The AI Bubble Has Two Problems — and They Could Collapse Together

Weak demand for AI services and a massive oversupply of computing infrastructure could turn today’s boom into tomorrow’s financial crisis

Photo by Enchanted Tools on Unsplash

When will this damn AI bubble burst?

How many warning signs does the market need before it finally admits that something is seriously wrong?

Honestly, I don’t know.

Nobody knows exactly when a bubble will burst. By the time it becomes obvious to everyone, the damage has usually already been done. That is what makes bubbles so dangerous. People keep investing, prices keep rising, and every warning sign is dismissed because “this time is different.”

But watching the AI industry right now is becoming increasingly strange.

Instead of slowing down, companies are continuing to spend more money, build more data centres, buy more chips, and sign even bigger contracts. The entire industry seems to be pressing the accelerator while heading directly toward a wall.

And yet, over the past few weeks, we have seen several developments that should have made investors extremely nervous.

How Does a Bubble Burst?

A bubble usually starts with excitement.

Investors believe that a new technology or industry is going to completely change the world. They start pouring money into it. Companies rush to build products and infrastructure to meet what they believe will be enormous future demand.

Prices rise.

More investors join in.

More companies build more capacity.

Eventually, supply can become much larger than actual demand.

That is when things become dangerous.

If investors suddenly realise that customers do not want as much of the product as expected, the entire story begins to fall apart. Companies start losing money. Investors become nervous. People begin selling their investments.

Then everyone rushes for the exit at the same time.

We have seen this happen before with tulip mania, the dot-com bubble, and the US housing market.

The difficult part is knowing when the bubble has reached that point.

Without a crystal ball, the best thing we can do is look for warning signs.

And recently, there have been some very big ones.

Elon Musk’s xAI Is Renting Out Its Compute

Elon Musk’s xAI has reportedly rented out a huge portion of the computing capacity at its Colossus data centre to competitors, including Anthropic and Google.

Anthropic reportedly agreed to rent around 222,000 GPUs for approximately $1.25 billion per month. Google agreed to rent around 110,000 GPUs for roughly $920 million per month.

The numbers are enormous.

What makes this particularly interesting is that the highest verifiable figure for operational GPUs at xAI’s data centre was around 230,000 as of September 2025.

If these figures are accurate, xAI may be renting out most of its own computing capacity to companies that compete with it.

That raises a very simple question:

Why?

Elon Musk has repeatedly argued that the future of AI will require far more computing power. He has even discussed building enormous networks of AI data centres.

So why would xAI rent out almost all of its existing capacity to competitors?

The obvious answer is that xAI may not be using enough of it.

If the demand for Grok and xAI’s own services is not high enough, renting out the hardware could simply be more profitable than using it internally.

And that is a serious problem.

The AI industry has spent years telling us that it cannot get enough computing power. Companies have been spending billions of dollars buying GPUs and building data centres.

But if xAI can suddenly rent out most of its infrastructure without seriously affecting its own business, perhaps the industry is not facing a shortage of compute after all.

Perhaps it has simply built far too much.

Meta May Be Facing the Same Problem

And xAI is not the only company apparently trying to make money from unused AI capacity.

Meta is also reportedly looking to sell or rent out excess computing power after failing to generate enough demand for some of its AI infrastructure.

This is particularly significant because Meta has invested enormous sums of money into AI.

The company has signed huge agreements involving AMD, CoreWeave and Nebius. It has also borrowed billions of dollars to expand its data centre infrastructure.

The basic strategy was simple:

Build a massive amount of AI infrastructure because demand is expected to explode.

But what happens if that demand does not arrive?

Suddenly, the company is left with billions of dollars of expensive hardware and data centres that are not being used enough.

At that point, renting out the excess capacity becomes an obvious option.

But there is another problem.

Who exactly is going to rent it?

There are only a handful of companies capable of spending billions of dollars on AI compute. OpenAI, Anthropic and Google are among the biggest buyers.

But if xAI and Meta are both trying to rent out excess capacity, while these companies are also building their own infrastructure, the market could quickly become flooded with computing power.

That would be a major problem for the entire AI industry.

The AI Bubble May Actually Be Two Bubbles

This is where things get even more interesting.

The AI boom may actually contain two separate bubbles.

The first is the obvious one: companies are valued at enormous levels based on the belief that AI will generate massive profits in the future.

The second bubble is the infrastructure industry supporting AI.

That includes GPUs, data centres and companies leasing computing power.

A lot of money is moving around inside this system.

AI companies need computing power.

Chip companies sell the hardware.

Cloud companies rent the infrastructure.

Investors fund the AI companies.

The AI companies then spend that money buying computing power from the same companies that helped fund them.

This creates a complicated financial cycle.

As long as money continues flowing through the system, everything looks fantastic. Revenues rise. Company valuations increase. More investment arrives.

But what happens when the demand stops growing?

The entire system becomes much harder to sustain.

AI companies such as OpenAI and Anthropic are still spending enormous amounts of money. Their future depends heavily on continued investment and access to increasingly large amounts of computing power.

If investors suddenly become less enthusiastic, or if customers stop paying for AI services at the expected rate, the whole system could become extremely fragile.

Oracle Is Another Warning Sign

OpenAI has reportedly agreed to purchase hundreds of billions of dollars worth of AI capacity from Oracle over several years.

Oracle, in turn, has invested heavily in building the infrastructure needed to support those contracts.

That is a huge bet on the future of AI.

The assumption is simple: AI demand will continue growing rapidly, and companies will happily pay for all this computing power.

But if investors truly believed that demand was guaranteed, you would expect Oracle’s stock to be performing extremely well.

Instead, Oracle’s share price has reportedly fallen sharply.

That does not automatically mean the AI bubble is about to burst.

Markets fall for many different reasons.

But when a company takes on enormous obligations based on the future growth of AI, and investors begin selling its stock, it is definitely something worth watching.

So When Does the Bubble Burst?

The frustrating answer is that nobody knows.

The AI industry may continue growing for years.

It is also possible that some companies will eventually generate enormous profits and prove the sceptics wrong.

But the current situation is becoming increasingly difficult to ignore.

We have companies building massive amounts of computing infrastructure.

We have AI companies struggling to generate enough revenue to justify their valuations.

We have major players reportedly trying to rent out excess computing capacity.

We have billions of dollars flowing between chip companies, cloud providers and AI laboratories.

And we have investors continuing to pour money into an industry that still has not clearly demonstrated how it will generate enough profit to justify all of this spending.

The biggest warning sign may be the sheer scale of the investment.

When companies are spending hundreds of billions of dollars based on the assumption that demand will arrive in the future, the consequences of being wrong become enormous.

The problem is that everyone involved has an incentive to keep the story going.

AI companies need more funding.

Chip companies need more customers.

Cloud providers need more demand.

Investors want their investments to become more valuable.

Nobody wants to be the first person to admit that the numbers may not add up.

So the game continues.

Everyone keeps building.

Everyone keeps investing.

Everyone keeps insisting that the real growth is still ahead.

Until one day, someone decides to leave.

And when that happens, everyone else may suddenly realise that the exit door is much smaller than they thought.

I cannot predict exactly when the AI bubble will burst.

Nobody can.

But I do know this: the industry is showing more and more signs of overbuilding, overinvestment and questionable demand.

The market may continue ignoring those warning signs for months or even years.

That is how bubbles work.

They do not burst simply because something is overvalued.

They burst when people finally stop believing that someone else will pay an even higher price.

And when that belief disappears, the entire game can change very quickly.

Sharing real experiences through words. Your feedback inspires me. Thanks for reading.