Will the AI bubble pop?

SECTION 1

What is an economic bubble, exactly?

Before asking whether AI is a bubble, it helps to agree on what a bubble actually is — because the word gets thrown around loosely in headlines.

An economic bubble is a situation where the price of an asset — a stock, a currency, a house, a company — rises far above what its underlying fundamentals (its profits, its cash flow, its practical usefulness) can reasonably justify, usually because buyers expect that someone else will pay an even higher price later. Prices keep climbing not because the value has genuinely increased, but because expectations of future price increases become self-reinforcing. Eventually, new buyers run out, confidence cracks, and prices fall quickly, often overshooting to the downside just as they overshot on the way up.

Economists generally describe a bubble unfolding in five recognizable stages, first popularized by economist Hyman Minsky and later adapted by researchers studying market cycles. Understanding these stages is useful because they repeat across very different assets — tulip bulbs in 1630s Holland, railway stocks in 1840s Britain, dot-com stocks in the late 1990s, and mortgage-backed securities in the mid-2000s.

Diagram 1 — The five stages of a classic bubble (Minsky model)

Most historical bubbles — tulip mania, railway shares, dot-com stocks, and housing in 2007 — followed a similar shape: a new technology or idea triggers a "displacement," early adopters profit, media coverage draws in the public during "euphoria," insiders quietly sell at the top, and a shrinking pool of new buyers eventually triggers a panic.

Where the AI market currently sits on this curve is exactly what economists, investors, and journalists disagree about in 2026. Some argue the industry is deep into the "euphoria" stage, pointing to valuations that have detached from confirmed revenue. Others argue AI is still in an early "boom" stage backed by genuine productivity gains, comparable to how electricity or the internet took over a decade to fully monetize after their initial hype cycles.

Bubbles are not a new invention of modern finance — they are one of the oldest, most repeated patterns in economic history. In the 1630s, prices for individual tulip bulbs in the Dutch Republic reportedly reached the equivalent of a skilled worker's annual salary before collapsing within weeks, an episode now used as shorthand for pure speculative mania. In the 1840s, British investors poured money into railway companies faster than track could realistically be built, and roughly a third of approved railway mileage was never completed once the mania faded. More recently, the 2007–2008 US housing bubble showed that a bubble does not need a flashy new technology at all — it can form around something as ordinary as residential real estate, once lending standards loosen and buyers assume prices only go up. Each of these episodes had a different trigger, but the same underlying mechanism: rising prices attracted more buyers, and more buyers pushed prices higher, until the chain of new buyers broke.

Two features make a bubble genuinely dangerous rather than just a normal, healthy market rally. The first is leverage — how much of the buying is funded by borrowed money, which turns a price decline into forced selling. The second is interconnection — how tightly the fortunes of many companies, banks, and investors are tied to the same asset, so that one failure cascades into others. Both features are directly relevant to the AI debate, and both are examined in detail later on this page.

SECTION 2

Key definitions you need before going further

This page uses financial and technical terms that mean specific things. Here they are in plain language, so students and non-specialists can follow the rest of the article without getting lost.

