China Is About To Pop The AI Bubble

China Is About To Pop The AI Bubble

The AI boom has created one of the biggest investment stories of the decade.

Microsoft, Nvidia, Amazon, Google, Meta and a growing list of companies are spending hundreds of billions of dollars building data centers, buying chips and developing increasingly powerful AI models.

The story is simple:

AI is the next great technological revolution, and the companies building it are going to make trillions.

But that story depends on one enormous assumption.

That the rest of the world will have no alternative.

China may be about to change that.

Because while the United States is spending close to a trillion dollars building the most expensive technology infrastructure in history, China is taking a very different approach.

It is building cheaper models, open-sourcing them and pushing them into the market at a fraction of the cost.

And if businesses discover that they don't need the most powerful AI in the world — they just need AI that's good enough — the economics of the entire AI boom could change.

This doesn't mean AI is useless.

It means AI could be incredibly useful while AI stocks are still massively overvalued.

That's the distinction investors need to understand.

---

The AI Boom Is Built On A Story

Look at what has happened to the stock market.

Nvidia became one of the most valuable companies in the world.

Microsoft, Amazon, Google and Meta have committed enormous amounts of capital to AI infrastructure.

Data centers are being built at a scale that would have seemed impossible only a few years ago.

And investors have largely accepted the same basic argument:

Spend now.

Build everything.

The profits will come later.

The problem is that nobody knows exactly what those profits are going to look like.

Companies are spending billions on AI, but the return on that spending remains difficult to measure.

If a company spends $50 million on AI, its board eventually has to ask a very simple question:

What did we get for the $50 million?

And increasingly, that question doesn't have a clear answer.

AI can make employees faster.

It can automate tasks.

It can write code.

It can analyze information.

But translating all of that into a measurable return on investment is much harder.

And that creates the first problem with the AI boom:

The technology is advancing faster than the business model.

---

AI Has A Problem Traditional Software Didn't

Traditional software has one of the greatest business models ever created.

Microsoft spends money developing Excel.

Once Excel exists, selling it to another customer costs almost nothing.

One more customer means more revenue without a proportional increase in costs.

That's why software companies can eventually generate extraordinary margins.

AI is different.

Every time you ask an AI model a question, somebody has to pay for the computing power required to answer it.

That means electricity.

GPUs.

Data centers.

Cooling.

Networking.

Infrastructure.

And increasingly expensive computing capacity.

In other words, AI has a cost attached to every interaction.

It's less like selling a copy of Microsoft Excel and more like running a restaurant.

Every time another customer walks through the door, you have to prepare another meal.

Except AI companies have an even bigger problem:

Some of them are losing money while serving those meals.

---

OpenAI's Numbers Show The Problem

The source cites OpenAI burning roughly $20.9 billion in 2025.

That is an extraordinary amount of money.

And the question isn't simply whether OpenAI can survive those losses.

The bigger question is whether the economics eventually improve enough to justify the enormous valuations being placed on the entire industry.

For decades, investors have been trained to accept losses from technology companies because of one assumption:

Scale eventually creates better margins.

Amazon lost money for years.

Technology companies spent heavily before becoming enormously profitable.

But AI has a problem that traditional software didn't have.

The more people use the product, the more computing the company has to pay for.

If revenue and costs continue rising together, the massive software-style margins investors are expecting may never arrive.

And if those margins don't arrive, the valuations become much harder to justify.

---

Then China Entered The Story

This is where things get interesting.

The United States is spending extraordinary amounts of money trying to build the world's most advanced AI.

China is spending dramatically less.

The source puts U.S. AI investment at approximately $764 billion this year, potentially rising toward $1 trillion next year.

China, by comparison, is estimated at roughly $102 billion this year and $123 billion next year.

That's roughly a 10-to-1 spending advantage for America.

But here's the problem:

AI doesn't necessarily need to be the smartest possible model to be commercially useful.

