Industry of Work

The core distinction

"The software industry you and I created is an industry of tools, tools used by humans. For the very first time this new type of software, neural networks, large language models, agents, robots, are not tools but they are skills, they're tasks, they do work. The industry of work is not one trillion, it is 100 trillion dollars. And that's the reason why we realized that this industry is in fact not a transition of the IT industry but a transition of every industry," argues Jensen Huang.¹

The prior software era produced tools: applications, operating systems, databases, platforms that amplify human capability but still require a human to direct each step, with value accruing to the toolmaker while the underlying work still depends on human labor. AI produces skills and tasks instead: an agent that handles support, discovers drugs, writes code, or manages inventory is not a tool a human wields, it is a substitute for a unit of human work. That shifts the addressable market from the technology industry, roughly a trillion dollars of software and hardware spend, to every job and every business function that currently requires human cognition, since, as Huang puts it, "every activity of GDP today is based on mankind's some type of brain activity." If AI can perform brain-work, there is no sector it leaves untouched.

The electrons argument

Huang extends the thesis with a physical-substrate comparison: "in the industry governed by atoms the size of the industry is limited, there's only so much atoms you can move around, it's heavy. But the industry of AI is the industry of electrons. It's governed by quantum mechanics. It can be infinitely large."¹ Industries that move atoms, manufacturing, logistics, construction, are bounded by mass, friction, and geography. Intelligence delivered as electrons faces no equivalent ceiling, which is part of why demand for cheaper intelligence can expand across every sector at once rather than saturating the way a physical-goods market eventually would; the constraint on the supply side becomes compute rather than raw materials.

Why this reframes competition

Three consequences follow from treating AI as work rather than as tools. Competition changes shape: when software was tools, competitors were other software makers, but when AI performs a skill, it competes with human labor directly, and the comparison shifts from which software does a task best to whether the task costs less done by AI than by a person. Addressable market grows by orders of magnitude, since sizing the AI market as a fraction of technology spending measures the wrong denominator; the real denominator is global intellectual labor across every function in every industry. And every industry becomes an AI industry in this framing, not just companies that sell software: any domain with expertise encodable in data, from art to drug discovery to logistics, can train a model on that expertise and end up with an artificial version of it, which makes AI as broad as domain expertise itself.