Nasdaq Is Soaring, But Burry Sees an AI Bubble - Why?

The Nasdaq Composite closed at a record 27,122.09 on September 21, 2026 after rising 2.26% in a single session. It was up 16.7% for the year. At almost the same time Michael Burry, the investor known for anticipating the US housing crash, was again describing the AI boom as a bubble. The market is effectively saying that AI profits will grow fast enough to justify today’s prices. Burry is asking whether the spending, accounting and financing underneath those profits are as strong as they appear.

Let's break down what Burry actually said, why the Nasdaq is rising, where his AI bubble argument is strongest and what the current numbers say about the probability of a genuine market bubble.

Did Michael Burry really tell investors to “sell” in September 2026?

No fresh one-word “Sell” message from Burry could be verified on September 22, 2026. His famous one-word post was published on January 31, 2023 before a Federal Reserve decision. On March 30, 2023 he publicly wrote that he had been wrong to say it. Using that old quote as though it were a new market call would therefore be misleading.

His current warning is still clearly bearish but it is more specific. In September 2026 Burry argued that an AI bubble was creating an extreme transfer of wealth and criticised what he called circular financing among technology companies. He has also questioned whether major cloud companies are spreading the cost of AI hardware over useful lives that are too long. That is not the same as predicting that the Nasdaq must fall immediately.

There is another important limitation. Scion Asset Management deregistered as an investment adviser in 2025. Its comprehensive quarterly portfolio is no longer visible through routine Form 13F filings. Burry now discusses selected positions and ideas through Substack and social media but those disclosures may not show his complete portfolio, position size or hedges.

The last regular Scion filing for the quarter ended September 30, 2025 reported put options covering 5 million Palantir shares and 1 million Nvidia shares. The filing displayed underlying share values of about $912.1 million and $186.6 million respectively. Those numbers were not the premiums paid for the options and should not be described as $1.1 billion of cash invested in the trade.

That distinction matters because Burry’s warning should be analysed as an argument, not followed as a signal. His reputation makes the message interesting. The evidence determines whether it is useful.

Why did the Nasdaq reach a record high?

The latest rally had three immediate drivers. Oil prices fell sharply, the 10-year US Treasury yield eased to 4.95% and enthusiasm around AI returned. Lower oil reduced the market’s inflation anxiety while lower bond yields made future technology earnings slightly more valuable in today’s money.

AI-linked stocks then amplified the move. AMD rose 9.95% to a record close of $615.52 and crossed $1 trillion in market value for the first time. Intel gained 12.1% and Meta rose 11.4%. The Nasdaq Composite advanced 599.55 points to its first record close since June 2.

Sources: Nasdaq official index history, Associated Press market data and The Wall Street Journal. Closing values and year-to-date returns are through September 21, 2026.

The S&P 500 was still about 0.4% below its record. This tells us that the latest session was not simply a broad economic rally. Large technology and semiconductor companies contributed disproportionately.

The Nasdaq Composite and Nasdaq-100 must also not be treated as the same index. The Composite contains thousands of Nasdaq-listed companies while the Nasdaq-100 contains 100 of the largest non-financial companies listed on Nasdaq. Investors using Nasdaq-100 ETFs are usually more concentrated in mega-cap growth companies than the word “index” might suggest.

Michael Burry’s AI bubble thesis has four parts

Burry’s concern is not that artificial intelligence is useless. His case is that a real technology can still attract too much capital, be financed badly and produce disappointing returns for investors who paid too much. Railways, the internet and fibre-optic networks all transformed the economy even though many companies financing their early build-outs destroyed capital.

AI spending may be running ahead of monetisation

Amazon, Alphabet, Meta and Microsoft now expect combined 2026 capital expenditure of roughly $720 billion to $745 billion. Not every dollar is exclusively for AI. The totals also include data centres, networking, logistics, satellites, offices and other infrastructure. Even with that caveat the scale is extraordinary.

Sources: Amazon Q2 2026 results and earnings call, Alphabet Q2 2026 earnings call, Meta Q2 2026 results and Microsoft FY2026 Q4 earnings call. Company definitions of capital expenditure are not perfectly identical.

A simple hurdle-rate test shows why investors should care. If $720 billion of annual investment eventually needs to earn a 10% operating return it must support about $72 billion of additional annual operating profit. At a 30% operating margin that would require around $240 billion of incremental annual revenue.

This is not a forecast and the spending will create assets that produce revenue for several years. It is a way to frame the burden of proof. AI usage does not only need to grow. It needs to become profitable enough to justify an infrastructure programme approaching three-quarters of a trillion dollars in one year.