AI bubble: A situation in which the market value assigned to AI companies, and the amount of money being invested in AI infrastructure, significantly exceeds what current or reasonably near-term AI revenue and profit can justify.
Capital expenditure (capex): Money a company spends building long-term assets — for AI, this mostly means data centers, servers, chips, and power infrastructure. Capex shows up on the balance sheet, not immediately as an expense on the income statement.
Circular deal / circular financing: An arrangement where a supplier invests in a customer, and that customer uses some of that same capital to buy products or services back from the supplier (or a close partner). This can make demand look larger and more independent than it really is.
Compute: The raw processing power — chips, servers, and data centers — needed to train and run AI models. "Buying compute" means renting or buying access to this processing power.
Forward P/E ratio: A stock's price divided by its expected future earnings per share. A high forward P/E means investors are paying a large premium today for profits they expect years from now — a classic marker of speculative pricing when it climbs quickly.
Foundation model / LLM: A large-scale AI system (a "large language model" is one type) trained on huge amounts of data, which can then be adapted for many different tasks — chatbots, coding assistants, image generation, and more.
GPU (graphics processing unit): A specialized computer chip, originally built for video game graphics, that turned out to be extremely efficient at the mathematical operations AI models need. Nvidia is the dominant maker of AI-grade GPUs.
Hyperscaler: A company that operates massive, global-scale cloud computing infrastructure — commonly used to refer to Microsoft, Amazon, Google, and Oracle in the context of AI data centers.
Inference cost: The ongoing cost of actually running a trained AI model to answer real user requests, as opposed to the one-time cost of training it. High inference costs can quietly erode profit margins even when revenue is growing.
Letter of intent (LOI): A non-binding announcement that two companies plan to do a deal, without a signed, enforceable contract yet. Many headline-grabbing AI "deals" worth tens of billions of dollars are LOIs, not confirmed, booked revenue.
Market capitalization: The total value of a public company's shares — share price multiplied by the number of shares outstanding. It reflects what investors are willing to pay, not necessarily the company's actual worth.
Unicorn: A privately held startup valued at $1 billion or more by its investors, based on the price paid in its most recent funding round — a valuation that isn't tested by public markets and can be far from realizable cash value.
Valuation multiple: A shorthand way to compare how expensive a company is relative to a financial measure, such as price-to-sales or price-to-earnings. Rapidly expanding multiples, without matching growth in the underlying numbers, is a classic bubble signal.

SECTION 3

The dot-com bubble: the closest historical precedent

Almost every article about the AI bubble eventually compares it to the dot-com crash of 2000–2002 — and for good reason. It's the most recent example of a transformative technology attracting speculative capital on a massive scale.

Between 1995 and early 2000, the rise of the commercial internet drew enormous investment into any company that could plausibly attach ".com" to its name. The Nasdaq Composite index rose roughly fivefold in under five years, and forward price-to-earnings ratios for the Nasdaq-100 reportedly reached around 60 times earnings at the peak in March 2000 — an extraordinary premium that assumed years of flawless growth. Famous cautionary tales from that era include Pets.com, which spent heavily on Super Bowl advertising and collapsed within roughly two years of its 1998 founding, and companies like Webvan and eToys, which burned through hundreds of millions of dollars in venture capital before folding. When the bubble burst, the Nasdaq fell by more than 75% from its peak, and it took roughly 15 years for the index to reclaim its 2000 high.

The key structural problem in the dot-com era was that most of the money-losing companies had little revenue and, more importantly, no credible path to profitability — investors were pricing in a story, not a business model. Only a small fraction of dot-com-era companies, by some estimates around 14%, were actually profitable at their peak valuations.

Today's AI leaders look different on that specific metric. Nvidia, for example, reported roughly $215.9 billion in revenue for its most recent fiscal year with a net profit margin near 53%, according to industry analysis — real, audited earnings, not a story about future earnings. That is the strongest argument AI bulls make: this isn't 1999's speculative frenzy over companies with no revenue, it's a boom built partly on companies that are, in fact, making enormous amounts of money today.

Comparison of key metrics between the dot-com bubble (2000) and the AI boom (2026)
Metric	Dot-com peak (March 2000)	AI boom (early–mid 2026)
Forward P/E of leading tech index	~60x (Nasdaq-100)	~23x (S&P 500), still the richest since the dot-com era
Share of market-leading companies with real profits	~14% of dot-com companies profitable	Leading chip and cloud firms report large, audited profits; many AI labs and startups do not
Flagship stock's multi-year gain	Cisco up roughly 30x over the 1990s	Nvidia up roughly 13x since late 2022; up ~880% over three years by some counts
Financing style	Heavy reliance on IPOs and public equity from companies with little revenue	Heavy reliance on private mega-valuations, vendor financing, and circular investment deals
Post-peak outcome	Nasdaq fell over 75%; took ~15 years to recover the prior high	Not yet known — still an open, actively debated question in 2026

Why this comparison matters, and where it breaks down

The dot-com comparison is useful for spotting warning signs — detached valuations, FOMO-driven retail buying, and companies chasing a trend without a business model. But it can also be misleading if applied too literally. The internet's core infrastructure (fiber optic cable, mostly) took years to be used at full capacity after 2000, whereas AI data centers are reportedly running near capacity almost immediately given how compute-hungry current AI models are. That single difference is central to the bull case explored in Section 5.

