We are not just in the middle of the experiment. We are absolutely being experimented on. Maybe, unintentionally. But even if you don’t have massive investments in it, you might still get affected if the bubble bursts.
Yep, the AI bubble. (I refuse to call it SI, just in case you were wondering). And I am not even talking about the extinction of humanity. Not enough empirical data to support that. But there is something else that is frightening and that we have enough observational data to be scared of.
The economics.
OK, what evidence do we have, why should it worry us, and have we seen similar bubbles burst right in front of us before.
History of Bubbles
The Internet Was Real. The Expectations Ran Ahead.
It’s the late 1990s. Sorry, if you didn’t exist yet. But some of us remember. The Internet was a new, amazing, almost magical thing. Investors poured money into internet businesses and telecommunications infrastructure like there was no tomorrow.
Ah, that Tomorrow. We always think it’s in the future, but it surprises us. It waits to jump at us from behind the corner to become the Today and ask for the dividend we promised it yesterday.
But yesterday, we were full of hopes and dreams. We had no idea they were just that – hopes and dreams.
That’s exactly what happened with the dotcom bubble. The technology was transformative, but investors dramatically overestimated how quickly demand and profits would justify the investment.
The NASDAQ Composite rose about 215% between September 1998 and March 2000. It subsequently fell roughly 78% from its peak to its trough over the next two years. The collapse destroyed enormous amounts of market wealth, although the banking system remained comparatively unscathed.
The telecommunications industry massively expanded fiber-optic networks in anticipation of rapidly growing internet traffic. But by 2000 and 2001, demand had flopped, prices collapsed, and a lot of telecom companies went under.
Left behind were miles of fiber-optic networks. We repurposed them, of course, but the fact that an asset remains useful does not mean the debt used to build it will be repaid in full.
Another great example of a burst bubble is even closer to home. Literally.
The Mortgage Bubble
In the early and mid-2000s, US housing prices rose fast. Very fast. Many believed that house prices would keep rising, or at least would not fall enough to cause serious trouble. Refinancing seemed like an attractive option for both lenders and borrowers.
But lenders were too optimistic about borrowers’ ability to repay. Optimistic enough to help set the stage for a devastating wave of defaults.
Yes, lenders assumed risks. They had been distributed throughout the financial system. But they were not eliminated. And when the housing market crashed, risks backfired and asked for the money.
When US housing prices started declining around 2006, refinancing became harder, and mortgage-backed securities lost value, this downward spiral sucked in many players – far beyond the original borrowers and lenders. And didn’t stop.
The crisis spread from housing into the wider financial system, contributing to the Great Recession. Unemployment skyrocketed, millions of families suffered.
Everything is connected in the economy.
Did it help to have the borrowings backed up by one of the most tangible products out there, the real estate? Not in the slightest.
The Warnings
So why are these two bubbles such useful examples for our inquiry about the next one – the AI bubble?
The dotcom bubble wasn’t simply a case of investors failing to predict which internet companies would succeed. The mortgage crisis wasn’t simply a case of people failing to predict falling house prices. The underlying failure was that the possibility of extraordinary future success was treated as evidence that such success was the most probable, and a chain of financial decisions built on assumptions that depended on one another.
The system treated a favorable scenario as pretty much the only one possible.
Hopes and Dreams. Hopes and Dreams.
So, Natasha, let’s talk about AI already!
Fair enough.
The AI Bubble. Is It Really a Bubble? And What Happens if It Bursts?
Here is how I define a financial bubble, drawing on the economic literature:
A financial bubble develops when excitement about an investment opportunity drives asset prices and financial commitments beyond what realistic expectations of future economic returns can justify. The opportunity may be genuine, the market may be growing, and the underlying technology or asset may have real value. But investors begin making increasingly large commitments based on optimistic assumptions about future demand, revenue, and profits, while inadequately accounting for costs, risks, and less favorable scenarios. As expectations reinforce one another, investment can grow far beyond what sustainable returns will support. When the expected returns fail to materialize, or confidence in them collapses, asset prices fall and investors suffer losses.
