Three channels of hidden AI concentration in UK pensions. One practical strategy for what trustees can do about it.
12 March 2026
We published our first analysis showing that UK pension funds have significant hidden exposure to AI companies. Since then, several readers -- including trustees, advisers, and scheme managers -- have asked the obvious question: so what do we actually do?
This is an attempt at an answer. Not a prediction about whether AI will succeed or fail. Not financial advice. A structural strategy for managing concentration risk that already exists in most UK pension portfolios, whether anyone chose to put it there or not.
The problem is real. The good news is that the tools to address it already exist.
Most discussions about AI concentration focus on equities. That misses two-thirds of the picture. UK pension funds are exposed through three channels simultaneously, and the channels are connected.
The Magnificent Seven -- Apple, Microsoft, NVIDIA, Amazon, Alphabet, Meta, Tesla -- now make up 33.5% of the S&P 500 and 24% of the MSCI World Index. Any pension fund tracking a global equity index is automatically concentrated in these seven stocks. Not by choice. By arithmetic.
This is not normal. Man Group's analysis shows concentration levels not seen since the Great Depression and the late-1990s tech bubble. Historically, whenever the spread between the largest and next-largest market decile reached today's levels, the largest decile has never outperformed over the following five years.
This is the less-reported story, and arguably the more important one for UK DB schemes, where 69% of assets are in bonds.
JP Morgan estimates that AI-linked companies will account for 14% of its investment-grade bond index by 2026. Mercer explicitly compared this to the TMT boom of the early 2000s, when telecoms companies "engaged in bidding war for mobile licences and then underperformed expectations." During that bubble, 85-95% of fibre optic cable laid went unused after the burst.
The parallel today: AI companies loading SPVs with debt, in a bidding war for data centre capacity, using hardware that depreciates in months but is financed over years.
The Financial Times reported in December 2025 that tech companies have created $120 billion in off-balance-sheet data centre spending through Special Purpose Vehicles. Meta's "Beignet Investor" SPV alone channelled $18 billion from PIMCO and $3 billion from BlackRock. These SPV bonds get investment-grade ratings and flow into the same bond funds pension schemes hold.
The funding ratio is actually the good news here. UK DB schemes are in their strongest position in years. That creates a window of opportunity: schemes can afford to act now, from a position of strength, before they are forced to act from a position of weakness.
Here is the structural problem in the UK regulatory architecture:
The Bank of England has flagged the risk clearly. Their December 2025 Financial Stability Report warned of "deeper links between AI firms and credit markets" and "increasing interconnections" that could amplify losses. But the BoE has no direct authority over pension fund asset allocation.
The Pensions Regulator has the authority to set investment standards for DB schemes. Their 2025 Annual Funding Statement warns trustees to "keep in mind the potential for heightened trade and geopolitical uncertainty." But they have issued no specific guidance on AI or technology concentration.
The FCA regulates fund managers and introduced concentration risk disclosure rules in January 2024. But they do not regulate pension trustees directly.
The Bank of England sees the risk but cannot act. The Pensions Regulator can act but has not looked. The FCA covers the middle but not the ends. Nobody is responsible for the aggregate picture.
This is not a criticism. It is a description. The regulators are doing their jobs within their mandates. The gap exists between the mandates.
The good news is that several major UK pension managers are already addressing this. Nobody is alone in recognising the problem.
Their GBP 36 billion Sustainable Multi Asset Fund (2 million members) is actively reducing US stock exposure and increasing UK and Asian allocation. Callum Stewart, Head of Investment Proposition: "We recognise the specific risks associated with US equities, such as tariff measures and the concentration in large tech stocks."
The state-backed DC scheme is exploring methods to reduce younger savers' dependence on "a single high-growth story." DC savers in growth phases can have 17-24% of their pension in seven stocks.
Pensions Expert reports that schemes managing GBP 200 billion or more are collectively shifting away from US equities through geographic diversification, hedging, and adding downside protection.
This is the practical part. Every step below uses tools and frameworks that already exist in UK pension governance. Nothing here requires regulatory change, new legislation, or heroic assumptions. It requires attention.
Most trustees do not know their total AI exposure across all three channels. The first step is simply to ask. Write to your investment managers and ask them to provide:
1. What is our total exposure to Magnificent Seven stocks across all equity mandates, including through index funds?
2. What proportion of our bond holdings are in AI-related or technology issuers, including through investment-grade indices?
3. Do any of our private credit, infrastructure, or alternative mandates include exposure to AI data centre SPVs, directly or through fund-of-fund structures?
If they cannot answer question three, that is itself useful information. Document the fact that you asked, and document the gap. Under the DB Funding Code of Practice (effective September 2024), trustees are required to assess their supportable risk levels. Asking these questions is part of that process.
Run a scenario. What happens to your funding level if:
AI stocks fall 30% (roughly the TMT correction of 2000-2001 in its first year). AI-linked bond spreads widen by 150 basis points. What is the impact on your liability-driven investment strategy?
AI stocks fall 50%. SPV-backed bonds lose liquidity. Bond spreads widen across the technology sector. This is the Oliver Wyman scenario from January 2026. What does it do to your funding ratio?
All three channels correct simultaneously -- because they are all driven by the same underlying story. Equity losses coincide with bond spread widening and private credit illiquidity. This is the scenario nobody models because it treats the channels as independent. They are not.
The point of stress testing is not to predict the future. It is to know what you do not know.
The industry advisers are already publishing practical diversification strategies:
Consider equal-weight equity indices (e.g. S&P 500 Equal Weight) alongside or instead of market-cap-weighted indices. This mechanically reduces concentration in the largest stocks without requiring active management or sector views.
John Roe at LGIM advocates higher European allocation, noting that "Europe's economy is much closer in size to the US than its listed equity market would suggest." Japan identified as a structural diversifier.
Hiroki Hashimoto flags that M7 concentration also creates dollar concentration. Separating asset allocation from currency hedging decisions gives trustees an additional lever.
Actively monitor the technology and AI weighting within investment-grade bond portfolios. The 14% AI-linked share of the IG index means passive bond strategies carry the same concentration risk as passive equity strategies. Consider bespoke bond mandates that cap single-sector exposure.
This may be the most important step. Under Regulation 4 of the Occupational Pension Schemes (Investment) Regulations 2005, pension assets must be "properly diversified" to "avoid excessive reliance on any particular asset, issuer or group of undertakings."
Trustees who hold passive global equity in 2026 without having specifically considered AI concentration risk are potentially exposed. But trustees who have considered it, documented their reasoning, taken professional advice (as required by Section 36 of the Pensions Act 1995), and made a deliberate decision -- even if that decision is to maintain current allocations -- are on much stronger ground.
We want to be clear about what this is not.
This is not a prediction that AI will fail. AI is genuinely transformative technology. These companies may be worth every penny of their current valuations and more.
This is not an attack on index investing. Passive strategies have served pension members well for decades. They remain excellent tools.
This is not a criticism of trustees. Most trustees are volunteers, giving their time to protect other people's retirements. That deserves respect, not finger-pointing.
What this is: a flag that the market structure has changed in ways that create concentration risk through channels that existing governance may not be monitoring. The 125% aggregate funding ratio means UK schemes can address this from a position of strength. That is a good position to be in. Use it.
Our detailed analysis of individual fund exposure, with methodology and sources:
Your Pension's Hidden AI Bet →
Our analysis of communication architecture in insurance (the same structural blindness in a different industry):
AgileMesh Lattice maps hidden concentrations across multi-channel portfolios.
We built this tool. It works. Let's talk.