NVIDIA's Jensen Huang on Building the Dynamo of the Intelligence Age
In a Nutshell
Jensen Huang argues that AI is shifting computing from retrieval to generative intelligence, where every interaction produces original, personalized content rather than stored data. This transformation requires massive infrastructure—AI factories built on five layers (energy, chips, data centers, models, applications)—that he compares to the dynamo powering the Intelligence Age. The core takeaway is that AI will not eliminate jobs but elevate them, as people using AI outperform those who don't, making adoption essential across every industry and role.
These notes were generated by AI and may contain inaccuracies.
The current AI revolution is described as bigger and faster than the industrial revolution, representing the largest infrastructure buildout in human history. At the center of this buildout is the AI factory, with NVIDIA enabling this transformation.
AI is most commonly understood through chatbots accessed via web browsers. Users provide a prompt and receive a response. Over the last two to three years, AI capabilities have significantly evolved. Two years ago, ChatGPT demonstrated the ability to understand input, perceive and understand information, and translate or generate information into something else.
Examples of AI translation capabilities include:
- Text to text: summarizing a PDF
- Text to image: generating an image from a story
- Image to text: describing what's happening in a picture
Two years ago, AI was largely capable of translation, known as generative AI. The significant development within generative AI is the transition from generation to thinking—generating internal thoughts, reasoning, step-by-step reasoning, and problem solving. This also enabled generating intelligence to control tools, whether digital tools like browsers, spreadsheets, Photoshop, PowerPoint, AutoCAD, or eventually mechanical systems such as robotics and self-driving cars.
Two years after the foundational ChatGPT technology, the industry now has agentic systems capable of understanding, reasoning, and doing work. Two years ago, AI was interesting and novel but not particularly valuable. Now, because it can do work, AI has become valuable and can be paid for its services. Companies are paying AI by the hour—$30/hour or $20/hour—to perform useful work, making it the fastest growing software business in history.
When companies build great businesses, they consider both downstream implications for industries like healthcare, financial services, life sciences, manufacturing, logistics, transportation, retail, advertising, and entertainment, as well as upstream industrial implications.
Modern computing emerged 64 years ago with IBM System 360. For 40 years, computing remained largely the same: a pre-recording, retrieval-based system where data is written to files and retrieved from disk drives when needed. Data centers are named for their storage function, not their computing function.
AI changes this paradigm fundamentally. Every time AI is used, new information (context) and a prompt (query) are provided. The AI understands the context and query, reasons about them, and produces output based on the specific circumstance. This means content is generated originally every single time, in real time.
The computer has been reinvented as generative. In the future, every pixel, sound, video, image, advertisement, TV commercial, and news story will be originally generated rather than retrieved. Each person's experience will be different based on their interests, context, identity, and how they asked the question.
Currently, AI serves about a billion people. As AI becomes agentic and capable of doing work independently, agents will communicate with each other, team up, and solve problems. Inside companies, hundreds of thousands of agents may run simultaneously, talking to each other and solving problems within guardrails and sandboxes.
In the future, the internet used by a billion people will likely be used by several billion—potentially 100 billion—agents working around the clock. These agents will include company agents, employee agents, self-driving cars, robots, and manufacturing systems. All commands and thoughts generated between agents will be originally generated.
This layer of intelligence-generating computing will cocoon the Earth, similar to how electricity and the internet have. Three hundred years ago, Siemens produced a machine that generated electricity from motion. Power generators now form a grid that cocoons the planet. Thirty-five years ago, a networking scheme created in the United States became the internet, which also cocoons the world. Now, energy, communications, and intelligence will all cocoon the planet as commodities.
The machine invented 300 years ago is called the dynamo, which converts motion (atoms) into electrons. NVIDIA's machines take electrons in and produce numbers (tokens) out. These numbers can be reformulated into language, math, proteins, human biology, physics, climate, weather, robotics, self-driving cars, and all forms of intelligence.
Each AI factory rack contains 72 chips, weighs two tons, costs $4 million, and contains 1.5 million parts. Each gigawatt of capacity represents approximately $50 billion in investment but generates $300-400 billion in intelligence value. The return on investment is extremely fast.
The industrial structure of AI can be understood as a five-layer cake:
-
Energy layer: The bottom layer, representing the greatest opportunity in several generations for energy industries. This includes sustainable energy investments in nuclear, wind, solar, and hydrogen. A trillion dollars is being invested this year alone into this ecosystem.
-
Chips and computers layer: Includes chips, computers, networking, switches, and silicon photonics.
-
Infrastructure layer: Includes land, power, shell, money, and data center operations—all in scarce supply.