SECTION 4

The case that AI IS a bubble

Here is the strongest version of the argument that the AI boom has become a speculative bubble, based on evidence reported through 2026.

1. Spending is far outpacing confirmed revenue

Multiple market analyses in 2026 estimate that global AI investment will exceed $2.5 trillion for the year, with roughly half of that flowing into data centers, chips, and other infrastructure. Against that, estimates of actual enterprise AI revenue realized in the same period run closer to $100 billion — leaving what several analysts describe as a gap of several hundred billion dollars between money spent and money earned back. A gap this size does not automatically mean failure, since infrastructure is typically built ahead of demand. But it does mean the industry is making a very large bet that revenue will eventually catch up, and every quarter that gap persists adds financial pressure.

2. "Circular deals" are inflating the appearance of demand

One of the most-discussed concerns in 2026 is the web of circular investments among the largest AI players. In simplified form: Nvidia has committed to invest up to roughly $100 billion into OpenAI; OpenAI, in turn, commits hundreds of billions of dollars to buying compute from Oracle, Microsoft Azure, and Nvidia chips. Separately, Amazon has invested billions into Anthropic, and Anthropic commits a similar order of magnitude back to Amazon's cloud infrastructure; Google has made a comparable investment into Anthropic, which also runs workloads on Google Cloud. Analysts at firms including Wedbush Securities have publicly described this pattern as "circular investing" — money moving in a loop among a small cluster of companies, each transaction simultaneously counted as investment income for one company and infrastructure revenue for another.

Diagram 2 — Simplified map of circular AI investment deals (2025–2026)

A simplified view of the AI industry's most-discussed circular deals: chip and cloud suppliers invest equity into AI labs, which then spend a large share of that capital buying compute back from the same suppliers or their close partners. Figures are drawn from company announcements and financial press reporting through mid-2026 and include both signed contracts and non-binding letters of intent, which analysts caution should not be treated as equivalent.

3. Debt is climbing to fund the build-out

Analysts have flagged that AI-related corporate debt reportedly reached around $1.2 trillion by mid-2026, making it the largest single segment of the investment-grade bond market — larger than debt issued by the US banking sector. Rising reliance on debt to finance speculative-adjacent infrastructure is a pattern that historically raises the stakes of any downturn, because falling asset prices combined with fixed debt obligations can force distressed selling.

4. Institutional warnings are stacking up

This is no longer just a retail investor talking point. Through 2026, the International Monetary Fund and the Bank of England have both issued public warnings that an overheated AI-linked stock market could trigger an "abrupt" correction with spillover effects on the broader economy, given how concentrated recent stock market gains have become in a small number of AI-linked companies.

5. Executives themselves are hedging their language

Even leaders inside the industry have started using more cautious language. OpenAI's CEO has publicly acknowledged that near-term profitability is not the company's immediate priority, with capability and market share taking precedence — a framing that, whatever its long-term merit, means investors are explicitly being asked to fund growth ahead of proven returns, echoing dot-com-era rhetoric about "growth first, profits later."

6. Inference costs may be rising faster than efficiency gains

A less-discussed but structurally important concern: even when an AI lab's revenue grows, the cost of actually running its models for every user query — its inference cost — can grow just as fast or faster, especially as models get larger and more capable. Anthropic's CEO has pointed to inference cost as a primary constraint on margin expansion. This matters because it changes the underlying question investors are pricing. A pure "platform" business, like a social network, gets cheaper to serve each additional user over time. A "utility" business, where every additional user costs real, ongoing computing resources, does not automatically get more profitable just because it grows — and as of 2026, which of these two models best describes leading AI labs has not yet been settled.

