{"id":65711,"date":"2026-11-13T16:22:18","date_gmt":"2026-11-13T20:22:18","guid":{"rendered":"https:\/\/www.cambridgeassociates.com\/?p=65711"},"modified":"2026-11-13T16:22:20","modified_gmt":"2026-11-13T20:22:20","slug":"ai-ambitions-pension-obligations","status":"publish","type":"post","link":"https:\/\/www.cambridgeassociates.com\/en-eu\/insight\/ai-ambitions-pension-obligations\/","title":{"rendered":"AI Ambitions. Pension Obligations."},"content":{"rendered":"

Artificial intelligence (AI) seems to be everywhere. It writes emails, answers questions, builds software, and if you believe every earnings call, will soon transform just about every industry. For investors, the story has largely played out in equities, where a small group of companies has driven a meaningful share of market returns. But AI\u2019s influence is no longer limited to equity markets.<\/p>\n

Building AI requires data centers, chips, power, and networking equipment\u2014basically, a great deal of capital. Increasingly, the companies funding that buildout are doing so through investment-grade corporate debt. For US-defined benefit plans, many of the same bonds are held in liability-hedging portfolios and are used to set the discount rate to value pension obligations.<\/p>\n

So, what happens if the AI trade disappoints? A reversal could affect pension funded status through familiar channels such as credit deterioration and equity declines, but also through a less obvious one: the composition and yield of the AA-rated corporate bond universe used in liability measurement. This piece examines AI risk from multiple angles and the potential steps plan sponsors can take to manage them.<\/p>\n

Why AI concentration matters in the AA-rated corporate universe<\/h2>\n

Under accounting guidance, pension discount rates are generally based on high-quality fixed income investments. In practice, many actuaries and plan sponsors rely on AA-rated corporate bond universes or AA yield curves to estimate those rates. Given the relatively small size of this subset relative to the broader credit universe, concentration within that market can have a material impact on pension liabilities.<\/p>\n

Concentration in the index has been a concern for quite some time, as fewer corporate issuers have had AA ratings. However, this concentration has become much greater recently, predominantly as more AI-related firms have come to rely on debt markets to build out the infrastructure needed for AI compute.<\/p>\n

Figure 1 illustrates the growing share of AI-related debt in the relevant AA-rated corporate bond universe. Given the longer duration nature of pension liabilities, the Bloomberg US Corporate Long AA Index was used for reference, but the general trend applies to all maturities. While the precise exposure depends on how \u201cAI-related\u201d is defined,<\/span> 1<\/a><\/sup> issuance linked to AI has become a larger portion of the opportunity set. For sponsors using AA-based discount rate methodologies, this creates greater sensitivity to the credit profile of a relatively small set of issuers. The concentration of AI-related names in the index has nearly doubled since December 31, 2024, increasing from 21% to 40% as of September 30, 2026. With the amount of capital needed to continue expanding the AI buildout, this concentration is likely to grow over time.<\/p>\n

\"Stacked<\/p>\n

So far, this has been a positive development for US pension plans as AI firms have come to market with higher yields to incentivize investors for increased debt supply and uncertain investment returns. Using the Blomberg US Corporate Long AA Index as a proxy, the average yield on the index is 6.50%. However, the average yield for AI-related names is 6.83%, while the yield on non-AI names is 6.27%, indicating a 56-basis point (bp) gap between AI-related and non-AI related firms (Figure 2). This has increased the discount rate, meaning that pension liabilities have declined during this period.<\/p>\n

\"Bar<\/p>\n

The current higher yields benefit investors and reduce pension liabilities. However, if a meaningful share of higher-yielding issuers were downgraded out of the AA-only universe, their removal could lower the discount rate and increase reported liabilities, all else equal. Depending on the duration profile of the liability and the type of yield curve methodology used, the increase in liability could range from 0.5% to more than 2.0%, assuming only AI names get downgraded.<\/p>\n

