Tech
ChatGPT hits 1 billion users — why AI is a winner-take-all game
This week SpaceX became the largest IPO in history and OpenAI quietly filed to go public. The frenzy looks new. The rulebook is 30 years old.
The numbers are dizzying. ChatGPT became the fastest product ever to reach a billion monthly users — about three years from launch. SpaceX began trading at a ~$1.77 trillion valuation. Anthropic is nearing $50 billion in run-rate revenue. Two frontier AI labs are now in the IPO pipeline at once.
It is tempting to explain all this with hype. But there is a quieter, sturdier explanation, and Kevin Kelly wrote it down in 1994 in Out of Control. He called it the network economy: in a connected world, value comes not from owning things but from connection itself.
The fax-machine rule
One fax machine is worthless — there's no one to fax. The second machine makes the first useful. The millionth makes every machine before it more valuable. Kelly's point: networks obey increasing returns. The more people use a thing, the more reason there is to use it, which pulls in more people. Success breeds success — what he listed as one of his "Nine Laws of God": cultivate increasing returns.
A large language model is a fax machine on steroids. Every user generates feedback, edge cases, and data that make the model a little better — which attracts more users. That loop, not the demos, is why one or two players can swallow a market. And it does not add up; it multiplies. In Kelly's words, the network economy runs on math where the curve bends upward: each new node doesn't just add itself, it adds a connection to everything already there. Industrial economies obeyed diminishing returns — the second factory was a little less profitable than the first. Network economies invert that. The second billion users are cheaper to serve and worth more than the first, because they thicken the web that all the others are already standing on.
This is why the AI race feels less like a footrace and more like a snowball rolling downhill. The leader isn't merely ahead; the leader is accelerating away, because being ahead is itself the engine. That is the uncomfortable heart of "winner-take-all": it is not a moral judgment about the winner being best, only a structural fact about which way the slope tilts.
The core idea, in one line
In a network economy, value lives in connections, not in what you own — so the system tilts toward winner-take-all.
Has this happened before?
Kelly wasn't theorizing in a vacuum. He was watching the fax machine and the early internet, and the pattern he named had already played out a dozen times. The fax is the cleanest case: the machine sat half-dead in catalogs for a century until enough offices had one that owning one became unavoidable, and then adoption went vertical almost overnight. The same shape repeated with the telephone, with credit-card networks, with the QWERTY keyboard that locked in not because it was best but because everyone already knew it.
The software era made it sharper still. Windows didn't win because it was the most elegant operating system; it won because applications were written for it, which drew users, which drew more applications — the flywheel again. By the time a rival had a better idea, the better idea had nowhere to land. The lesson Kelly drew is the unsettling one: in a network economy, the best product frequently loses to the most connected one. Quality sets the loop spinning, but connection is what makes it unstoppable.
So when ChatGPT crosses a billion users faster than any product in history, the right reaction isn't only awe at the technology. It's recognition. We have seen this movie before — just never at this speed, and never with a product that gets materially smarter every time someone uses it.
Why the giants still lose the edge
Kelly's second warning is for the winners. Innovation, he argued, happens at the edge — at the messy boundary, not the optimized center. Big companies get trapped on a "local peak": very good at what they already do, unable to climb down and cross the valley to a better hill. That is why a two-year-old lab can blindside a trillion-dollar incumbent. The incumbent is optimizing; the upstart is exploring.
This is the catch hidden inside increasing returns: the same loop that crowns a winner can also calcify it. A dominant model accumulates users, revenue, and infrastructure it cannot afford to disrupt, so it keeps polishing the hill it is already on. Meanwhile the real breakthrough is forming at the fringe — in a lab small enough to bet the whole company on an idea that would be irrational for the giant to chase. That a firm barely two years old can reach a $50-billion run rate is not a fluke. It is the edge doing exactly what Kelly said it does.
He also saw the ownership model dissolving into an access model. You don't own the intelligence; you rent access to it by the token. The IPO wave is, in part, investors pricing that shift — betting on who will own the rails of an access economy. When SpaceX prices at $1.77 trillion and two AI labs queue for the public markets at once, the market isn't just valuing today's revenue. It is valuing position on the rails everyone else will have to rent.
What it means for you
Kelly's framing gives you a sharper lens than "AI is booming." Ask instead: where are the increasing-returns loops, and who controls them? A valuation isn't betting on a clever demo. It's betting that a feedback loop has already started to spin — and that, once spinning, it is very hard to stop.
You can carry the same lens past the stock ticker. As a builder, the question stops being "is my model the smartest" and becomes "does every user make my product a little better for the next one" — because without that loop, quality alone will not save you. As a user, notice that you are not just a customer but fuel: the tool you pick today is being made better by the crowd that picked it, which is exactly why the crowd keeps growing. And as someone trying to read the headlines, treat each new billion-dollar figure not as a verdict on intelligence, but as a bet on which flywheel has already started to turn. The technology is what makes the demo dazzle. The network economy is what decides who is still standing in ten years.
Framework from Kevin Kelly, Out of Control (网络经济 / 递增收益 / 边缘创新), via the Out of Control knowledge skill. News figures: June 2026 reporting. Commentary, not investment advice.