Depreciation can delay the appearance of costs

When a company purchases a server it does not normally charge the full cost to profit immediately. It records an asset and recognises depreciation over the asset’s estimated useful life. A longer useful life creates lower annual depreciation and higher near-term accounting profit.

Consider $100 billion of computing equipment. Depreciating it over six years produces about $16.7 billion of annual expense. A three-year life produces about $33.3 billion. The $16.6 billion annual difference does not change the initial cash outflow but it can materially change reported profit.

This is the strongest accounting part of Burry’s argument because AI chips improve quickly. If older processors lose economic value faster than their accounting schedules assume then current profits may look better than the underlying economics.

However the criticism cannot be applied blindly to all data-centre spending. Buildings, power systems and networking equipment can operate far longer than GPUs. Microsoft said roughly two-thirds of its latest quarterly capital expenditure was for shorter-lived assets such as CPUs and GPUs while the remainder was for longer-lived assets. Alphabet said about 60% of its Q2 technical infrastructure investment was in servers and 40% was in data centres and networking equipment. The correct question is whether each asset class has a realistic life, not whether all AI infrastructure should be written off in three years.

Circular financing can make demand look stronger than it is

Circular financing occurs when money moves through a small group of related companies in a way that supports each participant’s reported demand. A chip supplier may invest in an AI developer. The developer may use that funding to purchase cloud capacity. The cloud provider may then use the revenue commitment to justify buying more chips.

Each transaction can be legally separate and economically rational. The risk appears when investors count every leg as independent end-customer demand even though the same pool of capital is moving around the ecosystem. If the final customer is not generating enough cash then the chain eventually needs more outside funding.

This is not proof that AI revenue is artificial. Microsoft said almost 90% of full-year Microsoft Cloud revenue came from customers outside frontier-model companies. It also said all sequential growth in commercial remaining performance obligations during its latest quarter came from customers outside that group. Alphabet reported that nearly 90% of the Fortune 100 were using Gemini Enterprise. Those figures support the argument that adoption extends beyond a small loop of AI labs and their financiers.

High expectations create asymmetric risk

A company can deliver excellent growth and still produce a weak investment return if its valuation already assumes an even better outcome. This is the central valuation risk across US technology stocks.

The Nasdaq-100 was 68.51% technology by Nasdaq’s classification on June 30, 2026. Its 10 largest securities accounted for about 45.0% of the index. Nvidia alone represented 7.60%. This concentration helps when a few AI leaders deliver but it also makes the index more sensitive to earnings disappointments, regulation or a change in the valuation investors are prepared to pay.

Source: Nasdaq-100 official fact sheet dated June 30, 2026. Weights change with prices and index rebalancing.

Concentration is not automatically a flaw. An index should allow successful companies to become large. The danger is assuming that owning 100 companies means the economic risk is evenly spread across 100 independent businesses.

The strongest case against Burry: AI earnings are already real

The current market is not the dot-com bubble copied and pasted into 2026. Many of today’s largest AI companies have profitable core businesses, large customer bases and access to internal cash flow. That gives them more resilience than the loss-making internet companies that depended on repeated equity issuance around 1999.

The latest earnings evidence is also powerful. FactSet estimated on September 18 that S&P 500 information technology earnings would grow 63.3% year on year in Q3 2026. Semiconductor and semiconductor-equipment earnings were expected to rise 126%. Even excluding semiconductors the sector’s expected growth was 24.2%.

Source: FactSet Earnings Insight dated September 18, 2026. Q3 figures are estimates and can change.

The broader S&P 500 forward P/E of 19.1 times was below its five-year average of 19.8 times and only slightly above its 10-year average of 19.0 times. That does not prove every AI stock is reasonably valued. It does show that a claim of universal market mania is too broad.

There is also measurable monetisation:

• Alphabet’s Q2 revenue increased 24% and Google Cloud operating margin reached 35.6% as Cloud revenue rose 82%.
• Microsoft’s fiscal 2026 revenue surpassed $331 billion while Azure exceeded $100 billion of annual revenue and grew 41%.
• Meta’s Q2 revenue rose 28% as ad impressions increased 14% and average price per ad increased 12%.
• Amazon’s AWS grew 37% in Q2 to $42.2 billion of revenue.

These companies are not spending against zero revenue. The real debate is whether incremental AI revenue and cost savings will earn an adequate return on the next $720 billion to $745 billion of investment.

A valuation model shows how both Nasdaq bulls and Burry can be right

Market returns depend on two variables: how fast earnings grow and what multiple investors place on those earnings. A simple three-year model makes the tension clear.

Assume a high-growth technology basket starts at 30 times forward earnings. The figures below are illustrative and are not a forecast for the Nasdaq Composite or any individual stock.

Source: INDmoney Research illustrative calculation. Price change equals earnings growth over three years multiplied by the ending P/E divided by the starting P/E. Dividends are excluded.

