Megacap "Big Tech" concentration in benchmark equity indexes has long been a headache for portfolio investors and passive index trackers, but direct exposure to the single theme of AI takes it to another level.
Despite considerable uncertainty about AI's eventual payoff and impact, investors simply buying what is billed as a diversified S&P 500 stock index .SPX now take on 50% exposure to AI-linked firms — hyperscalers, chipmakers and AI infrastructure companies central to the theme.
Insulating portfolios from an AI accident with bonds and fixed income may once have been a plain-vanilla solution. But concentration is also rife there, as hyperscalers borrow like there's no tomorrow and command a growing share of U.S. investment-grade (IG) credit indexes.
AI-related bond issuers account for a third of all net issuance year-to-date, and Apollo's credit team estimates the five big AI hyperscalers alone could grow from less than 5% of the IG index today to nearly 10% by 2030.
Put another way, it expects the five big hyperscalers plus SpaceX SPCX.O to have a bigger share of IG indexes than the top five U.S. bank issuers by the end of the decade.
So, tilt to the junk-bond market instead? Caution is warranted there, too. Apollo points out that data-center financing is now one of the fastest-growing segments of the high-yield market, and borrowers such as Core Scientific CORZ.O and TeraWulf WULF.O are raising capital to transform existing bitcoin-mining facilities into AI data centers.
"Almost overnight, direct data center exposure in the HY (high yield) market has gone from virtually zero to nearly 5% of the index, which, as a stand-alone sector, would make it the sixth-largest sector constituent in HY."
And that's before we get to GPU-backed financing in the leveraged loan market or huge swathes of AI financing in private markets.
The upshot is that investors accustomed to diversifying across asset classes and sectors that traditionally perform differently in periods of acute stress are now taking on large chunks of the same risk wherever they go. Chipmaking, real estate, power and construction are all now loaded with the same theme.
"A portfolio with hyperscaler bonds, data center debt, power infrastructure, semiconductor exposure, GPU-backed financings, or AI-related equity issuance may appear diversified but contains assets highly related to the same underlying factor," Apollo said.
AI better not be a bubble, because there aren't many places left to hide if it pops. Carving out individual stocks and sectors may be necessary, and paper investments with no exposure to AI at all may, conversely, be bid up by funds desperate to diversify.