7. Power and hardware bottlenecks are real physical constraints

Unlike a website, an AI data center needs enormous, continuous supplies of electricity, and in many regions power grid capacity is already struggling to keep pace with new data center construction. Analysts have flagged local power shortages in several major data-center hubs as a growing risk to the AI build-out timeline. A bubble narrative gains extra weight when the physical world — power plants, transmission lines, advanced chip manufacturing capacity — cannot expand as quickly as the financial capital chasing it, because that mismatch forces companies to either delay projects or bid up scarce resources, both of which strain the economics of the bet.

SECTION 5

The case that AI is NOT (yet) a bubble

Now the strongest version of the opposing argument — because a fair explainer has to give both sides a real hearing.

1. Real, audited profits already exist at scale

Unlike the vast majority of dot-com-era companies, several of the biggest AI-linked companies report substantial, verified profits today, not projected profits years out. Nvidia's roughly 53% net margin, mentioned earlier, is not a forecast — it is reported financial results. Microsoft, Amazon, and Google all generate large profits from businesses well beyond AI, meaning that even a serious AI slowdown would not threaten their survival, only their growth rate.

2. Infrastructure build-outs historically precede full monetization

Bulls point out that major infrastructure waves — railroads in the 19th century, electrical grids in the early 20th century, and fiber optic networks in the 1990s — were all overbuilt relative to demand at the time, then "grew into" that capacity over the following decade. Under this reading, today's enormous AI data center spending isn't irrational, it's simply early: companies are racing to secure the compute capacity a genuinely transformative technology will eventually require, even if usage hasn't fully caught up yet. Unlike unused dot-com-era fiber, AI data centers built today are reportedly running near full capacity almost immediately due to the sheer computational hunger of current models.

3. Enterprise adoption is measurable and growing

Corporate earnings data through 2026 shows continued double-digit revenue and earnings growth across most S&P 500 sectors, with analysts pointing to productivity contributions from AI tools in software development, customer service, and data analysis as a real, if still-early, contributor. This is different from dot-com-era "e-commerce will change everything" narratives that had little measurable proof at the time.

4. Some major banks remain confident

Not every institution is sounding the alarm. Goldman Sachs and JPMorgan have both published research arguing that current AI-related growth is fundamentally supported by real demand and earnings, rather than pure speculation, even as they acknowledge that valuations in some pockets of the market are stretched.

5. A broad market crash is not most analysts' base case

Even AI-bubble skeptics tend to stop short of predicting a dot-com-style collapse across the whole market. The more common view through 2026 is that a correction is more likely to be concentrated in the most overvalued, least-profitable pockets of the AI ecosystem — smaller AI startups, AI-adjacent stocks with weak fundamentals, and companies overly dependent on circular financing — rather than a systemic crash that drags down fundamentally profitable companies like Nvidia or Microsoft.

Bear case, in one line

Spending is outrunning revenue, debt is rising, and a small circle of companies is trading capital back and forth in ways that inflate apparent demand — a pattern that has preceded every major financial bubble in modern history.

Bull case, in one line

Unlike the dot-com era, the companies at the center of this boom already generate enormous real profits, and infrastructure build-outs have historically looked "overbuilt" right before demand caught up to justify them.

Diagram 3 — Illustrative AI infrastructure capex vs. realized AI revenue, 2023–2026

Illustrative visualization built from figures reported across multiple 2026 market analyses (McKinsey, Intellectia AI, and financial press estimates of hyperscaler capex and enterprise AI revenue). Exact figures vary by source and methodology; the consistent pattern across sources is a widening gap, not the precise dollar amounts.

SECTION 6

Real stories from the current AI boom

Numbers only tell part of the story. Here are three real, ongoing situations that illustrate what "is this a bubble?" actually looks like on the ground.

