the ai industrialists

Notes on Context, No. III

A dispatch from the Canton Fair, Guangzhou

Today I walked the Canton Fair alongside more than 86,000 international buyers, and I covered maybe a fifth of the floor. Everything there was physical. At one end were the building blocks: powders, resins, pellets, chemicals. Then came the molds, motors and production lines that shape them. At the far end were batteries, vehicles and humanoid robots throwing punches at the crowd. Under one roof stood the whole ladder, from raw material to machine.

Walking it, one idea kept coming back. The next great builders won't just write software. I call them AI industrialists.

The phrase gets used now and then as a jab at the people running AI labs. I mean something different. An AI industrialist is a person or team that puts intelligence to work in the physical world: building factories, running them, or serving the millions that already exist. Not the ones who make the models, but the ones who make them make things.

That's less new than it sounds, because AI was born industrial. The company that came to define this era doesn't sell software. NVIDIA sells chips. Marc Andreessen in The Techno-Optimist Manifesto, described AI as "literally making sand think." Every model we use runs on that sand, in data centers that need land, power, cooling and supply chains reaching back to halls like this one. The first wave of AI industrialists built the machines that make intelligence. The second wave will make that sand build.

Andreessen also wrote that the physical frontier, at least here on Earth, is closed. I think intelligence is reopening it.

Factories are where that starts. Code is abundant now. Context isn't, and a factory is full of it. A general-purpose robot knows how to do a task, but a robot on a production line has to understand its machines, its materials and the way that particular plant runs. Hayek argued that the most valuable knowledge lives at the edges, with the people closest to the work. On a factory floor, that's literally true. It lives in the sensors, the operational data and the heads of experienced operators. It doesn't download.

What's changed is timing. Robots, vision models and sensors have become cheap and capable at the same time. A line that once needed shifts of inspectors and planners can now run with cameras catching defects, AI predicting failures and one expert overseeing what used to take a department. Building a factory is still hard, and it still runs on energy. Whoever has cheap, clean power gets to industrialize. But running a factory no longer needs an army.

Some AI industrialists will build and run factories themselves. Many more will serve the ones that already exist: digitizing their operations, capturing what their best people know, training robots on their specific lines and bringing AI into every layer of their stack. Every factory sits on a mountain of context it has never used. Turning that into intelligence is an industry of its own.

Industrialists have always remade the world. The ones in the 1800s didn't invent steam or steel. They put them to work in mills, railways and ports, and that changed how nations lived. This time, the power is intelligence.

The tech capitals won't miss this. They'll lead it. Silicon Valley was named after a chip, and Shenzhen grew out of factory floors. Both started in hardware, and both are coming home to it with intelligence in hand.

But the bigger change may happen elsewhere. Places the last industrial wave skipped, small nations that import almost everything they use, finally get a real shot at making their own goods, because a small, AI-run factory makes sense at their scale.

So the question I'm carrying through the next four days isn't "what app should we build?" It's "what should we make, and where?"

Intelligence was born in a factory. Now it's going back to build more of them.