The simple version
If you're in a pension scheme -- and most working people are -- part of your money is almost certainly invested in a handful of AI companies. Not because anyone chose to bet your retirement on AI. But because the stock market indices that pension funds track have become dominated by seven companies: Apple, Microsoft, NVIDIA, Amazon, Alphabet (Google), Meta (Facebook), and Tesla.
These seven companies are now 33% of the S&P 500. One third. All driven by the same story: that AI will transform everything.
Maybe it will. Maybe it won't. But your pension shouldn't be a one-way bet on the answer.
The numbers
We analysed four major UK pension arrangements using public data:
| Fund | Assets | AI Exposure | At Risk* |
|---|---|---|---|
| LGPS (local govt workers) | GBP 390bn | GBP 43.9bn (11.3%) | GBP 18.4bn |
| USS (university staff) | GBP 76.8bn | GBP 5.4bn (7.0%) | GBP 2.2bn |
| Railpen (railway workers) | GBP 37bn | GBP 4.1bn (11.1%) | GBP 1.7bn |
| BT Pension Scheme | GBP 33.2bn | GBP 1.9bn (5.7%) | GBP 0.7bn |
*"At risk" = potential loss in an AI bubble deflation scenario modelled by Oliver Wyman (January 2026), where AI stocks fall 50% and AI debt spreads widen.
The exposure comes through three hidden channels:
- Stocks: Passive index funds automatically buy whatever's biggest. Right now, that's AI.
- Bonds: AI companies have issued $1.2 trillion in debt. It's now 14% of the main investment-grade bond index -- bigger than all US banks combined.
- Private credit: Over $1 trillion of AI infrastructure debt is flowing through private credit. The servers backing these loans depreciate in about a year, but the loans assume they last 7-15 years.
The law says trustees must act
This isn't a suggestion. It's the law.
Investment Regulations 2005, Regulation 4: Pension assets must be "properly diversified" to "avoid excessive reliance on any particular asset, issuer or group of undertakings" and "avoid accumulations of risk in the portfolio as a whole."
Seven companies sharing the same AI-dependent revenue model making up a third of your equity index is exactly the concentration Regulation 4 was written to prevent.
The Pensions Regulator's own 2025 Annual Funding Statement names "artificial intelligence adoption" as a dynamic that "may impact investment strategies."
Case law: In Nestle v National Westminster Bank (1993), the court held that failure to review investments regularly is a breach of trust. In 2001, Merrill Lynch paid an estimated GBP 70 million to settle a claim from the Unilever pension fund over a concentrated investment approach.
Section 33 of the Pensions Act 1995: The investment duty of care cannot be excluded. Trustees cannot contract out of this obligation.
A trustee who holds passive global equity in 2026 without having specifically considered and documented their AI concentration risk is exposed to a breach of Regulation 4 and the prudent person standard.
The 2008 parallel
Oliver Wyman (January 2026) mapped the structural risks:
- $6 trillion in AI infrastructure debt held in opaque, off-balance-sheet vehicles
- Tech companies have moved $120 billion of data centre spending off their books into Special Purpose Vehicles
- These SPVs are funded by the same asset managers that run your pension money: BlackRock, PIMCO, Apollo, JPMorgan
- GPU chips depreciate in ~1 year but are being amortised over 7-15 years
- In a default, the collateral is worth a fraction of the loan value
This mirrors the structured product risks that caused the 2008 financial crisis. Different technology. Same structure. Same opacity.
What this isn't
This isn't a prediction that AI will fail. AI might transform everything. These companies might be worth every penny.
This is a statement of fact: your pension is concentrated in a way that the law says it shouldn't be, and most trustees don't know it because the concentration is hidden inside index funds and bond portfolios.
What you can do
If you're a pension scheme member
Ask your scheme: "What is our total exposure to AI-related companies across equities, bonds, and private credit?" If they can't answer, that's the problem.
If you're a trustee
Run the x-ray. Map your holdings through to the underlying AI concentration. Document that you've considered it. Fiduciary duty is about process. Get proper advice (Section 36 requires it).
If you're an adviser
Your clients need this analysis. We built the tool. It's available.
If you're a journalist
The data is public. The analysis is free. Contact [email protected].
A note about stories
The financial system runs on stories. Money is a story. Markets are stories. AI is the latest story.
Stories aren't bad. They're how humans coordinate. Cave paintings were stories. Pensions are stories about the future.
But when one story gets so big that a third of the global stock market depends on it, and nobody checks whether the people whose retirements are at stake know about it -- that's not a market failure. That's a communication failure.
The fix isn't complicated. Look at what you're exposed to. Understand it. Talk about it honestly. And be kind to the people on the other side of the table -- the trustees are mostly volunteers, the fund managers are doing their best, and nobody has this figured out.
Transparency. Honesty. Kindness. The rest follows.
Get the full analysis
Full methodology and individual fund reports available. The data is public. The tool is built. Let's talk.
Contact: [email protected]This analysis was produced by AgileMesh using public data, published research from Oliver Wyman, M&G Investments, Mercer, JP Morgan, Sage Advisory, the Bank of England, and The Pensions Regulator. Fund data from published annual reports. Full methodology and individual fund reports available at agilemesh.net.
AgileMesh builds cognitive layers for enterprises. The Lattice platform maps hidden connections in complex systems. This is what it does.