In 2016, Geoffrey Hinton told the world to stop training radiologists. The Nobel laureate and AI pioneer argued that deep learning would render human image analysis obsolete within five years. Medical students reconsidered their career paths. The prediction seemed not just plausible but inevitable.

A decade later, the Mayo Clinic employs over 400 radiologists, a 55% increase since Hinton made that forecast. AI has completely permeated medical imaging. Every scan gets processed by algorithms. Yet hospitals are hiring more specialists, not fewer.

Jensen Huang stood on stage at the World Economic Forum in Davos last week and explained why.

What Huang Said at Davos

Speaking with BlackRock CEO Larry Fink on January 21, Huang called AI the foundation of “the largest infrastructure buildout in human history.” When Fink asked about job displacement, Huang pointed directly to radiology.

Huang stated that AI has now diffused into every aspect of the field. The impact is 100%. But, reasoning from first principles, the number of radiologists has actually gone up. He explained that the purpose of a radiologist’s job is to diagnose disease and help patients. Studying scans is merely a task. By enabling radiologists to process scans infinitely faster, AI freed them to spend more time with patients and consult with other clinicians. Hospital throughput increased. Revenue grew. Institutions hired more staff to keep up with the expanded capacity.

The same pattern is emerging in nursing, Huang noted. The United States faces a shortage of roughly five million nurses, partly because they spend half their time charting. AI-powered transcription and automated record-keeping allow nurses to see more patients. The bottleneck clears. Demand expands.

The Numbers Behind the Paradox

The American College of Radiology published a workforce study in February 2025 using CMS data from 2014 to 2023. In 2023, 37,482 radiologists were enrolled to provide care to Medicare patients. The study projected that number will grow 25.7% to 40.3% by 2055, depending on residency expansion.

At Mayo Clinic, the radiology department now runs more than 250 AI models. A dedicated team of 40 AI scientists, researchers, and engineers develops and licenses these tools. John Halamka, president of Mayo Clinic Platform, told the New York Times that within five years, not using AI in radiology will constitute malpractice.

Hinton himself acknowledged his prediction missed. He told the Times he spoke too broadly and was wrong on timing, though he maintains AI and radiologists will eventually work together to improve both efficiency and accuracy.

Why Efficiency Creates Demand

This pattern has a name in economics: Jevons Paradox. When steam engines became more fuel-efficient in the 19th century, coal consumption increased rather than decreased. Better efficiency meant more use cases became economically viable.

The same logic applies to AI. Before algorithmic analysis, many diagnostic imaging procedures were either too slow or too expensive to perform at scale. Wait times for scans stretched into weeks. Subtle abnormalities went undetected because radiologists lacked time to examine every image carefully.

AI cleared the constraint. More patients can now get scans. Earlier detection becomes possible. The addressable market expands dramatically.

David Autor, a labor economist at MIT, told the Times that predictions of AI job destruction typically underestimate the complexity of what people actually do. Radiologists do far more than study images. They advise surgeons, consult with patients, interpret findings in context of individual medical histories, and make judgment calls that require decades of training.

The Blueprint for Physical AI

Huang framed this as the template for the next decade of what he calls Physical AI and agentic workflows. Unlike traditional software that records data, these systems act on it. Humanoid robots and autonomous agents do not simply replace a worker on an assembly line. They manage multi-step processes, predict equipment failures, and self-correct schedules in real time.

The International Federation of Robotics released its 2026 trends report identifying agentic AI as a key development. These systems combine analytical AI for structured decision-making with generative AI for adaptability, enabling robots to work independently in complex environments.

Deloitte estimates humanoid robot shipments will reach 15,000 units in 2026, nearly triple the 2025 level. Applications span manufacturing, logistics, and healthcare. These machines are not designed to eliminate workers. They are designed to operate in environments built for humans, handling tasks that are dull, dirty, or dangerous.

A factory equipped with these systems does not shrink its workforce. It changes what workers do. A technician becomes an orchestrator, supervising an entire fleet rather than performing a single repetitive task.

The Demographic Dividend

This shift arrives at a critical moment. The manufacturing sector faces severe labor constraints. Nearly one-third of skilled trade workers are over age 55. The Bureau of Labor Statistics projects an estimated 2.1 million manufacturing jobs could go unfilled by 2030 if current trends continue.

EY’s latest U.S. AI Pulse Survey found that only 17% of companies using AI report job losses as a result. For every company cutting headcount, more are reinvesting productivity gains into upskilling and expansion. Dan Diasio, EY’s global consulting AI leader, said the most advanced AI adopters are using the technology as a catalyst for value creation rather than as a cost cutter.

MAGNET’s 2025 Ohio Manufacturing Survey found that 70% of manufacturers expect their headcount to grow in 2026, despite tariffs and supply chain volatility.

The Investment Thesis

The opportunity lies with companies using AI to expand output rather than reduce costs. Look for the radiologists of other industries: firms in manufacturing, logistics, and energy that deploy agents to multiply the work of existing teams.

Huang noted at Davos that 2025 was one of the largest years in venture capital history, with most funding flowing to AI-native companies. These firms span healthcare, robotics, manufacturing, and financial services. The application layer, Huang said, is where economic benefit will materialize.

The narrative of replacement assumes a fixed amount of work. That assumption is false. Human demand for better healthcare, faster delivery, and cleaner energy is effectively unlimited. AI is the tool that finally allows supply to meet it.

Watch for companies increasing headcount to manage new AI-driven scale. That signal reveals where capacity is expanding. The future belongs to the orchestrators.

If you liked this, you’ll love the daily dose we send to our Substack crew. It’s chill, it’s sharp, and it actually helps. Hop on here.

Sources