Recently, many people suddenly noticed something:
DeepSeek cut prices again.
And not by a little.
The kind of cut where you just flip the table over. By the end of May, prices dropped to a quarter of what they were.
Many people's first reaction: Chinese AI companies are starting a price war.
But I increasingly feel that understanding this only as a "price war" is way too shallow.
Because what's really happening here might be this: tokens are becoming industrialized.
What does that mean?
For the past two years, the global AI world has operated under a quiet assumption: high-quality tokens are expensive.
Because: models are expensive, GPUs are expensive, training is expensive, electricity is expensive.
So everyone defaulted to the idea that AI must be a high-margin industry.
Until Chinese models started slashing prices like crazy.
And for the first time, many people discovered: tokens might actually be like steel, display panels, solar panels, lithium batteries — entering a terrifying process of industrial cost reduction.
Behind this story is something deeply Chinese.
What do I mean?
American AI companies often follow a path of "high performance, high margins, high valuation." A bit like luxury goods.
But once Chinese companies start competing, things tend to look different: "First, crush the cost."
Then: massive scale, infrastructure-ization, supply-chain-ization, engineering optimization, labor optimization, power optimization. Eventually grinding the entire industry into "cabbage-price industrial capability."
Over the past twenty years, China has done this repeatedly. Solar power, EV batteries, drones, display panels, e-commerce, high-speed rail... The pattern is roughly the same.
Early stage: others think it's high tech. Later stage: China industrializes it. End result: profits vanish, but production capacity blankets the world.
Today, tokens are starting to look more and more like this story.
Because tokens are not fundamentally mysterious. They are, in the end, "data processing capability produced by an industrial system." And what is an industrial system best at? Reducing costs.
So now an especially interesting dynamic has emerged: American frontier models may still maintain the strongest capability. But Chinese models are closing in fast — maybe a few months behind, maybe still a bit weaker in certain areas. But the price is already shockingly low.
So developers around the world are facing a very pragmatic choice: "Do I need the world's strongest, or do I need strong enough + ten times cheaper?"
This question is deadly.
Because in most of the business world, what ultimately matters is not "theoretical peak performance" but "overall cost-effectiveness."
As tokens get cheaper and cheaper, many AI applications that were previously "too expensive to run" suddenly become viable.
In the past, AI was like a five-star hotel. Now it's starting to look like tap water.
Developers used to worry: "Is this agent going to burn dozens of dollars a day?" Now the attitude is shifting to: "Whatever, let it run."
And so token consumption begins to explode further. Which in turn drives even larger data centers, cheaper inference chips, more aggressive engineering optimization. The whole system enters a kind of industrial flywheel.
The most interesting part is: what this competition ultimately comes down to may no longer be just the model.
It's about: who has cheaper electricity; who has more data centers; who has cheaper engineers; who has a more complete supply chain; who can better tolerate thin margins.
In other words: AI competition is increasingly looking like modern industrial system competition, not just lab competition.
Many people still think of AI as "a few brilliant scientists changing the world." But what it increasingly resembles is "an entire national industrial system collectively entering the field to produce tokens."
In the internet era, China's greatest strength was "application industrialization." In the AI era, what might be truly terrifying about China is: token industrialization.
And as token prices keep falling, developers around the world will ultimately vote with their feet. Because the vast majority of companies, in the end, have to do the math.
by Tuya