THE INFRASTRUCTURE BET

Stargate: a $500 billion wager on future demand

In January 2025, OpenAI, SoftBank, Oracle, and the Abu Dhabi-backed investment vehicle MGX announced Stargate, a joint venture aimed at building roughly $500 billion of AI data center capacity in the United States, with around $100 billion committed for immediate construction. Oracle became the primary compute provider, and the project was publicly framed as critical national infrastructure. Supporters call it visionary — building the physical backbone a genuinely transformative technology will need. Critics point out that the entire plan depends on assuming that demand for AI compute will keep growing fast enough, for long enough, to fill facilities this large — an assumption that, if wrong, leaves an extraordinary amount of stranded capital in half-built data centers.

THE CIRCULAR DEAL, UP CLOSE

Nvidia, OpenAI, and the "left hand, right hand" criticism

When Nvidia announced plans to invest up to $100 billion into OpenAI, Requisite Capital Management's Bryn Talkington summarized the skeptical view bluntly on CNBC: Nvidia effectively hands OpenAI $100 billion, and OpenAI hands a large share of it straight back to Nvidia by buying its chips. Oracle's arrangement follows a similar shape, reportedly worth around $30 billion a year, where Oracle builds the data centers and OpenAI pays to use them — while Oracle itself is a major buyer of Nvidia hardware. None of this means the underlying technology is worthless, but it does mean that some of the "demand" being celebrated in earnings calls and press releases is, in accounting terms, money that never really left the same small circle of companies.

THE BUSINESS-OWNER REALITY

Enterprise adoption is real, but returns are uneven

Away from the trillion-dollar headline deals, small and mid-sized businesses worldwide have spent 2025 and 2026 adopting AI tools for customer support, content creation, coding assistance, and data analysis. Surveys of enterprise AI adoption through this period consistently show a similar pattern: a meaningful share of companies report measurable productivity gains from narrow, well-scoped AI use cases, while a larger share of ambitious, company-wide "AI transformation" projects fail to deliver a clear return on investment within their first year. This gap — between the value of AI in daily use and the very high price attached to AI-linked companies on the stock market — is itself one of the central pieces of evidence in the bubble debate: the technology can be genuinely useful and the financial valuations can still be overextended at the same time.

THE INVESTOR'S DILEMMA

Michael Burry-style skepticism versus index-fund reality

Prominent short-sellers and value investors who correctly called earlier bubbles have publicly questioned AI-linked valuations and, in some cases, taken bearish positions against specific AI-heavy stocks through 2025 and 2026. At the same time, because AI-linked companies now make up such a large share of major stock indices, the vast majority of ordinary retirees and pension holders around the world are exposed to this exact question whether they intend to be or not, simply by holding a diversified index fund. This is the crux of the investor dilemma explored in Section 7: avoiding AI entirely is not really an option for most passive investors, which makes understanding the mechanics of this bubble debate — rather than simply picking a side — the more useful skill.

SECTION 7

What this means for you, by audience

"Will the AI bubble pop?" is not one question — a correct, useful answer looks different depending on whether you're studying, investing, keeping the books, or running a company.

🎓 For students

The financial question (are AI stocks overvalued?) and the career question (should I learn AI skills?) are separate, and it's easy to confuse them. Even if AI-linked stocks correct sharply, the underlying technology adoption inside companies is unlikely to reverse — much like internet usage kept growing straight through the dot-com crash. The safest position is to build durable, transferable skills: data literacy, the ability to evaluate AI outputs critically, and fluency working alongside AI tools in your actual field, rather than betting your education on any single company's stock or any single AI product staying dominant.

📈 For investors

Distinguish between the technology and the ticker. A company can build genuinely useful AI products while its stock is still priced for perfection — and the reverse is also true. Watch capex-to-revenue ratios, the share of a company's growth that depends on circular or vendor-financed deals, and how much debt is funding expansion. Diversification across sectors, not concentration in AI-linked names, is the standard risk-management response to a market this concentrated in a handful of stocks.

🧮 For accountants & finance professionals

The technical accounting questions matter more than the headlines. How is AI infrastructure spending capitalized versus expensed, and over what depreciation schedule? Are "committed" deal values booked as revenue, or are they non-binding letters of intent that shouldn't appear as recognized income yet? How are circular investment-and-purchase arrangements disclosed, and do they inflate reported revenue growth in ways that could mislead a reader of the financial statements? These distinctions are exactly where bubble-era accounting scandals have historically originated.