A company doesn't necessarily need the world's most intelligent AI to answer customer-service emails.

It doesn't need the world's most powerful model to process insurance claims.

It doesn't need frontier-level intelligence to summarize documents or automate repetitive office work.

It needs something that works.

And if China can provide that capability for dramatically less money, the economics change.

---

The Price Gap Is Getting Difficult To Ignore

Consider one example from the source.

The same coding task was given to two AI models.

Claude Opus, one of Anthropic's leading American models, reportedly cost around $2.33.

The Chinese GLM model cost approximately $0.31.

The two models took roughly the same amount of time to complete the task.

That means the Chinese model was roughly 7 times cheaper.

And the broader point is even more important.

China isn't relying on one model.

Models such as DeepSeek, Qwen, Kimi, MiniMax and GLM are increasingly appearing throughout the global AI rankings.

The United States may still have the world's most capable frontier models.

But China is producing a huge number of models that are good enough for real-world business applications.

And that could be much more important than having the absolute smartest model.

Because businesses don't buy intelligence for bragging rights.

They buy it because it saves money.

---

China's Secret Weapon Is Distillation

So how can China compete while spending so much less?

One answer is distillation.

Training a frontier model from scratch can require billions of dollars and enormous computing resources.

But smaller models can learn from the outputs and capabilities of models that already exist.

Think of it as taking an enormous textbook and compressing its useful information into a much smaller book.

The result doesn't necessarily need to understand everything the original system understands.

It just needs to perform the tasks customers actually care about.

And China has been particularly aggressive about making these models open-source.

That creates a strange dynamic for the American AI industry.

The United States spends billions developing increasingly sophisticated models.

Chinese companies can study the results, compress them into cheaper systems and then distribute those systems to developers around the world.

So America's enormous AI spending could unintentionally be helping create the foundation for a cheaper global AI ecosystem.

---

The Real Threat Isn't That China Beats America

This is the part investors need to pay attention to.

China doesn't necessarily need to build a model that's better than America's best model.

It only needs to build one that's good enough and dramatically cheaper.

Imagine two AI models.

One is 100% as capable and costs $1.

The other is 90% as capable and costs $0.10.

Which one does a business choose?

For many applications, the answer is obvious.

And that's why China's AI strategy creates such a serious problem for the American investment thesis.

Because the stock market isn't pricing AI as though it will merely be useful.

It's pricing AI as though American companies will capture an enormous portion of the economic value created by it.

That requires high prices.

High margins.

Huge demand.

And limited competition.

China threatens the last two.

---

The Trillion-Dollar Infrastructure Bet

Now look at what companies are building.

Microsoft, Amazon, Google and Meta are committing enormous amounts of money to data centers and computing infrastructure.

Nvidia is selling the chips needed to power that infrastructure.

Cloud companies are buying those chips.

AI companies are renting the computing capacity.

And the entire ecosystem is feeding on itself.

That creates an important question:

How much of this demand ultimately comes from genuine end customers?

The source highlights Nvidia's relationships with smaller cloud providers, sometimes called "neoclouds," which borrow money to purchase Nvidia GPUs and then rent computing capacity back into the broader ecosystem.

If companies within the same ecosystem are financing one another's expansion, reported demand can look much stronger than the underlying end demand.

That doesn't mean the demand is fake.

It means investors need to determine how much of it is sustainable.

---

The Market Is Already Sending A Strange Signal

There's another detail worth watching.

The companies selling the picks and shovels of AI have been rewarded enormously.

Chipmakers and infrastructure providers have seen huge increases in their valuations.

But the companies spending the money — the hyperscalers — haven't received the same level of market reward.

That's strange.

If companies are spending hundreds of billions of dollars on AI because those investments are going to generate enormous profits, eventually investors should expect those profits to show up in the companies doing the spending.

But that hasn't happened at the same scale.

And companies aren't exactly rushing to disclose their AI revenue separately.

They report cloud revenue.

Advertising revenue.

