---
title: The Cracks in the AI Bubble Are Already in the Stock Market
description: AI chip, memory and power stocks have lost between a fifth and 62% since their 2026 peaks while orders keep rising. Where the AI bubble is cracking.
author: Darie Nani (Editor-in-Chief)
date: 2026-11-16T20:02:47.896Z
updated: 2026-11-16T21:18:22.825Z
canonical: https://www.sovereignmagazine.com/article/ai-bubble-cracks-stock-market
image: https://cdn.nanimediahouse.com/ai-bubble-red-tickers-v2.webp
categories: Markets, Artificial Intelligence, Economy
content_type: Analysis
region: United States
publication: Sovereign Magazine
schema_type: Article
---

The companies that make the chips, memory, power equipment and fiber for artificial intelligence have lost between a fifth and 62% of their stock market value since their 2026 peaks. Intel is down 61% from its June high, Corning 59% and Caterpillar 46%, and the Philadelphia semiconductor index has lost a third of its value since late June. Nvidia, the company most closely tied to AI, is down 35% from its October high.

The Bank of England warned in October 2025 that technology stocks focused on AI looked stretched. The first wave of selling came in July, and by the time the International Monetary Fund named an AI correction as one of its three main risks to financial stability in October, the second was starting. The companies selling into the AI build still say they cannot make enough.

So far the rest of the US economy is holding. Output grew at an annual rate of 3.4% in the third quarter and unemployment is 4.1%. The question now is how long the damage stays inside the stock market. In 2000 and 2007, losses in the markets took months to reach jobs and output.

## Chip, Memory and Power Stocks Have Lost Between a Fifth and 60% of Their Value

The fall came in two waves. Memory makers, chip-equipment groups and most of the industrial companies supplying data centers peaked in late June. Caterpillar and GE Vernova both closed at their highs on 30 June, Micron on 25 June, Intel and the semiconductor index on 22 June. Nvidia, AMD and the Nasdaq kept rising into the autumn and peaked together on 6 October.

**Chart: AI chip, memory and power stocks are down between a fifth and 62% from their 2026 peaks**

Change in share price from each company's highest 2026 close, %.

| Company | Change from 2026 peak |
| --- | --- |
| SK hynix | −61.6% |
| Intel | −60.6% |
| Corning | −58.5% |
| Vertiv | −49.9% |
| Caterpillar | −46.4% |
| Micron | −40.8% |
| GE Vernova | −35.0% |
| Nvidia | −34.7% |
| AMD | −22.3% |

Source: Yahoo Finance closing prices, 2026 peak to 16 November 2026

The second wave was faster. On 5 November, two days after the midterm elections, the S&P 500 fell 4.4%, [its worst day since April 2025](https://www.sovereignmagazine.com/article/bank-of-england-rate-rise-nasdaq-correction-isa-pension), with all 11 sectors lower. AMD fell 17% that day, two days after reporting data center revenue up 97% and a day after its shares had risen on those results. The Bank of England had raised rates the same day, and US Treasury yields and oil prices were already high, so the election was one cause among several.

Nvidia has fallen 35% from its October high. Part of that is its own: its chief executive, Jensen Huang, [resigned last week](https://www.sovereignmagazine.com/article/jensen-huang-resigns-nvidia) while its board was investigating reports about him. Nvidia reports its quarterly results on Tuesday after the US close.

Measured by its whole membership, the market has held up better than the AI names. The S&P 500 was still up about 8% for the year at Friday's close, and an equal-weighted version of the index, which gives the smallest member the same weight as Nvidia, beat the normal index by about six percentage points in October. The yield on the 10-year Treasury, which moves opposite to bond prices, was 5.12% on Monday, up from 4.80% earlier this month, which raises the cost of every loan priced off it, including the ones paying for data centers.

## Orders for AI Chips and Equipment Are Still Rising

Microsoft's contracted future revenue from commercial customers stood at $677 billion at the end of September, up 73% from a year earlier, and its chief financial officer, Amy Hood, told analysts that customer demand continues to exceed available supply. Intel's data center revenue rose 78% in the third quarter, and the company says it is short of the parts it needs to make more server chips. Caterpillar's order backlog reached a record $78 billion, almost double a year earlier, and sales of its power generation equipment to end users rose 75%, which the company put down to demand for turbines and engines used at data centers.

No large cloud company has cut its spending plans because demand fell away. Microsoft trimmed its forecast for 2026 slightly because deliveries of third-party capacity slipped into 2027, and Alphabet and Amazon both say they will spend more next year.