🏢 For business owners

Whether or not the stock market bubble pops, your AI adoption decisions should be evaluated on their own merits: does this specific tool save time or money in a specific workflow, measured against its actual cost? Avoid over-committing to long-term, high-cost AI infrastructure contracts based purely on competitive fear of missing out. Treat AI vendor pricing as likely to fall over time as competition increases, and prefer flexible, shorter-term commitments over long lock-ins while the market is this volatile.

SECTION 8

Ten warning signs analysts are watching in 2026

These are the specific, checkable indicators that market analysts, central banks, and researchers point to when discussing AI bubble risk.

Widening capex-to-revenue gap. Infrastructure spending growing significantly faster than confirmed, recognized AI revenue, quarter after quarter.
Rising reliance on circular deals. A growing share of "demand" coming from suppliers investing in their own customers.
Debt-funded expansion. AI-linked corporate debt climbing as a share of total investment-grade credit markets.
Market concentration. A shrinking number of companies accounting for a growing share of total stock market gains.
Letters of intent treated as revenue. Non-binding announcements being reported in the media, and sometimes in investor materials, in ways that blur the line with signed, binding contracts.
Retail investor FOMO. A surge in individual investors buying AI-linked stocks or funds specifically because prices are rising, rather than based on fundamentals.
Executive hedging language. Company leaders publicly de-emphasizing near-term profitability in favor of "capability and scale."
Power and energy constraints. Data center power shortages that could cap how much AI infrastructure can actually be deployed and used, regardless of capital available.
Falling AI product prices amid rising infrastructure costs. Intense competition pushing AI subscription and API prices down even as the cost of building the underlying infrastructure keeps climbing — a margin squeeze.
Institutional warnings. Central banks and international financial institutions publicly flagging AI-linked valuation risk, as the IMF and Bank of England both did through 2026.

SECTION 9

What actually happens if the AI bubble pops?

It helps to separate three different possible outcomes, because "the bubble pops" gets used to describe very different scenarios.

Scenario A: A contained correction

The most commonly cited scenario among 2026 market analysts: overvalued, weakly profitable AI startups and AI-adjacent stocks fall sharply, some fail outright, and valuations reset closer to fundamentals — while profitable core companies like Nvidia, Microsoft, and Google see their stock prices fall but their underlying businesses survive largely intact. This resembles what happened to strong dot-com-era survivors like Amazon, which fell more than 90% from its 2000 peak but recovered and went on to become one of the world's most valuable companies.

Scenario B: A broader market slowdown

Because AI-linked companies now represent such a large share of total stock market value, especially in the US, a serious AI correction could drag down broader indices, retirement accounts, and consumer confidence even for people who never directly invested in AI stocks — echoing warnings from the IMF and Bank of England about spillover risk into the wider economy.

Scenario C: A "soft landing"

The optimistic case: AI revenue and enterprise adoption gradually catch up to current spending levels over the next several years, valuations grow into their current prices rather than collapsing, and today's aggressive infrastructure buildout is remembered as prescient rather than reckless — similar to how the fiber optic overbuild of the late 1990s eventually became essential capacity for the streaming and cloud computing era of the 2010s.

What history suggests either way

One consistent lesson from the dot-com crash: the technology itself did not disappear. Internet usage kept climbing straight through the crash and the recession that followed it. If AI follows a similar path, a financial correction in AI-linked stocks would not necessarily mean AI tools stop being used or improved — it would mean weaker companies fail, capital becomes more disciplined, and the strongest AI businesses eventually emerge in a healthier position.

SECTION 9B

A global picture: the AI bubble question isn't only a US story

Most headline numbers in this debate come from US markets, because that is where the largest AI labs and chipmakers are listed. But the risk, and the opportunity, extend well beyond the United States.

United States

The US remains the center of the AI investment boom, home to Nvidia, Microsoft, Google, Amazon, OpenAI, and Anthropic. It is also where stock market concentration risk is highest: a small group of AI-linked companies now represents a historically large share of total US stock market value, which is precisely why the IMF and other institutions have focused their warnings on spillover risk to the broader American economy, including retirement savings held in index funds that are more exposed to these companies than many investors realize.

