Photo: Christian David / Wikimedia Commons, CC BY-SA 4.0, cropped

Will the AI Bubble Burst in the US Stock Market? 6 Warning Signs, 4 Reasons It Hasn't, and What History Says

The short answer

No one knows if or when the AI bubble will burst in the US. Anyone who tells you the date is guessing.

What we can say is where things stand on October 28, 2026, and it's a striking split:

Asia has already shown what different versions of "the AI bubble bursting" look like. The US hasn't had any of them yet. Read our full breakdown: Did the AI bubble burst in Asia?

Below are the six things that worry us, the four reasons the boom keeps going, and what history suggests.

6 warning signs

1. Valuations are near dot-com levels

The Shiller CAPE ratio, which compares stock prices with 10 years of inflation-adjusted earnings, was about 40.7 for the S&P 500 at the October 26 close. The all-time high was 44.2 in December 1999, just before the dot-com crash. Its long-run average is under 20.

A high CAPE doesn't tell you when a fall comes. Stocks can stay expensive for years. But it does tell you future returns from here are likely to be lower than the past decade's, and that there's less cushion if earnings disappoint.

2. AI spending is bigger than the railroads

The Wall Street Journal estimated in September that AI investment now equals about 3.6% of US GDP. For comparison: the 1800s railroad boom peaked at about 2.2%, the 1950s interstate highways at 1.1%, and the 1996–2003 internet build-out at 1.1%.

Individual budgets are staggering. Alphabet guided $195–$205 billion of capital spending for 2026, and its heavy spending pushed free cash flow negative last quarter. Microsoft is expected to spend more than $50 billion in a single quarter. Both report results tonight. See who actually gets paid from AI capex.

3. More of it is paid for with debt

In earlier years, Big Tech paid for AI out of its own cash. In 2026 that changed. Hyperscalers began selling long-dated bonds, including a 100-year bond from Alphabet, and off-balance-sheet commitments such as data-centre leases have been estimated at around $3 trillion. Our explainer on hyperscalers' $1.2 trillion in lease obligations goes deeper.

Debt changes the game. Equity investors can wait for profits. Lenders want to be paid on schedule.

4. Credit cracks are appearing

The bond market is often the first place stress shows up, and it's showing up:

• CoreWeave, an AI cloud provider, saw its 2032 bonds fall below 78 cents on the dollar in October, pushing yields to about 15%. See our coverage.
• Oracle was downgraded by S&P Global in July to one notch above junk, and the cost of insuring its debt (its five-year credit default swap) hit a record 261 basis points on October 8.
• A Kansas City Fed president said in September it was important to ask whether the AI ecosystem was becoming "too big to fail."

5. Circular money

Some of the AI boom's revenue comes from companies funding their own customers. Nvidia's planned investment in OpenAI, which then buys computing power that runs on Nvidia chips, is the best-known example. When money goes round in a circle, it can make demand look stronger than it is. The Journal described it as "virtuous on the upside, and vicious on the downside." We explain it in Nvidia's OpenAI financing and the round-tripping question.

6. The Fed is raising rates

Bubbles rarely end on their own. They usually end when money gets more expensive. The Fed raised rates in September for the first time in three years, to 3.75%–4.00%. US inflation was 3.7% in September, and markets price a strong chance of another hike in December. Higher rates hurt long-duration growth stocks the most, and make debt-funded data centres more expensive to finance.

4 reasons the boom hasn't burst

1. The demand is real, so far

This is the biggest difference from 2000. Today SK Hynix reported quarterly revenue up 328% and operating profit of about $62.5 billion, and said AI memory demand will outpace supply through 2030. TSMC guided fourth-quarter revenue to $52–$54 billion. Microsoft's contracted-but-unbilled backlog was $678 billion last quarter, up 84%, and Google Cloud grew 82%.

Dot-com companies sold dreams. Today's AI suppliers are selling physical chips that are sold out.

2. The biggest spenders are profitable

Microsoft, Alphabet, Amazon and Meta are among the most profitable companies in history. If AI returns disappoint, they can cut budgets and survive, painfully. That's not true of the speculative Chinese AI listings that just crashed, or of the many dot-com firms that had no profits at all.

3. US stocks are nowhere near China's valuations

China's STAR50 peaked near 250 times earnings. The Nasdaq-100 trades around 30 times. Expensive, yes, but in a different universe.

4. Markets have already absorbed several scares

The Nasdaq had its worst day since July on October 14, falling 1.88%. AI stocks swung hard on debt worries. Yet the S&P 500 recovered to near its record within two weeks. Each scare that doesn't turn into a crash suggests buyers still believe the story, though that belief can change quickly.

What history says about how bubbles end

Two lessons stand out. First, bubbles that burst while central banks are tightening tend to fall the hardest, and the Fed is tightening now. Second, the companies at the centre of a bubble can be both a real revolution and a terrible investment at the wrong price. The internet changed the world. Investors who bought the Nasdaq in March 2000 still waited 15 years to break even.

Three ways this could play out

1. Slow deflation. AI profits keep growing while stock prices go sideways for a few years, letting earnings catch up with valuations. Painless, but boring.

2. A sharp correction. A trigger, like a disappointing earnings season, a big capex cut or a credit event, sends AI stocks down 30%–50% and the S&P 500 down 20% or so. Strong companies recover; weak ones don't. This is roughly what Korea just went through.

3. A full burst. Multiple debt-funded AI firms fail, banks and private-credit funds take losses, and the selling spreads well beyond tech. This would look more like 2000–2002. It requires several things to go wrong at once.

We won't put odds on these. Anyone who does is guessing.

What a sensible investor can do

You don't need to predict the top to protect yourself:

• Know how much AI you own. If you hold an S&P 500 or Nasdaq fund, you already own a lot. Our 15-minute AI exposure check shows you how to measure it.
• Rebalance, don't panic. If AI stocks have grown into a bigger slice of your portfolio than you planned, trimming back to your target mix locks in some gains without betting everything on a crash.
• Diversify beyond the top 10. Equal-weight, international, value and bond funds don't depend on the same handful of companies. See S&P 500 concentration risk.
• Keep your emergency fund in cash. Money you need in the next few years shouldn't be in volatile stocks.
• Avoid leverage. Leveraged ETFs and margin magnify drops like Korea's 39% into wipeouts.
• Stay invested for the long term. Selling everything to dodge a crash means you also have to guess when to get back in. Most people get both calls wrong.

The bottom line

The AI bubble has burst in China and cracked in Korea. In the US, it has only wobbled. The warning signs, record-level valuations, debt-funded spending, credit stress and a tightening Fed, are the same ones that preceded past busts. But the profits and demand behind US AI are far more real than China's mania or the dot-com era.

The smartest response isn't to predict the pop. It's to make sure your portfolio can survive one.

Frequently asked questions