Or in 6th grade language: A financial bubble happens when excitement pushes investment prices far above what the profits they can realistically generate are worth.
So is AI already a bubble?
It sure already shows signs that it is or will soon become one.
Let’s take a look.
- Excitement about enormous future potential? Check.
- Financial commitments based on projected future returns? Check.
- Evidence that investment could exceed what future returns can justify? Check.
- Insufficient consideration of costs and downside scenarios, including local opposition, electricity constraints, and permitting delays? Check.
- Warning signs around infrastructure costs, debt, and uncertain returns, with expectations reinforcing one another? Check.
- Potential for a sharp reversal and investor losses? Check.
If it looks like a bubble risk, walks like a bubble risk, quacks like a bubble risk… It’s the freaking BUBBLE RISK!
Well, don’t take my word for it.
A July 2026 research paper from the Bank for International Settlements estimates that competitive pressure could drive AI investment to roughly 1.5 times the economically efficient level under its baseline assumptions. This is just a model result, not proof. The paper also warns that debt and connected investments could amplify losses if expected returns disappoint.
The BIS’s broader 2026 assessment similarly highlights the risk that AI-related spending could outrun earnings and free cash flow, leaving companies vulnerable if productivity gains or revenue fall short of expectations.
The risk is clear.
Some will say: Boo your risks. Isn’t AI already generating revenue, providing useful services, and improving productivity? That’s true, of course. The main question remains – will the profits eventually generated justify the money committed today?
And that’s not the only question that bothers economists and your humble servant.
- What if businesses don’t use AI as much as investors expect?
- What if AI companies can’t earn enough money to cover the cost of running their products and building data centers?
- What if AI technology improves so quickly that expensive chips and data centers become outdated before investors earn their money back?
- What if investors stop putting money into AI companies, and those companies can’t borrow more or repay their debts?
- What if one major AI company runs into trouble and pulls down the other companies, investors, and lenders that depend on it?
Those outcomes would not mean AI had failed. They could mean the industry needed less infrastructure than investors expected.
People made enormous financial commitments based on expectations about the AI future, while underestimating what’s going to happen if those expectations turn out to be wrong.
Will the AI Bubble (if There is One) Burst?
Unlike some AI investors, I am not going to make an assumption I can’t back with data. I will not speak definitively even if it costs me your attention.
We can’t know for sure.
But what we can do is SPECULATE!
The AI data center buildings and connections may survive. Land, substations, power connections, cooling systems, and network access may retain value. A facility could potentially be adapted for conventional cloud computing, enterprise workloads, or other computing services. But location, design, available power, and conversion costs will determine how practical that is.
The equipment may lose value much faster than the buildings and other infrastructure. Specialized accelerators and servers can become less competitive as hardware improves.
If a data center cannot attract enough customers to cover its financing and operating costs, the owner may default, sell at a discount, or restructure the debt.
And let’s not forget the impact on the communities. Some oppose data center projects over electricity demand, water use, land use, or other local impacts. Others accept them under specific conditions. But what happens to those agreements if these facilities stop being… well, data centers?
In Case Of Bubble Bursting, Look For Exit
So, what may happen if shait hits the fan? And how will it affect you and me? Who will pay?
Investors.
Shareholders in AI companies, chipmakers, cloud providers, and related businesses could lose a lot. Venture capital and private-equity investors could see their stakes written down or become worthless.
Banks and other lenders.
Banks may face losses on direct loans and indirect exposures through other financial institutions. Private-credit funds, insurers, pension funds, and investment funds may also be affected.
Workers.
Companies facing losses may cancel projects, freeze hiring, cut contractor spending, or lay off employees. The damage could reach construction firms, equipment suppliers, utilities, and local businesses that benefited from the infrastructure boom.
You and Me.
Even people who never bought an AI stock may have exposure through retirement, pension funds, and index funds. If it goes down hard enough, job security could be shaken, and credit could become less accessible.
The economy.