Europe

European markets have comparatively little direct exposure to AI-linked "Big Tech" style valuations, since the continent has few companies of similar scale in the sector. That gives European investors some insulation from a US-centered correction, but it also means European businesses are almost entirely dependent on American and Asian AI infrastructure providers, raising separate strategic and regulatory questions that European Union policymakers have been actively debating around data sovereignty and AI regulation.

Asia

Asia plays a dual role in this story. Taiwan, through TSMC, manufactures the advanced chips that make the entire AI boom physically possible, making it arguably the single most critical link in the global AI supply chain — and a geopolitical risk point in its own right. China, meanwhile, has pursued a somewhat separate AI ecosystem, partly due to US export restrictions on advanced chips, and Chinese AI labs have shown they can build competitive models with less compute than some Western labs assumed was necessary, a development that itself puts pressure on the "spend enormous capital to win" thesis underlying much of the Western AI buildout. Gulf state sovereign wealth funds, notably in the UAE and Saudi Arabia, have also become major financial backers of Western AI infrastructure projects such as Stargate.

Emerging markets

For students and business owners in emerging markets, the AI bubble question intersects with a separate, practical issue: access. Many of the most capable AI tools are priced and marketed primarily for wealthier markets, and the infrastructure gap — reliable power, broadband, and cloud access — remains a real constraint on how quickly businesses in lower-income regions can adopt AI tools regardless of how the financial bubble debate resolves. This makes the "technology versus valuation" distinction covered earlier in this guide especially relevant: the pace of useful AI adoption in emerging markets is likely to be shaped more by local infrastructure and cost than by the ups and downs of Nasdaq-listed AI stocks.

SECTION 10

What experts and institutions are saying

A snapshot of positions from across the spectrum, as reported through 2026.

Summary of expert and institutional positions on AI bubble risk
Voice	General position
International Monetary Fund	Has warned that an overheated, AI-driven stock market could produce an abrupt correction with broader economic spillover.
Bank of England	Flagged a growing risk that the AI-linked segment of financial markets could see a sharp downward correction.
Goldman Sachs	Has generally argued that AI-driven earnings growth is supported by real fundamentals, while acknowledging pockets of stretched valuation.
JPMorgan	Broadly supportive of the growth thesis, while separately noting the rapid rise of AI-linked corporate debt as a market segment to watch.
Alpine Macro (Oxford Economics)	Argued the market does not yet show classic signs of speculative excess, but flagged 2026–2027, if financial conditions loosen further, as a period of elevated risk.
Economist Ruchir Sharma	Has warned that a correction becomes more likely if interest rates rise and cheap capital becomes scarcer.
Wedbush Securities (Matthew Bryson)	Has described Nvidia's investment pattern as falling "squarely into" circular investment territory, while noting it could also build a genuine competitive advantage if the bet pays off.

The honest summary: there is no consensus. Reasonable, well-informed people looking at the same data are reaching different conclusions, which is itself a sign that the market is at a genuinely uncertain point rather than an obviously overheated one.

SECTION 11

Test yourself: the AI bubble quiz

Ten questions to check what you've learned. Your score is calculated instantly and never leaves your browser.

Question 1 of 10

YOUR RESULT

0 / 10

SECTION 12

Frequently asked questions

Is the AI industry definitely in a bubble?

No one can say this with certainty. Spending on AI infrastructure is growing far faster than confirmed revenue, which matches the early pattern of past bubbles. But unlike many dot-com companies, today's largest AI firms, especially chipmakers, report substantial real profits. Most economists describe the situation as elevated risk rather than a confirmed bubble.

What is a circular AI deal?

A circular deal is when a supplier invests money into a customer, and that customer then spends the same or similar amount buying products back from the supplier. In AI, this shows up when a chipmaker invests in an AI lab, and the lab uses that capital to buy chips or cloud computing from the same supplier or its close partners.

