Jensen Wong argues agentic AI became genuinely useful, driving thousandfold compute demand while creating jobs, not destroying them.
Anthropic's Claude Code came out and was the first agentic system able to do really productive work, like software coding. Coding is the codification of anything you want to automate over and over, so it matters to every company, not just software engineers. That is when AI became useful.
Moving from generation to understanding, reasoning, planning and using tools takes about a thousand times more computation. Multiply that by roughly a hundred times more people wanting to use it, and GPU consumption goes through the roof. Even GPUs sold four or five years ago are rising in price faster than good wine.
It changes all the time. Two years ago there was enough energy but a real shortage of chips, and it depends which chips. Today's systems contain seven different types of chips, built with just about every chip company and systems maker in the world, so a bottleneck sits somewhere in the largest supply chain in the world.
Vera Rubin is about twice the width of the stage. Each rack costs roughly four or five million dollars, weighs three tons and contains one and a half million parts, and a data center holds a football field of racks, with silicon photonics, the most advanced memories, three-dimensional packaging and cooling.
Under the CHIPS Act everybody was reluctant to build in the United States. Jensen told President Trump he would place half a trillion dollars of orders with suppliers and bet they would come. They all came to build here, creating hundreds of thousands of jobs for the next four or five years.
People called them circular deals. NVIDIA put in roughly a dollar while the company still had to raise another nine, and that anchor investment gave other investors confidence NVIDIA was behind it. Everyone who invested alongside NVIDIA in CoreWeave, Nebius and Nscale is extremely happy, because NVIDIA can see the demand and the pipeline coming.
In the last three to six months both companies, and most AI-native companies, turned gross margins extremely positive. When what you make is highly profitable your goal is to make more of it, which is why both are racing for capacity. Cursor is excellent too, right across the AI-native ecosystem.
No. America should always have a head start, and the United States has the right to the first, the most and the best. But all American companies should compete globally, because the aim is to maximize exports and revenues; tax revenues fund defense, and economic security contributes to national security.
Not with another Mythos. You do not defend against a super force with another super force, but with an abundance of cheap force: open-source models run as swarms of white blood cells, trained to detect threats, find where a threat comes from and close the door. Asymmetry is what you want.
No. A decade later AI is completely integrated into radiology, and computer vision is superhuman at reading scans, but the job was not wiped out. Radiologists study more scans, accept more patients and diagnose better, radiology became one of the biggest profit centers, and hospitals want to hire more radiologists.
“AI actually reinvented the computer industry. AI invented a whole new industry.”
“Everybody should have AI. The global south should have AI.”
“the United States benefited from the last industrial revolution for a good reason. Not because we invented it, but because we applied it.”
“even without seeing you, I can tell you you need AI. None of you should have a nuke.”
“The way you defend against a super force is not with another super force. It's with an abundance of cheap force.”
“No, I think you should move fast, but you shouldn't break things.”
“Leading in the Age of AI: A Conversation with NVIDIA CEO Jensen Huang | Global Conference 2026.” https://www.youtube.com/watch?v=dUFis1dte14. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/leading-in-the-age-of-ai-a-conversation-with-nvidia-ceo-jens.html.