AI bubble: will it burst? What it means for you
Will the AI bubble burst? See what a bubble is, the numbers that worry the market, the warning signs and what changes for you if it bursts.
NinaKalkazu.org
Listen to this article
Narrated by Thalita
One headline says the AI bubble will burst. The next week's says artificial intelligence is just beginning and will change everything. It sounds contradictory, but both can be right, and this article shows why, with what changes for those who just use ChatGPT, Gemini or Claude day to day.
Quick summary: the answer in 30 seconds. Nobody knows if the AI bubble will burst, or when. Tech giants spend far more on data centers than AI brings in today, and increasingly with borrowed money. If the bet fails, stocks fall and companies go under. But the technology usually survives the burst.
Index
- What is a bubble, anyway?
- Why do so many people think the AI bubble exists?
- And who says it's not a bubble?
- Is it one bubble or several?
- What signs show the AI bubble is close to bursting?
- What happens if the bubble bursts?
- What changes for you
- Frequently asked questions
What is a bubble, anyway?
A bubble happens when the price of something rises far above what it will return, pushed by the expectation that it will rise further. While everyone believes, the price climbs. When doubt appears, many people try to sell at once, and the price plummets.
Think of jabuticaba seedlings that everyone buys to resell at a higher price, not for the fruit. The day someone asks how much the tree produces, the line of sellers appears.
The most famous case is the dot-com bubble, of internet companies. According to a Goldman Sachs study on the 25 years since the burst, the Nasdaq, the American tech company exchange, lost 34% one month after the March 2000 peak. At rock bottom, in October 2002, the drop reached nearly 80%.
AI entered this conversation because the money invested in it grows much faster than the money it returns.
Why do so many people think the AI bubble exists?
Because the numbers don't add up yet. Cloud giants spend hundreds of billions of dollars a year on data centers, and AI revenue grows much more slowly. Moreover, part of the spending now is paid with debt, and the money circulates among the same companies.
A data center is a warehouse full of computers running day and night. It works like a power plant: instead of energy, it manufactures AI answers. Cloud companies, like Amazon, Microsoft and Google, own these warehouses and rent their capacity over the internet.
Spending that doesn't pay for itself yet
A study by economists Jessica and Jonathan Wachter calculates that the five American companies that invest the most spent $404 billion on AI infrastructure in 2025, already adjusted for inflation since 2022. For 2026, the forecast is nearly double. The authors warn they risk going under if profit doesn't keep pace.
To get an idea of the scale: Meta announced in July 2026 the expansion of a single data center in Louisiana to 5 gigawatts, with over $50 billion in investment. That's just over a third of the capacity of Itaipu, which has 14,000 megawatts.
On the other side is revenue. Consulting firm Bain estimated in 2025 that the sector will need to bring in $2 trillion a year by 2030 to pay for the computing that demand requires. Even adding what companies save with their own AI, $800 billion would still be missing.
Debt entered the game
Giants used to pay for data centers with leftover cash. That changed. The term that shows the shift is free cash flow: what's left after paying all bills and investments, like the salary that remains after rent, groceries and installments.
In the second quarter of 2026, Alphabet, owner of Google, brought in $119.8 billion. After day-to-day costs and $44.9 billion in equipment, free cash flow was negative at $5.9 billion. Negative forces using reserves or borrowing.
A common way to borrow is issuing debt securities: the company receives money from investors and promises to return it with interest. In October 2025, Goldman Sachs already counted $139 billion in securities issued that year by AI-related companies.
Money going in circles
There are also circular deals. In 2025, Nvidia announced the intention to invest up to $100 billion in OpenAI, gradually, as OpenAI installed 10 gigawatts in Nvidia systems. Part of the money leaves the chip maker and returns to it as chip sales.
The International Monetary Fund (IMF) analyzed 14 such deals since 2023: about $300 billion in equity purchases, $650 billion in debt and $1.4 trillion in future revenue promises. The warning, from October 2026, is that one company's stumble can bring down the others in a chain.
In one sentence: spending is gigantic, revenue is still small compared to it and debt is rising.
And who says it's not a bubble?
Analysts like Eric Sheridan, from Goldman Sachs, and part of the market. The argument is that, unlike 2000, the largest AI companies are profitable, demand for the most advanced models is real and the technology does something unprecedented: intellectual work. The price may be exaggerated, but it wouldn't be a house of cards.
