Central Bank and the AI Risk: The Pattern Making a Comeback

The precedent: December 1996

In December 1996 the Chairman of the Federal Reserve uttered two words before the American Enterprise Institute: «irrational exuberance». He was describing a precise mechanism.

Inflated equity valuations, growing investor leverage, capital concentrated in a handful of technology companies. The market ignored the diagnosis for more than three years.

Then came March 2000. The Nasdaq lost much of its value over the following twenty-four months, and the capital cycle changed shape.

A central banker had mapped the pattern well in advance. His institution chose to observe, while leverage continued to accumulate.

Thirty years on, the structure returns. The actors and the technology change. The geometry remains identical.

The current pattern: Bailey's letter

On August 31, 2026, Andrew Bailey, Governor of the Bank of England and Chair of the Financial Stability Board, wrote to G20 finance ministers. He warned that a collapse in growth in the artificial intelligence sector could trigger a «future market correction» of global scope, as reported by the BBC^[1].

His central argument is surgical. The risk arises from the interaction of three variables.

• Elevated equity valuations
• Growing investor leverage
• Concentration of money in a few large technology companies

Bailey added a detail that deserves attention: the growing «cross-investment» between AI companies and hyperscalers. This cross-financing architecture amplifies any future correction.

A group of 100 companies, including Google, Microsoft, Anthropic and OpenAI, had requested just weeks earlier that cyber defenses be strengthened. The signal converges from multiple directions.

The mechanism: why concentration amplifies

The correlation here is obvious. The causal mechanism deserves precision.

When a few companies hold reciprocal capital stakes, their balance sheets become interdependent. The capital of A finances the revenues of B, which in turn inflate the valuation of A. The structure works on the way up and accelerates on the way down.

Add leverage. Investors borrow to amplify their exposure to the same names. A 20% drop in concentrated stocks transmits to portfolios built on debt against those very stocks.

The Financial Stability Board officially monitors treasury officials, banks and regulators in the United States, United Kingdom, France, Germany, Canada, Japan, Australia, China and Saudi Arabia. When its chair writes, he is mapping a systemic risk, rather than commenting on a single stock.

That is the point. The central bank is watching the plumbing of the system.

The geopolitical layer: energy and war

Bailey also cited the volatility generated by energy supply shocks linked to the war between the United States and Iran. This layer changes the picture.

An AI market correction, overlaid on an energy shock, hits two channels at once: financial wealth and the real cost of productive inputs. Risk models tend to treat them as separate events.

History counsels caution. In 1973 the Arab oil embargo quadrupled the price of crude within a few months, and Western equity markets lost ground for two years. Energy shock and financial correction reinforced each other.

The current combination bears the same signature. An overvalued sector, a source of external volatility, leveraged balance sheets.

The central bank knows this signature. It cites it because it recognizes it.

AI becomes state infrastructure

There is a second geopolitical layer. Artificial intelligence is shifting from a commercial product to a national security infrastructure.

The Pentagon now has its own version of conversational AI models, as documented by TechCrunch^[2]. The race for «sovereign AI» is pushing governments to build internal capabilities and reduce dependence on foreign suppliers.

The technical literature feeds the same concern. Recent research on arXiv^[3] studies the behavior of AI agents when faced with security controls. Bailey's cyber warning finds its foundation here.

When the same technology drives both equity valuations and national security, the risk of a correction becomes a strategic risk. Private capital is financing critical public infrastructure.

My position

Here is the thesis. Bailey's warning will be filed away by markets as noise, exactly as happened after 1996, and concentration will continue to rise until the breaking point.

The consensus treats these letters as institutional theatre. It is wrong. When the FSB Chair writes to G20 ministers, he is documenting a risk that VAR models today exclude.

This is a regime change, rather than a cycle. The thirty-year bond bull run is over, and portfolio construction based on stock-bond decorrelation belongs to the previous regime.

What would change my reading? One precise data point: a measurable reduction in cross-investment between AI companies and hyperscalers, visible in quarterly balance sheets. As long as that figure rises, the risk rises with it.

Three implications for capital

The diagnosis remains abstract until it touches allocation. Here is my translation.

1. Family offices and sovereign wealth funds (36-month horizon): reduce overexposure to concentrated AI stocks and diversify toward real assets and emerging markets outside the dollar's orbit.
2. Chief Risk Officers (18-month horizon): incorporate into models the joint scenario of an AI correction plus an energy shock, correlating the two events rather than treating them as independent.
3. CFOs and Investor Relations (12-month horizon): revisit the growth narrative built on tech valuations. A downward revision stress-tested today costs less than one forced upon you tomorrow.

Each horizon carries its own urgency. Sequence matters as much as direction.

The forecast

I am framing the thesis in a verifiable way. The Financial Stability Board will publish by December 2027 a formal report classifying concentration in the AI sector as an explicit systemic risk for global financial stability.

Confidence: 70%. Horizon: December 2027. Verification: the publication of an FSB document that openly names AI concentration among the systemic risks being monitored.

The signal that would disprove the thesis remains simple. The absence of such a report by that date, accompanied by a reduction in AI-hyperscaler cross-investment in public balance sheets.

What to watch

• Hyperscaler quarterly balance sheets: the cross-financing line item directed toward AI companies
• Aggregate margin debt on US markets, a direct indicator of leverage
• The price of crude oil and energy volatility premiums linked to US-Iran friction

These three indicators will confirm or disprove the pattern. The first to move will set the timing.

The divergence between market calm and regulatory alarm always resolves. The question remains how.