Summary

• The Pensions Regulator asks trustees to assess direct and indirect AI exposure across their full portfolios.
• Its guidance recommends stress scenarios, examination of changing asset correlations and challenges to investment advisers.
• The regulator has not prohibited AI investments or established that pension schemes have already incurred widespread AI-related losses.

The Pensions Regulator has urged UK pension trustees to examine their exposure to artificial intelligence across entire investment portfolios, warning that holdings spread between technology, infrastructure, property and energy may be more closely linked than conventional diversification measures suggest.

In guidance issued on 22 October, the regulator asks schemes to test how their investments would behave if enthusiasm for AI-related assets weakened and prices fell across several sectors together. Trustees should consider direct and indirect exposures, challenge investment advisers and take steps proportionate to their own circumstances.

The regulator has not banned investment in AI businesses or infrastructure, ordered pension funds to sell holdings or concluded that a collapse is inevitable. It is asking trustees to examine concentration risks before changing market conditions make those connections costly.

A pension portfolio can include shares in technology groups, debt issued by infrastructure businesses, property investments and utility holdings. These appear to represent different sources of returns, yet demand for AI computing may influence all of them through purchases of chips, data centre construction and the electricity required to run those facilities.

The regulator draws on concerns raised in the Bank of England’s July 2026 Financial Stability Report about growing connections within the AI supply chain and the incomplete visibility of some financing arrangements. When investments rely on overlapping assumptions about future demand, the apparent protection from holding several asset classes can weaken during a downturn.

Exposure extends beyond technology shares

AI infrastructure is financed through equity, corporate borrowing and other investment structures. A pension scheme holding a broad equity index may be exposed to leading technology firms, while another fund in the same portfolio could lend to a business developing the computing facilities those firms need.

Separate fund managers may each monitor concentration within their own mandates without seeing repeated holdings or correlated exposures elsewhere in the scheme. The Pensions Regulator therefore wants an aggregate assessment that brings those positions together rather than relying solely on reports from individual funds.

Passive strategies create a particular issue when investments track an index weighted by market capitalisation. As the market value of a group of AI-related companies rises, their combined weight can increase automatically, even if trustees have not made a separate decision to add technology exposure.

Active management does not necessarily avoid the problem. Several managers can invest in the same businesses, or hold different securities whose returns depend on the same underlying infrastructure spending. The number of mandates may give an impression of diversification that is not matched by the risks actually taken.

Indirect exposure is harder to identify because businesses outside the technology sector may depend on future demand for AI facilities. A utility could expect additional electricity sales, for example, while a property investment might depend on tenants operating data centres. Their prospects are not identical, but weaker investment or delayed infrastructure projects could affect both.

Trustees need information from their advisers and managers about those dependencies, including any financing relationships that are difficult to see in headline asset allocations. Identifying a connection does not establish that an investment is unsuitable; it provides evidence for deciding whether a scheme is comfortable with the combined risk.

Stress testing must allow correlations to change

The guidance recommends scenario analysis involving a severe but plausible decline in AI-related assets. Trustees should compare the outcome with existing stress tests and assess whether the scheme can withstand the losses and any resulting demand for liquidity.

Historical relationships between investments are not fixed. Shares, property and infrastructure debt may behave differently in ordinary trading conditions but decline together when investors reassess one source of growth. An assessment based only on past correlations could therefore understate losses during an abrupt change in expectations.

Leverage and liquidity can deepen that problem. A fund with borrowing or collateral obligations may have to raise cash after prices fall, potentially selling assets at an unfavourable time. The regulator wants trustees to ask their managers how such mechanisms would affect a scheme under pressure.

The emergence of new AI companies and additional debt issuance also changes exposure over time. Trustees should not assume that a concentration analysis remains adequate indefinitely, particularly where passive fund weights or private market financing arrangements evolve.

Testing should reflect each scheme’s investment objectives and risk appetite, rather than applying one universal allocation rule. A pension arrangement with a long investment horizon may have a different capacity to absorb volatility from one facing near-term cash commitments.

Communication forms part of resilience

The regulator also highlights communication with members of defined contribution schemes. A sharp market fall could prompt some savers to switch investments or stop contributing at an unhelpful point, potentially crystallising losses or reducing participation in a recovery.

Trustees should consider whether they have a proportionate, clearly explained communication plan that can be used promptly during periods of stress. Information should help members understand what happened without claiming certainty about short-term market movements or encouraging reactive decisions.

Those responsibilities sit alongside existing governance requirements. Trustees remain accountable for evaluating investment risks even when portfolio management and financial advice are delegated to specialists. They should be able to explain the assumptions behind their chosen investment strategy and the information used to assess it.

None of this establishes that all pension schemes carry excessive AI exposure, or that AI investment lacks long-term economic value. The regulator acknowledges the technology’s potential productivity benefits while emphasising that substantial financing has changed the composition and possible connections of global markets.

The practical outcome sought is a clearer view of where investment risks overlap. Pension trustees may decide to retain their allocations after conducting the analysis, but they should do so with evidence about their full exposure and how it could behave when conditions turn against the AI investment cycle.