Navigating AI concentration risks in liability-aware portfolios | Insight Investment

AI-related concentration risks are rising. Although they remain modest across broad fixed income indices, pension plan sponsors may wish to monitor potential impacts on their LDI assets and discount rates.

The AI-related investment boom continues to reshape financial markets.

The hyperscalers (Amazon, Google, Microsoft, Meta, Oracle, and now SpaceX¹) alone project ~$5trn of additional capex out to 2030². As a result, AI-related issuance (which accelerated in the second half of 2025) has continued at pace. Year-to-date, issuers raised over $170bn in US dollar investment grade corporate bonds. Hyperscalers have also been increasingly issuing in other currencies, and we project a further ~$100bn of issuance across all currencies by early 2027.

Figure 1: AI-related issuance has been substantial through 2026 so far³

These firms and other AI-related issuers (like utilities, “neocloud” providers and others pivoting business models to the AI ecosystem) have also been busy tapping high yield, bank debt, private credit, joint ventures, vendor financing and equity issuance for finance.

Amid this “supply shock”, hyperscaler and tech credit spreads have widened, potentially creating compelling security selection opportunities for diligent bottom-up credit investors (Figure 2).

Figure 2: The AI “supply shock” has caused spreads to widen, even for the highest rated issuers³

AI fixed income concentrations are rising, particularly for liability-aware investors

The US investment grade market’s exposure to hyperscalers has risen from ~3% to ~5% over five years. This is dwarfed by the 20% to 56% concentrations that mainstream US equity indices face (Figure 3).

Liability-aware investors, like corporate defined benefit (DB) pension plans, may have more to consider, however. Many hyperscalers are AAA to AA rated (the least common ratings) so their higher quality index concentrations are significant.

Long US corporate bond strategies (often used to more closely match a pension plan liability’s interest rate sensitivity) have a modestly higher exposure to hyperscalers (Figure 3) than broad investment grade indices at ~8% (up from ~5% five years ago).

Figure 3: Broad bond indices still have modest hyperscaler concentration, but high quality indices are more exposed⁴

The impact of AI-issuance pension discount rates is worth watching

Concentration risks are significantly higher in AA long corporates where Hyperscalers account for a ~35% concentration (Figure 3 – above) and tech firms account for ~50%.

This is particularly notable for plan sponsors because AA corporate yields are often the basis for pension liability discount curves.

Discount-rate providers typically construct AA curves from screened bond universes using maturity-based averaging or regression techniques. Sometimes they exclude anomalously priced securities or offer issuer-capped curve variants, which can reduce the influence of highly concentrated issuers.

Currently, the impact of hyperscalers on pension discount curve is still relatively modest. Insight’s modeling indicates that if all AA hyperscalers were downgraded below AA, pension discount curves would likely flatten, with yields falling by up to 6bp at longer maturities (Figure 4). We estimate this may have a 20-40bp impact on pension benefit obligations for a typical plan.

Figure 4: Hyperscalers are pushing up longer-dated AA yields⁵

As AI-related issuance continues to ramp up, pension plans may wish to take a closer look at the construction of their discount curves, because the impact of hyperscaler issuance may vary. For example, if curve construction seeks to optimize discount rates for yield by filtering for the highest-yielding securities, this may produce more concentrated discount curves that are more sensitive to hyperscaler pricing and sudden downgrades.