The AI Industry Suddenly Tears Apart: One Foot on the Brake, One on the Accelerator

"Why are global tech companies frantically adopting AI on one hand, while starting to restrict AI on the other?"

Microsoft has started cutting Claude Code licences, and Uber was reported to have burned through a full year's AI budget in just four months. On the other hand, Alibaba's Qwen3.7-Max suddenly shot to second place on the global coding leaderboard, even surpassing GPT-5.5.

One is stepping on the brake.One is stepping on the accelerator.Why has the AI industry suddenly developed this sense of tearing apart?

Because many people have not yet realised:

AI's biggest crisis is no longer "not smart enough."It is – "too expensive."

01What Is Truly Frightening Is Not That the Model Is Stupid, but That It Is "Too Capable"

Over the past two years, everyone has been competing on model capability.
Whose parameters are larger, whose inference is stronger, whose coding ability is fiercer.
But now, the big companies have suddenly discovered:

What is truly frightening is not that the model is stupid, but that the model is "too capable."

Previously, an AI reply to one sentence consumed a few thousand tokens.
Now, in Agent mode, it reads code, calls tools, runs tests, and reasons repeatedly on its own.
For a single request, token consumption can surge by hundreds or even thousands of times.

So a particularly absurd phenomenon has emerged:
Last year's corporate KPI: "Why aren't employees using AI?"
This year's corporate KPI: "Why are employees frantically using AI again?"

A new term has even emerged: "Token Maxxing."
It means employees have started treating AI as unlimited labour and throwing every task into it.
Writing code, writing documents, running tests, running analysis, fixing bugs…
As a result, AI does indeed do more work – and the bills start exploding.

Many companies have suddenly discovered:

The efficiency gained from AI may not be as fast as the speed at which tokens burn money.

02The Jevons Paradox: Lower Model Costs Have Not Saved Enterprises Money

Behind this is actually a classic economic phenomenon: the Jevons Paradox.
The cheaper the technology, the more humans use it.

Lower model costs have not saved enterprises money.
Instead, they have led everyone to "call AI infinitely."
In the end, total costs rise rather than fall.

This is also why Microsoft began to contract third-party Claude Code.
On the surface, it is an ecosystem adjustment.
In essence, AI costs have already begun to affect the profit structure.

03Domestic Models Have Suddenly Ushered in a Genuine Great Opportunity

And just as global big companies fall into "AI cost anxiety," domestic models have suddenly ushered in a genuine great opportunity.
Because the core of industry competition has already changed today.

What was compared in the past: who is the smartest.What will be compared in the future: who can achieve 90% of the effect at the lowest cost.

This is also what is truly formidable about Qwen3.7-Max this time.
It is not simply a rise in the rankings.
But rather achieving three things at once:

Stronger capability;Faster inference;Lower resource consumption.

Note that these three things combined mean something completely different.
Because the AI industry is moving from a "performance race" into an "industrialisation race."

In the past, everyone bought AI like buying a supercar.
Whoever is fastest, whoever is flashiest.
In the future, everyone will buy AI more like buying industrial equipment.
Whoever is stable.
Whoever is cheap.
Whoever saves energy.
Whoever can be deployed at scale.
Only then can they truly cover the entire industry.

Many people are still discussing: have domestic models surpassed GPT?
But the real question is actually:

If domestic models can cut Agent costs to one-tenth of GPT's, what will happen?

The answer may be: AI will no longer be just a toy for internet giants.
It will truly enter: software companies, MES systems, industrial automation, enterprise processes, and private deployment.

Because the vast majority of enterprises do not need "the world's smartest AI."
What they truly need is:

an AI employee that is smart enough, can work 24 hours a day, and has controllable costs.

This is where the two pieces of news truly connect.

04AI Is Moving from the "Show-Off Era" into a True Era of Large-Scale Industrialisation

The "cost crisis" at Microsoft and Uber is not an AI bubble.
Quite the opposite.
It means: AI is moving from the "show-off era" into a true era of large-scale industrialisation.

And in the industrialisation era, what determines victory or defeat has never been parameters.
But rather:

cost, efficiency, supply chain, and scalable delivery capability.

Many people are still staring at model leaderboards.
But truly smart companies have already begun researching:
how to control tokens.
how to compress inference costs.
how to make AI as cheap as water, electricity, and gas.

Because the next round of the AI war may not be about who is more like a human.

But rather who is more like "infrastructure."