How is the AI bubble different from the dot-com bubble?

The largest AI-linked companies today, particularly chip and cloud providers, are already generating tens of billions of dollars in real profit, unlike most dot-com era companies which had little or no revenue. However, valuations for many private AI labs and AI-adjacent startups have risen just as fast as internet stocks did in the late 1990s, and overall market concentration in a handful of tech stocks is historically high.

What happens to the economy if the AI bubble pops?

A sharp correction could wipe out significant stock market value, slow venture funding, and trigger layoffs at AI startups and their suppliers, similar to the dot-com crash. Because AI spending is now a meaningful share of overall economic growth in some countries, a severe correction could also slow broader GDP growth. Most analysts see a partial correction in overvalued segments as more likely than a full systemic crash.

Should students worry about an AI bubble when choosing a career?

Career risk and stock market risk are different things. Even if AI stock valuations correct, practical AI skills, data literacy, and the ability to work alongside AI tools are likely to remain valuable across industries, because the underlying technology adoption is separate from short-term financial valuations.

Are Nvidia and Microsoft in the AI bubble?

Nvidia and Microsoft both generate large, verified profits and have diversified revenue beyond AI, which sets them apart from purely speculative companies. That said, a meaningful share of their recent growth and valuation is tied to AI demand, so a slowdown in AI spending would affect their results even if it would not threaten their survival.

How can investors protect themselves from an AI bubble bursting?

Common approaches include diversifying across sectors rather than concentrating in AI-linked stocks, favoring companies with confirmed profits and cash flow over pure growth narratives, avoiding excessive leverage or margin exposure to volatile tech stocks, and treating any single forecast about a market top or bottom with caution. This is general education, not personalized financial advice.

How should accountants and finance teams think about AI-related risk?

Accountants should pay close attention to how companies capitalize AI infrastructure spending, how vendor financing and circular investment arrangements are disclosed, and whether reported AI revenue reflects recognized cash flow or non-binding commitments and letters of intent, since these distinctions materially affect the accuracy of financial statements.

What signs typically appear before a bubble bursts?

Historically, warning signs include valuations detached from earnings, heavy reliance on debt or vendor financing to fund growth, widespread retail investor participation driven by fear of missing out, and a rising number of companies rebranding around a hot trend without a matching business model.

Will AI itself disappear if the bubble pops?

No. Even after the dot-com crash, the internet did not disappear — it kept growing while weak companies failed and capital became more disciplined. Most analysts expect a similar pattern for AI: continued technology adoption alongside a possible financial correction in overvalued stocks and startups.

What is the single best indicator to track going forward?

Most analysts point to the gap between AI infrastructure capital spending and confirmed, recognized AI revenue as the metric to watch most closely. A narrowing gap over the next several quarters would support the "this is a justified build-out" view; a continuously widening gap, especially alongside rising debt levels, would support the "this is a bubble" view.

SECTION 13

Conclusion: so, will the AI bubble pop?

The honest answer is that this remains genuinely unresolved, and anyone who claims total certainty in either direction is overstating their evidence.

What the evidence through 2026 does support is a more nuanced position: parts of the AI ecosystem — particularly unprofitable startups, companies overly dependent on circular financing, and stocks priced for flawless multi-year execution — show real bubble characteristics. Other parts — particularly the profitable chip and cloud infrastructure companies at the center of the buildout — rest on genuine, audited earnings that most dot-com-era companies never had. Both things can be true at once, and probably are.

For students, investors, accountants, and business owners alike, the most useful response isn't trying to perfectly time a market top or bottom. It's understanding the specific mechanics covered on this page — circular deals, capex-to-revenue gaps, valuation multiples, and the difference between a letter of intent and a signed contract — well enough to make informed decisions regardless of which way the broader market eventually moves.

Educational content, not financial advice. This page is provided by Learn Edition for general educational purposes. It does not constitute investment, accounting, tax, or legal advice, and figures cited are drawn from public reporting and third-party analysis current as of September 2026, which may change. Always consult a qualified, licensed professional before making financial, investment, or business decisions.