used by companies and cloud providers). Third was a newer idea he called “AI factories,” meaning purpose-built setups that turn large amounts of computing into useful AI outputs, similar to how a factory turns raw materials into products.

He also discussed “agentic systems,” which is shorthand for AI agents that can carry out multi-step tasks on their own (like a digital assistant that can plan, click through steps, and finish a job instead of only answering questions). The keynote touched on uses in areas like retail supply chains, live video and games, and robotics. Nvidia also mentioned work that combines AI with quantum computing efforts, including a program called Holoscan Quantum and partnerships.

Some claims mentioned in outside commentary, including a specific $1 trillion AI chip sales projection through 2027 and an “OpenClaw strategy,” are not clearly supported by the available keynote materials. The full keynote video is available on YouTube for people who want to check what was said.

Why it matters

Even if you never buy a Nvidia product, Nvidia’s chips and systems often power the AI features that show up in everyday apps and services. When Nvidia changes how it builds and packages AI computing, it can affect how fast AI gets deployed and how much it costs for companies to run.

Source: TechCrunch AI