Investors are still lining up for the companies that build AI models. Anthropic, the maker of Claude, is due to [set the price of its shares](https://www.sovereignmagazine.com/article/anthropic-ipo-price-range-roadshow) on Wednesday within a range of $70 to $78, and orders have reached several times the shares on offer, Bloomberg reported, citing people familiar with the deal. Nvidia, AMD, Micron and four other chip and cloud companies will also take $26 billion of Anthropic stock at the offer price.

The objection to reading those order books at face value is that buyers have been ordering more than they need. Paul Kedrosky, a venture capitalist and fellow at MIT, told the Financial Times that backlogs reflect expectations, with buyers "double and triple ordering to protect their position in the queue given perceived scarcity." If some of those orders are cancelled in 2027, today's backlog overstates the real demand, and the share prices are falling ahead of any proof either way.

Investors are now more worried about how the build is paid for and whether it will be allowed. Alphabet, Meta and Amazon have each reported negative free cash flow in recent quarters as their building outran their income, and they are borrowing to fill the gap. The price of insuring against default on debt issued by CoreWeave, the data center operator, had reached levels normally associated with companies in distress by mid-October.

## PC and Phone Sales Are Falling as AI Takes More of the Memory Supply

AI has already cut sales of PCs and phones, because they need the same memory chips. Memory makers have moved production toward the chips used in AI servers, and the price of the memory that goes into ordinary devices has risen sharply as a result. Worldwide PC shipments fell 22.4% in the third quarter from a year earlier, according to Gartner, which blamed higher prices and suppliers favoring their scarcest components for more expensive machines. IDC counted a 22% fall in smartphone shipments in the same quarter, the steepest it has recorded, with memory costs up about 350% in a year. We reported last month that SK hynix [expects the memory shortage to last through 2030](https://www.sovereignmagazine.com/article/how-long-will-the-memory-shortage-last).

Intel's results show both effects. Its total revenue rose 21% in the third quarter, carried by its data center chips, while its PC business fell and it forecast a weak fourth quarter, sending its shares down 28.6% on 30 October, [the company's worst day since 1974](https://www.sovereignmagazine.com/article/is-the-ai-bubble-bursting-october-2026). A shareholder lawsuit filed on 13 November in federal court in San Jose alleges that Intel did not disclose how much the PC slowdown was hurting the business before that forecast. Intel had told investors in July that PC demand would fall by a low double-digit percentage in 2026.

The device makers' shares have not followed their shipments down. Apple and HP trade above where they were at the end of June, while the memory makers profiting from the shortage have lost 40% to 62% from their peaks.

## The Trump Administration Pushed to Speed Up Data Centers

The AI build had the full backing of the federal government. Within days of taking office in January 2025, President Donald Trump announced the Stargate project with OpenAI, SoftBank and Oracle and revoked his predecessor's order on AI safety. In July 2025 he signed an executive order to speed up federal permits for data centers using more than 100 megawatts, which also offered federal land for them. In August the government took a 9.9% stake in Intel.

Republicans in Congress tried first to stop states from writing their own AI rules. A provision in the 2025 budget bill would have barred states from enforcing their AI laws for ten years, cut to five in a late compromise, and the Senate stripped it out by 99 votes to 1. In December 2025 the White House tried again. An executive order told the attorney general to set up an AI Litigation Task Force to challenge state AI laws that conflict with the order's policy, and told the Commerce Department to make states with what the order calls onerous AI laws ineligible for some of the remaining federal broadband money.

When states moved anyway, the president attacked them. After New York's governor, Kathy Hochul, paused state permits for large data centers in July, Trump wrote that "New York State has made a terrible decision." In August he wrote that "the only reason that communities throughout the U.S.A. should not want Data Centers is if they want to end up being backwards and poor." In a Reuters/Ipsos poll in June, 57% of Americans said they opposed a data center being built in their area.

The White House has also asked something of the industry. In March, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed a pledge to build, bring or buy their own power so that household bills do not rise to pay for data centers. The pledge is voluntary.

## States Are Now Pausing New Data Centers

Several states have now slowed the build themselves. New York's pause covers new data centers of 50 megawatts or more. Texas Governor Greg Abbott ordered an audit of the state's grid connection queue in August, where about 90% of the requests come from data centers, and paused approvals until it is done. Oregon has stopped issuing air and water permits for new or expanded data centers until July 2027.

On 3 November, [Democrats won control of Congress](https://www.sovereignmagazine.com/article/data-center-moratorium-midterms-ai), and Gina Hinojosa, Texas's governor-elect, promised to pause new data centers until the state legislature acts. She cannot do so before she takes office in January, and the new Congress has not set out what it will do. Democrats are split on permitting, and the bills that have passed with support from both parties so far are about who pays for power, not whether data centers get built. Trump's own orders remain in force.