If businesses cancel projects, investors become more cautious, and lenders tighten credit, the loop may only become stronger. Public debt may go up.
Governments may have to support struggling banks and businesses, help people who lose their jobs, or deal with lower tax revenues. That means less money coming in and more bills to pay.
Those scenarios may not yet be likely, but they are already realistic.
Could an AI crash break the economy completely?
A financial market correction by itself is not the same as the collapse of an economy. The dotcom crash caused serious losses and contributed to a recession, but it did not start a major banking collapse. The economic impact of an AI investment bust will depend on how large the losses are and how they spread through the financial system.
One plausible scenario is a serious slowdown in economic growth, which can still be devastating for people who lose their jobs, savings, or businesses.
Is This Dangerous Yet?
To see whether the AI boom is becoming disproportionately risky, we need to track the Big Five: Microsoft, Alphabet (Google), Amazon, Meta, and Oracle. We track them because they are the biggest publicly owned businesses that invest massively in AI infrastructure development. You can also track OpenAI and Anthropic, and AI infrastructure enablers Nvidia, AMD, and Broadcom. Focus on the following signs and track how they change over time.
Investment grows faster than results
Which means that companies spend more and more money on AI, but the money they earn from it doesn’t grow as quickly.
How to spot it. Read the Big Five’s quarterly reports. Compare how much they spend on new AI equipment and data centers with how much money their business brings in. Listen to their earnings calls (yes, you can). Are they showing real results, or mostly promising that the money will come later?
Debt replaces cash
The Big Five borrow more money because they aren’t generating enough cash to pay for their investments.
How to spot it. Look at their quarterly reports. Is the company’s debt growing? Is it spending more on interest? Is it borrowing again and again to keep its projects going? If a company needs more loans just to keep expanding, that’s a warning sign.
The Big Five and Co become financially tied together
AI companies, investors, suppliers, and customers depend on one another financially. If one gets into trouble, others may suffer too.
How to spot it. Follow the news about major AI deals and investments. Look for cases where one company invests money in another, and that company then uses the money to buy its products or services. Also, check whether a supplier depends heavily on just one or two big customers. If one company goes down, how many others will it drag along?
More infrastructure than customers need
The Big Five build data centers and buy equipment faster than customers’ needs grow. They may end up with expensive facilities that don’t bring in enough money.
How to spot it. Follow news about new data centers, delayed projects, canceled deals, and unused facilities. Compare what companies plan to build with what they actually complete and how much business they generate. Announcing a huge investment doesn’t mean the investment will be profitable.
Investors stop providing money
A business depends on investors and lenders to keep funding their growth. If that money dries up, even a promising business can run into trouble.
How to spot it. Watch for the Big Five and other AI-invested companies struggling to raise money, borrowing at higher interest rates, receiving credit-rating downgrades, or delaying projects because they can’t get financing. You can also follow reports about the cost of insuring corporate debt against default. If that cost rises sharply, investors may be getting more worried about the company’s ability to repay its debts.
Without context, or without more than three of these signs happening at once, they don’t paint the complete picture.
Watch for a pattern: investment keeps accelerating, demonstrated returns fail to catch up, debt grows, financing becomes more interconnected, and the cost or availability of capital begins to deteriorate.
Unfortunately, warning signs are not always obvious until financial stress is already building. By the time the headlines arrive, some losses may already have occurred.
And even then, a downturn does not automatically mean a devastating crash is imminent.
The AI boom may turn out to be one of the most important technological transitions of our time. It may also leave behind failed companies, underused infrastructure, and investors who committed far more money than the eventual returns could justify.
Both can become a reality.
And that brings us back to a question that applies far beyond AI: how do we make decisions when the future is uncertain and the cost of being wrong is high?
The answer is the same, as it has been for years. Validate your assumptions, and keep validating them as conditions change. Make decisions using the best available evidence, test what happens when your assumptions are wrong, and don’t commit money you can justify only under the most optimistic scenario.
And the bigger your decision or investment is, the more important it is to understand the downside before you let hopes and dreams set you on the path of loss.