-
Model layer: Includes companies like OpenAI and Anthropic, but extends beyond language models to structural domains including proteins, genes, cells, physics, and the physical world.
-
Applications layer: The top layer where AI enhances human condition in financial services, legal, accounting, transportation, logistics, and other domains. Last year saw $100 billion in venture capital investment in this layer—the single largest year of VC investment in human history.
The current trillion-dollar investment represents the beginning of what may become a $20 trillion per year ecosystem. The question driving investment is how important intelligence is, who needs it, and how much of it is desired. All forms of intelligence—whether for proteins, cars, robots, language, math, or science—must be generated by these machines.
Every country has a different attitude about AI based on their culture. Recommendations include being careful with analogies and science fiction stories involving Terminator, singularity, and related concepts.
Jensen Huang cautions against using analogies and science fiction stories that portray AI as Terminator or reference ideas like the singularity. Claims that AI has a 20% chance of ending humanity are described as complete nonsense. The idea that AI is mysterious and no one knows how it works is rejected outright. Huang points out that every year the technology demonstrably improves, which would be impossible if researchers did not understand how it functions.
Huang emphasizes that individuals may or may not lose a job to AI, but they will absolutely lose a job to someone who uses AI. The practical response is to focus on what is certain: learn to use AI rather than worrying about uncertain outcomes. This advice extends to parents advising their children—rather than running away from the technology, people should ensure they engage with the superpowers it provides.
NVIDIA is dedicating substantial computer science resources, investment, and passion to making AI safe for everyone. Evidence of rapid safety improvements is visible when comparing ChatGPT from two years ago to current versions. Hallucinations have been dramatically reduced, outputs are more accurate and contextually relevant, the system conducts research when needed, and it now reflects on multiple possible answers before responding. Huang compares this progress to automotive safety: cars today are far safer than those from a century ago because technology advanced to make them safer.
The technology industry, scientists, and engineers bear responsibility for building AI safely. Individuals have a parallel responsibility to encourage family, children, companies, and countries to engage with AI. The core message is that humanity will not lose to AI itself but to those who use AI effectively.
A trillion dollars has been invested into the global ecosystem this year, generating jobs across multiple sectors. The energy sector, chip sector, infrastructure layer (land, power, shell, finance), AI model layer, and upper application layer are all experiencing record job growth. Traditional job concerns are addressed by distinguishing between a job's purpose and its specific tasks.
Huang's role as CEO involves leading the company, yet most of his time is spent typing and talking—tasks AI can already perform in a superhuman way. Despite this, he reports being busier than ever, illustrating that automation of tasks does not eliminate the underlying purpose of the role.
Over a decade ago, a leading computer scientist predicted that radiology would be the first profession eliminated by AI because computer vision already possessed superhuman image recognition capabilities. The prediction was partially correct: computer vision penetrated every form of radiology. However, the outcome was the opposite of elimination. Radiology demand increased, the number of radiologists grew, and departments became more profitable because automation allowed them to handle more patients and scans.
The original warning caused some people to avoid radiology careers, yet more radiologists are needed. This example demonstrates the harm that can result from irresponsible predictions framed around task elimination rather than purpose elevation.
Recent claims that 90% of software coding will disappear, eliminating the need for software engineers, are contradicted by companies hiring more engineers than ever. A software engineer's purpose is to solve problems and innovate, not merely to type code. Coding is a task, not the job itself.
Radiologists existed before workstations and will exist after AI. Engineers existed before software coding and will exist afterward. The narrative that 50% of jobs will vanish stems from focusing narrowly on tasks rather than the purpose of work.
Countries and companies that fail to invest in AI miss out on both job creation and the elevation of their workforce. AI is not positioned to eliminate jobs but to elevate them. A plumber today receives task sheets or schematics; tomorrow, that same plumber is likely to function as a designer using AI-generated kitchen plans. A furniture salesperson can become an interior designer. A carpenter can evolve into a home designer. These shifts represent elevation of craft rather than displacement.
AI is described as the greatest force for eliminating the technology divide in Huang's career. Over 40 years in computer design, systems grew increasingly complex and the percentage of the population able to program them declined. Only about 2% of people know C++, whereas far more know human language. With modern AI, everyone can now program computers, reversing the prior exclusionary trend.
The conversation concludes with appreciation for Huang's perspective on the shift from retrieval to generation, customized knowledge, the intelligence revolution paralleling energy and telecommunications revolutions, new computer languages including protein, five layers of participation, and the movement from task execution to problem definition and purpose-driven work.
Keep Sequoia Capital in your library
Save the videos and channels worth coming back to, and find them again in one place.
