Every piece of writing about artificial intelligence arrives at the same narrow point: the chip that does the math. Not the ordinary processor in your laptop, but something called a GPU. Understanding what makes this chip different explains why a video game company suddenly became more important than most tech giants.
One Craftsman Versus an Assembly Line of Workers
The ordinary processor in your laptop is like a craftsman who builds a whole table by himself: cuts the boards himself, assembles them himself, varnishes them himself. He can handle any step, whatever it is, and jump between completely different tasks. But he's one man. He works things one after another, no matter how fast his hands move.
A GPU (graphics processing unit, literally a chip for processing images) is the opposite. Instead of one craftsman, you have an assembly line of a hundred workers, each of whom knows only one simple move: turn a screw. On its own, none of them is worth as much as the craftsman. But when you need to turn a pile of identical screws, the assembly line finishes the job while the craftsman is still cutting the first board.
Artificial intelligence does exactly that kind of work. To guess the next word in a sentence or the next pixel in a photo, it repeats the same simple multiplication of numbers, over and over, across huge tables at once. That's why the chip built to move graphics in video games turned out more useful for AI than the chip built for office tasks.
NVIDIA didn't invent artificial intelligence. It built the assembly lines and sells access to them. That's why the company became so important: not so much because it writes the best AI programs, but because almost anyone who writes such programs runs them through its machines to train them. It's the shovel seller in a gold rush: it doesn't dig itself, it sells the tool to everyone who digs, and it profits no matter who finds the vein. If you're the only one in town building assembly lines big enough for this kind of work, everyone who wants to move fast comes to you and pays whatever you ask.
That's why Google built its own assembly line. Its chips are called TPU (tensor processing unit, a chip specialized for exactly this kind of mass calculation). Amazon did the same with a chip named Trainium. Both companies decided it was cheaper to build their own factory than to keep paying NVIDIA for access to its. Most other companies in the industry don't have that choice. They pay.
How the Stage Is Set
When you hear that some company made a breakthrough in artificial intelligence, I ask first whose machines it happened on. The idea on its own is cheap. Without an assembly line, it stays on paper.
So the real battle in this industry isn't which program is smarter this week. It's who controls the assembly lines. For now, almost everyone pays the same factory owner, and that's exactly why the price of his stock interests more people than artificial intelligence itself.