Subscription revenue.

Other business lines.

But investors still don't have a clean picture of how much incremental revenue AI itself is generating compared with the enormous amount being spent on it.

That makes the investment story difficult to evaluate.

---

So When Does The AI Bubble Pop?

Here's where the comparison with the dot-com bubble becomes useful.

The bubble doesn't necessarily pop when companies stop spending.

It can pop before that.

During the dot-com era, companies continued spending heavily on internet infrastructure even after technology stocks had already begun collapsing.

Why?

Because the infrastructure spending was based on a story.

Once investors stopped believing the story, the valuations collapsed.

The spending followed later.

The same thing could happen with AI.

The trigger may not be some massive technological failure.

It could be one relatively boring sentence on an earnings call:

"We are moderating the pace of our AI infrastructure investment."

That's it.

If one major hyperscaler cuts its AI capital expenditure and Wall Street rewards the decision, every other company suddenly has permission to do the same.

The AI industry is heavily influenced by competition.

Nobody wants to be the company spending less than its rivals while AI is booming.

But if the first company cuts spending and its stock rises, the incentive changes.

Suddenly everyone wants to know:

Why are we spending this much?

---

Watch The Debt Market

There is another signal investors should watch closely:

Credit spreads.

Stock markets can run on narratives.

Bond markets care about something much simpler:

Will I get my money back?

When lenders become nervous about a company's ability to repay its debt, they demand higher interest rates.

The difference between a company's borrowing cost and the risk-free government rate is called its credit spread.

When spreads are tight, investors are comfortable.

When spreads widen, investors are becoming nervous.

Right now, credit spreads remain relatively calm.

But history shows that calm credit markets don't necessarily mean everything is healthy.

In 2007, credit markets were still relatively relaxed even as problems were already developing in the housing market.

The lesson isn't that an AI crash is imminent.

The lesson is much simpler:

Credit markets can be wrong.

---

The Most Important Number Might Be The Price Of AI

There's one final signal that deserves attention.

The price businesses pay for AI tokens has been falling.

According to the source, the AI token expenditure index was nearly 20% below its May peak.

And that is happening while the world's largest companies are spending more than ever on AI infrastructure.

Why would the price of AI fall during the biggest AI buildout in history?

One explanation is that customers are moving toward cheaper models.

And that's exactly where China has an advantage.

If businesses increasingly choose cheaper models that deliver most of the same functionality, the total amount of AI usage could increase while the amount businesses are willing to pay per unit of intelligence falls.

That's great for consumers.

It may not be great for the companies that spent trillions building the infrastructure.

---

China Could Pop The Bubble Without Even Trying

This is the most important conclusion.

China doesn't need to destroy American AI.

It doesn't need to create a model that is dramatically smarter than ChatGPT.

It doesn't need to beat Nvidia.

It simply needs to make AI cheap enough that the enormous profits investors are expecting become impossible to achieve.

That's a completely different kind of competition.

The United States is competing to build the most powerful AI.

China is increasingly competing to make AI cheap, accessible and everywhere.

And if the second strategy wins, the first strategy becomes much less profitable.

---

The Bubble Doesn't Need AI To Fail

This is where the entire story comes together.

AI can transform the economy.

AI can make companies more productive.

AI can become one of the most important technologies of the century.

And AI stocks can still crash.

Those statements aren't contradictory.

The dot-com bubble didn't happen because the internet was useless.

It happened because investors paid prices based on expectations that were far ahead of the profits companies could actually generate.

The same thing could happen with AI.

The technology can be real.

The revolution can be real.

And the valuations can still be wrong.

The biggest danger for the AI boom isn't necessarily that China builds something better.

It's that China builds something good enough for 10% of the price.

Because once businesses realize they don't need the most expensive AI in the world, the trillion-dollar spending race starts looking very different.

And when investors stop believing that today's spending will become tomorrow's profits, the bubble doesn't need a dramatic explosion.

It just needs the story to change.