The Pensions Regulator has asked pension trustees to examine their portfolios for hidden exposure to artificial intelligence, warning that holdings spread across several sectors may still depend on the same underlying technology investment cycle.
Guidance published on 22 October asks schemes to identify direct and indirect AI-related holdings, test their resilience against plausible market stress and challenge advisers on how diversification might perform if previously separate assets begin to move together.
The regulator’s concern is not confined to shares in major software companies. AI infrastructure requires power, data centres, property, networks, semiconductor production and financing. Pension portfolios can therefore accumulate exposure through technology equities, utilities, infrastructure funds, real estate and credit markets, even where each holding appears to occupy a different part of the economy.
Such connections are not automatically problematic. Long-term investment in AI could create substantial productivity and commercial benefits. The risk is that trustees assume exposure has been dispersed when a broad range of investments would all be affected by a disappointing AI adoption cycle, lower infrastructure demand or a sudden change in financing conditions.
The new guidance asks trustees to take a portfolio-wide view rather than relying on analyses prepared independently for each fund. A manager may monitor concentrations within its mandate while missing holdings or dependencies shared with other managers. Trustees are expected to ask how these exposures combine at scheme level.
Passive investment deserves particular attention. Funds weighted by market capitalisation automatically allocate more capital to companies as their valuations grow. When a narrow group of technology-related businesses becomes a larger share of the index, a nominally broad fund may carry increasing exposure to the same earnings and valuation assumptions.
Active portfolios have a different information problem. Multiple managers can hold similar securities or securities financed by connected businesses, while underlying assets in private markets may be less transparent. Reviewing fund names or high-level asset classes alone may not reveal those overlaps.
The regulator wants scenarios that test severe but plausible losses, changes in asset correlations, demands for liquidity and any amplification from leverage. Assessments should reflect each scheme’s structure, funding position, investment strategy and capacity to absorb volatility. There is no suggestion that trustees must abandon AI-related holdings or apply a universal exposure cap.
Communication planning is another consideration for schemes with defined contribution members. Sharp declines can prompt members to switch investments at disadvantageous moments, locking in losses rather than waiting for a possible recovery. Trustees can examine how to explain concentration and investment risk clearly without implying that AI exposure is necessarily inappropriate for every saver.
Regulators’ broader concern is that AI businesses, infrastructure providers and investors can share financing and demand assumptions even where their securities appear in different asset categories. A concentrated investment cycle can make conventional assumptions about market correlations unreliable just when investors depend most on diversification.
For pension employers and trustees, the immediate implication is a more demanding conversation with advisers. The relevant questions concern not merely which funds are owned but which economic assumptions, infrastructure demands and financing arrangements drive their performance.
The guidance has implications for governance records as well as portfolio construction. Trustee boards should be able to demonstrate what information they requested, which assumptions they challenged and how the discussion affected investment decisions. A broad assurance from a manager that a fund is diversified may be insufficient where several funds rely on the same technology spending cycle. Schemes also differ in their sensitivity to short-term losses: a mature defined benefit scheme paying pensions and a younger defined contribution default fund may require different liquidity assumptions, time horizons and member communications.
The regulator has framed its expectations around proportionate governance. Schemes should document the exposures they understand, the uncertainties they cannot fully measure, and the actions they would consider under different conditions rather than treating AI’s growth prospects as a reason to bypass established risk disciplines.
You must be logged in to post a comment.