## The Bank of England and the IMF Warned of an AI Correction

> "Equity market valuations appear stretched, particularly for technology companies focused on Artificial Intelligence (AI). This, when combined with increasing concentration within market indices, leaves equity markets particularly exposed should expectations around the impact of AI become less optimistic."
> — Bank of England Financial Policy Committee, October 2025

The Bank of England's Financial Policy Committee said in October 2025 that the risk of a sharp market correction had increased. After the first wave in July, the same committee said in September that AI-related and semiconductor stocks had fallen sharply but that there had been "no spillovers to core markets and no signs of broader systemic stress." It added that "the risk of a sharper correction with spillovers to core markets persisted."

The IMF's October report said that weaker investment or profits in AI "could trigger a correction in AI-related asset prices, exacerbated by rising leverage, concentrated exposures, and circular financing arrangements," and that countries with large holdings of AI assets were "likely to be more exposed to negative wealth effects or sharp market corrections."

Officials have called a problem contained at this stage before. In March 2007, the Federal Reserve chairman, Ben Bernanke, told Congress that "the impact on the broader economy and financial markets of the problems in the subprime market seems likely to be contained." The Nasdaq peaked in March 2000, and the recession that followed began a year later. In both cases the damage was in markets for months before it showed up in the official figures.

## The Rest of the US Economy Is Still Growing, but Leans on AI and on Rich Households

The latest figures show an economy that is not yet hurt. Output grew at an annual rate of 3.4% in the third quarter, with consumer spending and investment both rising, and employers added 165,000 jobs in October.

Some of that growth is AI. The Bureau of Economic Analysis said business equipment investment in the third quarter was led by computers and related gear. Jason Furman, a Harvard economist, estimated last year that spending on information processing equipment and software accounted for about 92% of US growth in the first half of 2025, while making up about 4% of the economy. Goldman Sachs estimates the measured boost this year is far smaller, because much of the equipment is imported.

There are three ways the losses in the stock market could reach the rest of the economy. The first is household spending. Shares made up a record 48% of American households' financial assets in the second quarter, after a $12.8 trillion rise in net worth that the Federal Reserve attributed mainly to stock market gains. The richest tenth of households account for about 45% of consumer spending, according to Moody's Analytics, although government survey data put the share far lower, and they own most of those shares, so a lasting fall in prices tends to show up in what they spend.

The second is debt. AI-related debt issuance worldwide reached about $450 billion by early September, more than double the total for all of 2025, according to a Morgan Stanley estimate cited by the Bank of England. The five largest cloud companies alone sold $222 billion of bonds in the first nine months of the year, the Bank for International Settlements said in October, and a further $75 billion to $100 billion of bonds and asset-backed securities linked to off-balance-sheet vehicles for data centers has been placed. Private credit funds had lent more than $200 billion to AI-related companies by the start of the year, up from close to nothing, by the BIS's count. Those loans do not trade on public markets, so losses on them take longest to appear, and banks are exposed both through construction loans and through their lending to the funds. A 10-year Treasury yield above 5% makes all of that more expensive to refinance.

Early estimates of what a debt problem will cost have been low before. In July 2007 Bernanke told the Senate Banking Committee that "some estimates are on the order of between $50 billion and $100 billion of losses associated with subprime credit products." Nine months later the IMF put potential losses from the crisis at $945 billion. Most AI debt is owed by companies with large cash flows, not by households with little to spare, so the lesson from 2007 is how quickly the estimates grew rather than how large the debt is today.

The third is the building itself. If moratoriums and higher borrowing costs stop projects, the losses land first on construction workers, power companies and the counties that have come to rely on data center taxes.

None of those has turned into lost jobs or falling output yet. Nvidia's results on Tuesday, HP's on 24 November and the government's second estimate of third-quarter growth on 25 November are the next figures that will show whether the damage is still contained.

## FAQ

**Q: Is there an AI bubble?**
Central banks and the IMF have warned since 2025 that AI stock valuations were stretched. Since their 2026 peaks, AI chip, memory and power stocks have fallen by between a fifth and 62%, which is the kind of correction those warnings described. The companies selling AI equipment still report more orders than they can fill, so the fall so far is in share prices rather than in demand.

**Q: Are AI stocks going to crash?**
Many already have. SK hynix is down about 62% from its June high, Intel about 61%, Corning about 59%, Caterpillar 46% and Nvidia 35% from its October high. The broader S&P 500 has fallen much less, because the rest of the index has held up.

**Q: What happens if an AI bubble bursts?**
The main risks to the wider economy are lower spending by wealthy households who own most shares, losses on the debt used to pay for data centers, and job losses if construction stops. The Bank of England said in September that the July fall in AI stocks had not spread to core markets, but warned that a sharper correction could.