Model, here, is the program that generates AI answers, like GPT behind ChatGPT. In the Goldman Sachs report, Sheridan notes that giants still generate plenty of cash, enough to buy back their own shares and distribute profit to shareholders, something rare in the dot-com era. His conclusion was cautious: perhaps AI just hasn't become a bubble yet.
| Point | Who sees bubble | Who doesn't |
|---|---|---|
| Spending × revenue | Spending runs far ahead | Infrastructure always comes before revenue |
| Profit | OpenAI and Anthropic aren't profitable yet | The giants that spend most are profitable |
| Debt | Growing fast and spreading | Still fits in giants' cash flow |
| Technology | Productivity gains don't show up | AI does intellectual work |
Both sides look at productivity. A survey by Stanford's economics institute (SIEPR), from February 2026, heard from nearly 6,000 executives in the US, UK, Germany and Australia. Over 80% of companies saw no AI effect on employment or productivity. Their expectation is modest: 1.4% productivity gain.
Is it one bubble or several?
Probably several. Those who study the sector, like a VentureBeat article, separate AI into three layers: applications that resell others' AI, labs that create models and the infrastructure of chips and data centers. In a burst, they wouldn't fall together or the same way.
| Layer | Examples | Risk | Why |
|---|---|---|---|
| "Wrapper" apps | Apps that put a screen over another company's AI | High | Google or Microsoft can offer the same function for free |
| Model labs | OpenAI, Anthropic, Google DeepMind, Mistral | Medium | Need lots of money and models become similar |
| Infrastructure | Nvidia, data centers, cloud | Lower long-term | Chips and buildings serve any AI use that succeeds |
What signs show the AI bubble is close to bursting?
There's no single alarm, but IMF and Goldman Sachs warnings point to signs anyone can follow: debt growing faster than profit, companies without proven returns, idle chips, rising interest rates and problems in a central piece of circular deals. Several together increase risk.
| Sign | What to watch | Where to see |
|---|---|---|
| Debt | New securities issues from giants | Economic news and IMF reports |
| Cash | Negative free cash flow | Companies' quarterly reports |
| Returns | Companies cutting AI budgets | Executive surveys, like Stanford's |
| Idle chips | AI chips sitting unused at companies | Market research, like VentureBeat's |
| Interest rates | US interest rate increases | US Federal Reserve decisions |
| Central piece | Problem at OpenAI, Nvidia or partners | Reports and announcements from these companies |
High interest rates weigh twice: they make debt more expensive and bring down stocks of companies that give little profit today and promise much in the future, like most AI companies.
Two signs are already flashing. In VentureBeat surveys from July to September 2026, 79% of 218 companies said their AI chips, the GPUs, sit idle half the time or more. In September, 91% of those planning more infrastructure had no return calculation reviewed by the finance department.
What happens if the bubble bursts?
The 2000 script is most likely: stocks plummet, indebted companies go under, tech jobs disappear for a while and the technology continues. The internet didn't end with the burst, and AI probably won't either. What's expensive is the path.
Fiber optics is the best example. According to Goldman Sachs, the US went from about 1.6 million kilometers of cable in 1996 to 16 million in 2000, much of it with debt. Telecom operators Global Crossing and WorldCom went bankrupt owing $25 billion and $100 billion, respectively. The cables remained and today carry video and cloud.
With AI, data centers and chips would also remain. What worries the IMF is today's debt, spread across funds that don't always know how exposed they are.
That's why both headlines from the beginning are right: the bubble can burst and the technology continue.
In one sentence: in a burst, money disappears before technology, and those who bet with borrowed money suffer most.
What changes for you
For those who just use AI, the effect would be indirect: some apps close, free plans may become more limited and tech jobs cool off for a while. In Brazil, a market tumble usually affects the dollar, and subscriptions priced in American currency would weigh more on the wallet.
| Your profile | What may change | What to do |
|---|---|---|
| Use AI for free | Lower limits, more ads, apps closing | Don't depend on just one service and save copies of what matters |
| Pay subscription | Price may rise or fall, depending on competition and exchange rates | For small apps, prefer monthly to annual plans |
| Work in tech | Fewer hires at AI startups | Invest in skills that work at any company |
| Invest | Sharp drops in AI companies | Know how much of your portfolio depends on the theme (this is not investment advice) |
Free plans are expensive for providers, and companies already test ways to pay the bill, as we showed explaining who pays for free AI with ads on ChatGPT. Without investor money, pressure increases. Data centers also affect the energy bill, and the artificial intelligence section explains the technology behind it.
Frequently asked questions
Will AI end if the bubble bursts?
No. A burst breaks companies that overdid it, but the technology continues, like the internet after 2000.
When will the AI bubble burst?
Nobody knows, and whoever says they know is guessing. You can follow the signs, but not predict the date.
Can free ChatGPT end?
It's unlikely, because the free plan attracts people to paid plans. In a pinch, it's more likely to become more limited or get more ads.
Should I sell AI company stocks now?
This article doesn't give investment advice. It helps to know how much of your money depends on the theme and, to decide, seek a certified professional.
To follow from here on
More useful than guessing the date is following the signs in the table and remembering the lesson from 2000: stock prices fall much faster than the technology's usefulness.
