The AI boom is going to go bust, reports The Wall Street Journal . But this leaves open questions such as what will cause it, how likely each risk is — and, if it happens, when and which stocks will be hit and by how much.

My guess is that the combination of higher interest rates and an abrupt change in capital-provider attitude — from fear of missing out to fear of not getting paid back — will be the reasons the AI bubble bursts.

Should that happen, there will be winners and losers. The losers will be indebted neoclouds and AI data center builders including CoreWeave, Oracle and SoftBank. Nvidia and memory makers might also suffer — however, those with little debt will not be hurt as badly. Finally, power equipment suppliers such as GE Vernova and cybersecurity firms could benefit from an AI bust.

Read on for a list of risks that could cause the AI bubble to burst and which stocks could be affected and by how much.

When Capital Suddenly Stops Flowing

As I remember from the dot-com bubble bursting , the market’s decision to stop funding money-losing companies can happen quickly and ruthlessly. When companies have nowhere to go to pay their bills, bankruptcy court is often their next stop.

Since chipmakers and investors fund the AI labs that pay the AI data center builders that borrow to buy the chips, one lab’s funding gap could cascade through the supply chain.

The warning lights are flickering. S&P cut Oracle to BBB- , one notch above junk, and CoreWeave had to raise its spread and yield to close a $2.6 billion loan. Still, deals are getting done: CoreWeave’s loan closed, and top labs continue to raise enormous rounds.

Polymarket put the chances of “AI bubble burst” at 10% by this year-end and 24% by mid-2027. If that happens, I estimate CoreWeave stock could fall 40% to 60%; Oracle could lose 25% to 35% of its value; and SoftBank could decline by 25% to 40%.

When Money Gets More Expensive

AI infrastructure is sensitive to higher interest rates because of the long-lived nature of its assets, the heavy borrowing needed to build and operate, and the long wait for profits. On Sept. 16 the Federal Reserve Bank raised rates to 3.75% to 4% , its first hike since 2023, and most officials expect more.

CoreWeave estimated each one-point rise adds about $30 million a year in interest on its floating-rate debt. This is not a current problem for hyperscalers with positive cash flow and investment-grade bonds.

However, if rates keep rising, that could change. There is about a 20% chance borrowing costs could become truly prohibitive, with the 10-year yield sustained above 5.5%, by mid-2027. If that happens, shares of leveraged cloud-services providers could lose ground — for example, CoreWeave (down 20% to 35%), Oracle (declines 10% to 20%) and heavily indebted Vistra (off 10% to 15%).

When The Grid Can’t Keep Up

A data center earns nothing until it is switched on, but the debt that built it needs to be paid back more urgently. The International Energy Agency projects global data-center electricity use will more than double to roughly 945 terawatt-hours by 2030, and finds that data centers account for about 50% of U.S. demand growth.

The strain is already visible: Oracle’s shares fell after it sent a force-majeure notice on its 2.45-gigawatt Project Jupiter campus in New Mexico after a key energy permit was denied and a gas pipeline slipped.

There is another side to this debate. Shortages delay revenue, but the IEA expects gas and renewables to step in to fill supply gaps. Thus, scarcity will increase the value of power as long as demand is high.

While there is a 70% chance scarce supply slows down the buildout, odds are about 10% that power alone ends the boom. Two stocks — GE Vernova (up 10% to 20%) and Vistra (up the same amount) — could benefit from scarcity pricing. Meanwhile, CoreWeave would likely lose 20% to 30% of its value due to delayed capacity carried on floating-rate debt.

When Paying Customers Don’t Materialize

Unless the AI model builders operate profitably or raise more capital, they will run out of funds to buy more AI computing from hyperscalers. If hyperscaler revenue lags spending, their free cash flow turns negative, hyperscalers cut guidance and chip orders drop.

This has not happened yet. The four biggest hyperscalers plan roughly $700 billion of 2026 capital spending , yet McKinsey’s 2026 State of AI survey finds only 37% of companies using AI report any earnings impact from AI — the same as last year.

In April, OpenAI missed internal revenue and user targets — knocking down Oracle and chip stocks. But Anthropic’s annualized revenue topped $65 billion , roughly seven times its level at the end of 2025, proof that real demand exists.

There is a 25% to 30% chance that a major spender cuts capex by 20% or more by mid-2027 — a scenario suggested by Rothschild, according to Reuters . This could take Nvidia down 25% to 35%, Oracle could lose 30% to 40%, given its heavy reliance on OpenAI contracts, and CoreWeave could tumble between 40% and 60%.

Two other scenarios — cheap open-weight models collapse token prices and wipe out frontier labs’ pricing, and rogue AI ends the world — seem too remote to try to quantify.

Signals Investors Should Watch

Here are some indicators investors should monitor closely:

  • Are Oracle’s credit-default swap rates rising?
  • Is CoreWeave able to obtain new financing at reasonable rates?
  • Will Anthropic price its IPO in November?
  • Will hyperscalers give a robust outlook for 2027 when they report results in late October?

If the answer to these questions turns out to be “no,” power-equipment makers and cybersecurity firms are the rare names that could gain from bad news — although their valuations are high.