The second scenario is especially useful. Earnings can rise by roughly 73% over three years and the investment can still return only about 11% a year if the valuation falls from 30 times to 24 times. Good businesses do not automatically create exceptional returns from every starting price.

This is where Burry may be early without being irrational. The Nasdaq can continue rising while earnings estimates climb. His warning becomes more relevant if capital expenditure keeps accelerating but cloud growth, AI pricing or free cash flow begins to slow.

Higher interest rates make AI valuations less forgiving

The Federal Reserve raised its target rate by 25 basis points to 3.75% to 4.00% on September 16, 2026 because inflation remained elevated. August headline CPI was 3.4% year on year while core CPI was 2.4%. The 10-year Treasury yield was still around 4.95% on September 21 even after easing during the market rally.

Technology valuations are sensitive to rates because more of their expected value comes from profits far in the future. At a 30 times P/E the earnings yield is only 3.3%. Investors may accept that when earnings are compounding rapidly. If growth weakens while a low-risk government bond yields close to 5% the valuation has less room for error.

This is not a direct comparison because equities can grow and bonds cannot. It is a measure of the opportunity cost. AI companies now need to produce enough growth to compensate investors for taking business risk when cash and government bonds offer meaningful yields.

Is the Nasdaq in an AI bubble?

The evidence points to a selective expectations bubble rather than a market-wide earnings illusion.

Burry is right about three risks. Capital spending has reached a scale at which small errors in demand forecasts can destroy large amounts of value. Useful-life assumptions deserve scrutiny because depreciation can shift costs between periods. Concentration means a handful of companies can make a diversified-looking index behave like a narrower AI portfolio.

The blanket bearish case is weaker. Cloud demand is growing at unusually high rates. Backlogs are large. Major platforms are profitable and technology earnings estimates have continued to rise. The Nasdaq rally is therefore supported by real operating progress, not only by a fashionable story.

Our view is that the market’s greatest risk is not that AI disappears. It is that AI succeeds while investors overpay for the companies funding it. The internet transformed the world after 2000 but many shareholders still suffered because the price paid before the crash assumed too much success too soon. Technology can be revolutionary while an investment made at the wrong valuation disappoints.

The most useful conclusion is not “Burry is right” or “the market is right.” The Nasdaq is pricing rapid earnings growth and high returns on AI infrastructure. Burry is highlighting the accounting and financing paths through which those expectations could fail. Investors should monitor the gap between the two.

What should investors track next?

Five indicators can test the thesis more reliably than social-media predictions:

1. Cloud growth against capital expenditure: Revenue growth should remain strong enough to absorb the rapidly expanding asset base.
2. Free cash flow after AI investment: Operating profit can rise while free cash flow weakens if capital expenditure grows faster.
3. Depreciation and useful-life changes: Longer lives can support reported earnings but investors should compare them with actual replacement cycles.
4. Demand outside frontier AI labs: Enterprise contracts, paid users and usage from conventional businesses provide stronger evidence than transactions within the AI funding ecosystem.
5. Earnings breadth: The rally becomes healthier if software, services and non-technology sectors contribute rather than semiconductors carrying most of the growth.

FactSet’s current estimate illustrates the last point. Semiconductor earnings growth of 126% lifts expected technology-sector growth to 63.3%. Remove semiconductors and the estimate falls to 24.2%. That is still strong but it shows how much of the acceleration is concentrated in one industry.

What does this mean for Indian investors?

Indian investors should separate three decisions that are often mixed together: belief in AI, exposure to US equities and the price paid for a specific stock. A person can believe that AI will reshape the economy while deciding that a particular company’s valuation already assumes too much.

The first check is total exposure. Direct ownership of Nvidia, Microsoft or Palantir can overlap with Nasdaq-100 funds, S&P 500 funds, global technology funds and AI ETFs. Counting the same company through every route prevents accidental concentration.

The second check is currency. A US asset’s return in rupees reflects both its dollar price and the USD-INR movement. Rupee depreciation can improve the Indian investor’s translated return while rupee appreciation can reduce it. Currency should therefore be treated as a separate return driver, not proof that the underlying stock performed better.

The third check is time horizon. Burry’s warning is mainly about valuation, capital allocation and the quality of reported earnings. Those risks can take years to appear. Short-term price moves can remain positive even while long-term expected returns become less attractive.

The Nasdaq’s record high is real. So are the profits and demand supporting it. Burry’s warning is also worth taking seriously because the capital required to sustain the AI boom has become too large to ignore. The deciding evidence will not be the next dramatic post or the next record close. It will be whether AI revenue, cash flow and productivity grow fast enough to earn an acceptable return on the infrastructure now being built.