AI is a platform shift whose five layers demand the largest infrastructure buildout in human history.
AI is described as a five-layer cake: energy at the bottom, then chips and computing infrastructure, then cloud infrastructure and services, then the AI models most people think of as AI, and on top the application layer, in financial services, health care or manufacturing, where the economic benefit ultimately happens.
We are only a few hundred billion dollars into what he calls the largest infrastructure buildout in human history. Trillions of dollars of infrastructure still need to be built, because all of these contexts must be processed so the models can generate the intelligence that powers the applications sitting on top.
No. Ten years ago radiology was the profession everyone expected to be wiped out, because the first superhuman AI was computer vision and scans were its largest application. Ten years later AI has completely permeated radiology and radiologists do use it on scans, yet the number of radiologists has gone up.
Because the purpose of the job is diagnosing disease, while studying scans is only a task. Doing scans infinitely fast lets radiologists spend more time with patients and other clinicians. The hospital then sees more of the people who waited a long time, revenues go up, and it hires more radiologists.
The United States is five million nurses short, and nurses spend half their time charting. Using AI for charting and transcribing patient visits, with partner company Abridge named as doing incredible work, lets nurses visit more patients, so patients get in sooner, hospitals do better, and they hire more nurses.
First, models became better grounded instead of hallucinating, reasoning step by step and evolving into agentic AI systems. Second, open models: DeepSeek, the world's first open reasoning model, was followed by many others. Third, physical AI, understanding nature, proteins, chemicals, fluid dynamics, particle physics and quantum physics.
Separate the purpose of a job from its tasks. A camera pointed at the two speakers would suggest they are typists, because they type all day, but that is not their purpose. For radiologists and nurses the purpose is caring for people, and automating the task makes that purpose more productive.
Treat AI as infrastructure, like electricity and roads. Training AI is not so hard these days, and open models combined with local expertise let a country build its own. Take advantage of your fundamental natural resource, your language and culture, and keep your national intelligence inside your own ecosystem.
His test: NVIDIA has millions of GPUs in every cloud, renting one is incredibly hard, and spot rental prices are rising even for two-generation-old GPUs. Investments are large because the infrastructure under every layer of AI has to be built, so he sees the opportunity as extraordinary rather than a bubble.
Europe's industrial manufacturing base is incredibly strong, so it can leap past the software era the United States led and fuse manufacturing with AI, moving into physical AI and robotics, a once-in-a-generation opportunity. Its deep sciences are strong too, but energy supply must increase to support the infrastructure layer.
“At the same time, BlackRock's annualized total return has been 21 percent.”
“And so I sold NVIDIA stock at a valuation of $300 million.”
“AI is actually essentially a five-layer cake.”
“We're now a few hundred billion dollars into it.”
“We're 5 million nurses short in the United States.”
“Get engaged in AI. And recognize that AI is likely to close the technology divide.”
“Conversation with Jensen Huang, President and CEO of NVIDIA | WEF Annual Meeting 2026.” https://www.youtube.com/watch?v=hoDYYCyxMuE. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/conversation-with-jensen-huang-president-and-ceo-of-nvidia-w.html.