AI demand is real; the credit concentration needs scrutiny

Jordan Stuart's essay for Insurance AUM Journal opens with the standard objection to AI capital spending: trillions of dollars into chips, data centers, power facilities and orbital compute, a boom that invites the late-1990s comparison. Stuart, a vice president and investment director, says the rhyme is wrong. The dot-com cycle built infrastructure first and waited for customers. At the end of 1999, only a third of U.S. adults were online. Worldwide, fewer than one person in twenty had internet access. Demand was a future event. AI's largest providers, he writes, already run billions of daily active users, and every new capability — coding agents, voice, video, agentic tasks — gives those users another reason to show up. The build-out is behind demand.

Stuart expects the largest AI firms' capital expenditures to top $750 billion this year. Demand, he says, is outrunning construction schedules — the supply problem named in the essay's title. On valuation, he sees none of the dot-com tells. Memory chipmakers are at the center of the build-out. Profits in some cases are up several hundred percent year over year, and many of the stocks still trade at single-digit forward price-to-earnings multiples. These are earnings stories. IPO volume in 2026 is again approaching 2021 levels, but Stuart warns that nominal dollar comparisons overstate the cycle; issuance, he argues, should be measured against the size of the market.

For a general account, the essay's argument lands first in the credit book. Voya counted AI-related issuance at more than 15% of investment-grade bonds. Several $10 billion private placements are tied to the same build-out, as this publication reported. The capex Stuart describes is more than an equity story. It generates the new bond supply and private placement flow insurers are being asked to absorb.

The supply argument cuts two ways for insurers. It supports the durability of AI-related borrowers, which have revenue, margins and profits rather than projections. It also means concentration is building as fast as the spending. When one theme accounts for more than 15% of new investment-grade issuance, the individual credit decision becomes a single economic bet. The demand may be real; the concentration is what deserves bubble-style skepticism.

The dot-com comparison fails on its own terms. Demand arrived before supply, so the AI build-out is a growth story, not a cautionary tale. It does not follow that concentration is harmless. The general account does not need to pronounce on the bubble. It needs to examine whether the investment-grade book has quietly become a one-theme portfolio.