Zhipu climbed 33 percent in a single session. The trigger, depending on which desk you ask, was a major US lab tightening access abroad: fewer doors open overseas, so the money rotated to whoever still had a door open. That’s worth sitting with for a second, because the move is being read as enthusiasm and it isn’t.
Capital is water. Wall off one side of the river and it doesn’t evaporate. It finds the next channel down the slope. A third of a company’s value appearing overnight isn’t a verdict on that company. It’s the sound of money looking for the path of least resistance and finding it somewhere new. That is the whole argument here, and everything below is evidence for it: the question everyone is asking about whether the market can absorb the enormous private AI companies assumes the value sits inside those companies, and the week’s quieter news says it is already draining out of them into the layer underneath.
Can the Stock Market Swallow the Private Giants
Hold the Zhipu move next to the bigger question being asked in more serious rooms this month. The Economist put it directly in asking whether the stock market can swallow the largest private technology companies: the biggest rocket builder, the biggest model labs, businesses valued like small nations and owned like startups, with a handful of names on the cap table and no ticker for the rest of us. The honest answer keeps coming back uncertain, and uncertainty in that particular question is itself informative, because the default assumption for a decade was that demand for these listings would be effectively unlimited.
Then Bloomberg reported that a failed allocation of private rocket shares prompted a wider review in Korea. That’s a small data point, but small data has a way of being more useful than the big kind, because it is specific enough to be wrong and therefore worth something when it is right. Set the two next to each other and they tell a single story from opposite ends. The Economist asks in the abstract whether the appetite exists. The Korean episode is a concrete instance of someone testing that appetite and finding a limit. There’s a ceiling, and somebody just found the underside of it with their head.
So the framing everyone reaches for is whether these giants are too big to go public cleanly. It’s a good question. It’s also, I think, the wrong one, or at least not the one that decides anything, because it assumes the value lives inside the giants. It assumes that if you can’t fit the whale into the boat, the value stays stranded out at sea with the whale.
The Value Is Already Leaking Out the Bottom
Look at what shipped this week underneath the headlines. VentureBeat covered a startup debuting a universal orchestrator for AI agents talking to each other rather than only to us. Separately, a method called PixelRAG claimed it could beat text parsers on accuracy while cutting an agent’s token costs tenfold. Neither of those made the front page. Neither is a giant. Both are plumbing.
But that’s the whole point. The first wave of this technology concentrated value in the model. The bigger the model, the bigger the moat, the bigger the private valuation nobody can buy into. The second wave is doing the opposite. It’s pushing value out of the model and into the connective tissue: into how agents coordinate, into how cheaply a task can be run, into the unglamorous question of what it costs to do the thing once you have decided to do it.
Notice how neatly those two releases fit together, because they are halves of one machine. An orchestrator only matters if running many agents is affordable, and a tenfold cost reduction only matters if you have something worth running many times. Each makes the other economically sensible, and neither requires permission from a frontier lab. That is what a second wave looks like: capability assembled sideways out of parts, by people who did not need to raise a nation-sized round to participate.
Why Cheaper Tokens Reprice the Floor
Ten times fewer tokens isn’t a feature. It’s a repricing of the entire floor. When the cost of doing the work drops by an order of magnitude, the work stops being a luxury good and starts being infrastructure, like water pressure or a dial tone, something you only notice when it’s gone. That transition changes what gets built, because it changes what is worth trying. At a high per-task cost, you only run an agent on work you already know is valuable, which means you only ever confirm what you suspected. At a tenth of that cost, you can run it on work that probably isn’t valuable, and that is where the surprises live.
This is why the rerouting matters more than the rocket. When one lab closes a door and a third of a valuation reappears somewhere else the next morning, it’s telling you the thing being traded was never scarce in the way everyone assumed. Genuine scarcity does not reprice that fast. What repriced was access, not capability, and access is a policy, not a property of the technology.
The scarcity at the top, the private giant nobody can own, creates abundance at the edges, because every closed door funds three open ones. Block the obvious channel and you don’t stop the flow. You just teach it a new route, and the new route is usually cheaper, because it was built by people who had to make it cheap in order to exist at all.
What the Giants Actually Own
There’s a version of this where you feel sorry for the public investor, locked out of the best companies of the decade. I’m not sure that’s the right read either. What the giants own is the headline. What they increasingly don’t own is the marginal cost of the work, which is collapsing in labs and startups whose names won’t trend.
The orchestrator that lets agents hand tasks to each other doesn’t care which model is the biggest. It cares which one answers, and how cheaply. That’s a filter, not a fan club, and filters route by relevance rather than reputation. A company whose advantage depends on being chosen by a routing layer is in a very different competitive position from a company whose advantage depends on being admired, and the difference shows up in the margin long before it shows up in the coverage.
None of this means the giants are in trouble. They have real assets, real research, and the ability to keep moving the frontier. It means the frontier is not the same thing as the market, and a business can own the first while progressively losing its grip on the second. That is not a dramatic failure. It is the ordinary way a leading position turns into a respected one.
The market keeps asking whether it can swallow the whales. It’s the wrong appetite to worry about. You don’t have to eat the whale to live off the ocean it swims in. The value isn’t trapped in the four or five companies too large to list. It’s dissolving into the water around them: into the cheaper token, the agent that talks to another agent, the door that opens the moment a louder one shuts. The people watching the whales will keep watching the whales. They’ll be very well informed about the part of this that matters least.
The giants will go public or they won’t. Either way the interesting money has already stopped waiting for them. It went looking for the next channel down the slope, and the slope, it turns out, runs everywhere the giants forgot to fence.
Leave a Reply