AI stock bubble: inside the Bank of England's warning

Editorial digest April 24, 2026
Last updated : 16:32

What did Sarah Breeden actually say?

Sarah Breeden is the Bank of England's deputy governor for financial stability. That is not a desk that issues loose warnings. On Thursday she told markets that record-high global equities "do not reflect the risks" now baked into the global economy, and that an "adjustment" is coming. She named three risk vectors: private credit, highly valued AI stocks, and "risky valuations" more generally.

Translated out of central-banker dialect: we think the market is wrong, and when it realises it is wrong, prices fall.

Officials at her level do not usually call tops. They hint, they hedge, they mention "pockets of exuberance". Breeden skipped the euphemisms. That alone is news.

Why is Intel breaking dotcom records the key tell?

While Breeden was warning London, a second data point landed in New York. Intel's share price surged to a level the Financial Times describes as surpassing its dotcom-bubble high, on the back of what its CEO called "fundamental" changes after a year-long turnaround. The fundamentals story may be true. The price story is harder to defend.

Consider what it takes for Intel — the company that largely missed the first AI chip wave, that was bailed out of the smartphone era, that surrendered its manufacturing lead to TSMC — to trade above its dotcom valuation. Either the company has reinvented itself more completely than its revenues yet show, or the market is paying dotcom prices for dotcom hopes. The second reading is the uncomfortable one.

Crossing a bubble peak is not, in itself, proof of a new bubble. Amazon spent years above its 2000 high while growing into the price. The warning sign is when many stocks cross that threshold simultaneously on the same thematic bet. That is where AI-exposed equities now sit.

What does Meta's 10% layoff actually tell us?

Meta will cut roughly one in ten employees. The BBC frames this as the company's largest redundancy round since 2023. The context is what matters: Meta has been pouring billions into AI infrastructure — chips, data centres, custom silicon, talent. The layoffs are not a signal that the AI bet failed. They are a signal that the bet has become so capital-intensive the company needs to fund it by cutting elsewhere.

Read alongside the stock-market story, the tension is clear. Investors are paying more for AI exposure, while AI-exposed companies are squeezing operating costs to keep financing the build-out. That is not unusual in a tech cycle. It is unusual when it happens across the entire Magnificent Seven at the same time, at valuations a deputy governor now describes as unanchored from risk.

There is another reading, less flattering to Meta. Productisation of AI was supposed to deliver revenue, not just cost. More than three years into the generative-AI cycle, the dominant corporate response from Big Tech looks the same as it did in 2023: more capex, more layoffs, more promises. At some point the question stops being when the returns arrive and starts being whether they arrive on the scale implied by today's prices.

Why is private credit the warning nobody is covering?

Breeden did not only name AI. She named private credit. This is the part of her statement that will not make headlines, and that is precisely why it matters.

Private credit — non-bank lending to corporates — has ballooned since 2008 because regulators pushed risky lending out of banks. The funds that now hold that paper face looser rules, lighter stress testing, and almost no public disclosure. The industry's pitch is that these loans are held to maturity by patient capital. The counter-argument — which the Bank of England has been making in increasingly direct language — is that nobody knows what these books really look like until somebody has to sell.

Linking AI and private credit in the same warning is deliberate. A meaningful slice of the AI build-out — data centres, GPU leasing, specialist cloud providers — is being financed off-balance-sheet through private credit vehicles. If the AI revenue story disappoints, the first casualties are not the listed tech giants. They are the private-debt funds that lent against AI cash flows which may not materialise on the schedule the model assumed.

That is a shadow-banking problem wearing an AI label.

Does the UK actually have exposure to this?

The UK retail investor is, in practice, a tracker investor. Pension auto-enrolment has quietly turned millions of British workers into leveraged holders of US mega-cap tech. Any default workplace pension or SIPP "global equity" option is typically 30%-plus exposed to the Magnificent Seven. The median UK saver does not know this, has never been asked to consider it, and has no mechanism to hedge it.

That is the domestic reason a Bank of England official cares about valuations set in Palo Alto. If Nvidia, Microsoft, Apple, Meta, Alphabet, Amazon and Tesla correct by a third, the British pension system takes a direct hit. Breeden's brief is financial stability. AI stocks are now a financial stability question for the UK.

Add the Guardian's story, from the same day's feed, of a young postgraduate spending her house-deposit savings to clear a student loan whose interest rate has ballooned. The macro picture and the micro one rhyme. A generation told to save, invest and defer consumption is watching the assets it was pushed into — equities, housing, education debt — reprice against it. The political consequences of a sharp correction would not be symmetrical.

What is different about this warning from the last ones?

Central banks have been muttering about valuations for three years. Very little has happened. Markets have learned that warnings are free and rate cuts are not. So why should this one read differently?

Three reasons.

First, the specificity. Breeden did not warn about "equities". She named AI stocks, private credit, and highly valued assets. Specific warnings age better than generic ones, and markets know it.

Second, the timing. The warning lands on the same day Intel retakes its dotcom-era high and Meta announces its biggest layoff in three years. Officials do not time speeches accidentally. The Bank is putting a marker down before a repricing, not after.

Third, the backdrop. IEA executive director Fatih Birol told the Guardian this week that the oil crisis triggered by the Iran war has changed the fossil-fuel industry for ever, and that the UK should forgo much of its potential North Sea expansion. Read that alongside the equity story and a pattern emerges: the 2020s are not the 2010s. Cheap energy, cheap money and cheap geopolitical risk are gone. Valuations priced against the old regime will struggle in the new one.

What happens if Breeden turns out to be right?

A 20% correction in US tech would shave roughly 8 to 10% off the equity sleeve of a typical UK workplace pension. That is annoying, not catastrophic, for anyone with a 20-year horizon. It is close to catastrophic for anyone already in drawdown who has been quietly overallocated to "growth" by default.

A wider correction that drags private credit with it is a different story. Banks would be hit indirectly through credit-line commitments to private funds. Insurers would be hit through illiquid-asset allocations that have become structurally large. The Bank of England's own stress tests have flagged these channels for years; the open question is whether the assumptions in those tests reflect the scale the market has actually reached.

The honest answer is: probably not. Private credit has grown sharply since the last stress regime was calibrated. AI-linked capex has gone vertical. The map the regulators are working from is older than the territory.

What should a UK reader take from all this?

Three things, stated plainly.

One — when a deputy governor for financial stability names specific asset classes as overvalued, treat it as signal, not noise. That is not how those officials usually speak.

Two — the AI trade and the private credit trade are now linked. A correction in one will transmit to the other. The 2026 version of the 2008 chain runs through GPU lease financing, not subprime mortgages, but the mechanics rhyme.

Three — if you hold a default UK pension, you are in this trade whether you chose to be or not. That is not an argument for panic. It is an argument for looking at an allocation most people never look at, and asking whether the risk you are carrying is the risk you would choose if you were starting from scratch.

Breeden's warning, for the people it concerns most, is information they have never been told they needed.