Many blamed OpenAI’s delay for the tech selloff. The real story might be somewhere else: a competitor doing almost the same thing for far less.
Published on July 11, 2026 · by Alex · Artificial intelligence · China · Valuations
Tech stocks keep falling and money keeps rotating, and much of the market attributed it to OpenAI's possible listing delay.
That may not be the story. While everyone watched OpenAI and SpaceX, GLM 5.2 appeared in China — a model that, according to several analysts, delivers performance very close to the best American models at a fraction of the cost.
For the last few years, large companies justified billions in investment with a promise: ever more powerful models. The argument worked while the quality gap was obvious.
Now customers are starting to ask a different question: is it worth paying several times more for an improvement you barely notice? And the moment that question appears, the whole business model changes.
| If the customer says no… | Consequence |
|---|---|
| Stops buying the expensive service | Less revenue for the large developers |
| Runs alternatives on its own servers | Less dependency and less recurring revenue |
| Cuts spending per task | Less need to keep investing at the same pace in data centres and chips |
Here is the part worth understanding. The hit does not land only on whoever sells the model. It lands on the entire chain built assuming a given rate of investment: chipmakers, memory, data centre builders and the energy companies that power them.
Which is why it is no accident that the Magnificent Seven fell hard on several days. The market stopped asking who has the best artificial intelligence and started asking who can make money from it.
If low-cost models keep closing on the performance of the most advanced ones, the valuations of the next big listings have to come down. Not out of pessimism: out of arithmetic. A company selling something its competition nearly gives away is worth less than one without that competition.
And that is the real concern among large funds: that we may not be watching a technology race, but the start of a price war nobody had in their models.
A cheaper model with close performance does not mean an equivalent one. Differences tend to show up in hard tasks, in reliability and in support, and that matters a great deal to a corporate customer. The open question is not whether cheap models are as good: it is whether they are good enough for most uses. That answer defines how large a premium market is left.
A Chinese model that, according to several analysts, delivers performance very close to the best US models at a fraction of the cost. It matters because it forces customers to ask whether paying several times more is worth an improvement they barely notice.
Because the whole chain was built assuming a certain pace of investment: chips, memory, data centers and the power that feeds them. If customers cut their spend per task, the need to keep investing at that pace drops, and the hit reaches every link at once.
It forces valuations down — not out of pessimism but arithmetic: a company selling something its competition offers at half the price is worth less than one without that competition. The market stopped asking who has the best AI and started asking who can make money from it.
This article is the written version of the Saturday analysis.
Watch the video on TikTok · Follow the channel