Plans using a bond matching approach to determine the discount rate are especially vulnerable as this methodology tends to pick out the highest yielding names. Although these methodologies generally impose issuer concentration limits, they do not impose sector-level limits or caps on aggregate exposure to hyperscalers. As such, downgrades of these issuers could therefore exclude a substantial share of bonds from the discount rate calculation, amplifying the liability impact noted above.<\/p>\n

What about total portfolio impact?<\/h2>\n

The balance sheet effect may extend beyond the liability calculation into the asset component. Many plans hold public equities, high-yield credit, REITs, infrastructure, and other risk assets with direct or indirect exposure to the same underlying AI theme. An adverse repricing of AI-related sectors could weaken funded status through multiple channels simultaneously: declines in growth-sensitive assets such as equities, wider spreads in credit holdings, and lower liability discount rates if downgrades alter the composition of an AA-based liability benchmark.<\/p>\n

Most pension plans hold some allocation to investment-grade credit. Typically, these mandates are broad and cover the entire investment-grade universe, not just the AA-rated names. Even in this case, 13% of the Bloomberg Long US Corporate Index is directly tied to hyperscalers or AI names.<\/span> 2<\/a><\/sup> While it would take a major event for AI names to fall out of the investment-grade index, a downgrade in these names would reduce the value of the fixed income portfolio.<\/p>\n

Portfolio implications<\/h2>\n

From a pure liability management perspective, there are limited options available as the use of AA-rated bonds for discount rate determination is generally required by US accounting guidance. Reducing reliance on a bond matching approach and migrating to using a larger portion of the universe could mitigate any future volatility and reduce the risk, although certainly not eliminate it. Measuring the liability on a broader index, such as bonds rated A or better can help smooth out the impact of the AI-issuance risk. While this methodology cannot be used for accounting purposes, it can create a more holistic view of the liability, which plan sponsors can use to design investment strategies, such as glide paths or any other aspects relying on a mark-to-market funded status.<\/p>\n

Perhaps more options exist on the asset side of pension plan management. A common theme has been to underweight credit risk in general, especially as spreads have remained near all-time lows (Figure 3). As of September 30, 2026, the average option-adjusted spread on the Bloomberg US Long Credit Index was just under 100 bps, compared to the long-run average of 166 bps.<\/p>\n

\"Line<\/p>\n

At current levels, taking credit risk is not ideal as it creates an asymmetric risk profile. Credit spreads may stay at these levels or even continue to grind lower by 10 bps\u201320 bps and break through all-time low levels not seen since 1998, but they can also widen by 200+ bps from a severe market downturn. Reducing this risk may be most prudent now and with the right approach, can help protect funded status from the AI concentration. Figure 4 estimates the return expectations from these credit events on both the Bloomberg US Long Credit Index and Bloomberg US Intermediate Credit Index.<\/p>\n

\"bar<\/p>\n

It\u2019s worth noting this phenomenon is different than equity-related risk, as equities can continue outperforming even as credit spreads remain range bound. Due to this, it may be more prudent to take AI exposure to equities than relying on credit markets.<\/p>\n

There are two options to consider when plan sponsors wish to reduce credit risk.<\/p>\n

    \n
  1. Reduce long credit exposure for intermediate credit exposure.<\/strong> As foreshadowed in Figure 4, this is a pragmatic approach to reduce duration times spread (DTS), helping reduce funded status deterioration from widening credit spreads. In essence, if the credit spread duration of the fixed income portfolio is lower than the credit spread duration of the liability, funded status should improve from a widening event. However, it is important to note that the full impact of the liability decrease may not be realized due to the AI-issuance downgrade discussed in this paper. Nonetheless, shifting the portfolio toward intermediate-duration bonds can reduce the impact of spread widening on fixed income holdings and help protect funded status. This shift does not eliminate that risk, however, because the Bloomberg US Intermediate Credit Index is also exposed to AI-related issuance. There are also a few other considerations for this trade:<\/li>\n<\/ol>\n