Yesterday a new data tool hit production with an uncomfortable number attached: $45.92 billion. That is the AI-linked investment exposure a vendor's tagging engine found sitting inside 24 life insurance companies — $1.4 trillion in assets, roughly a fifth of the U.S. life insurance industry, as of December 31, 2025. (The data comes from a LinkedIn post by the creator of the "Insurance Intel" tool, released October 21; the tagging methodology is his, the SPV deals are admittedly undercounted, and the figures here are his — treat $45.92 billion as a floor, not a ceiling.)
Why this clears the bar
In yesterday's essay I wrote about W. Robert Berkley Jr. putting numbers on AI as an underwriting exposure — identified, priced, ring-fenced. This is the investment-side twin of that story, and it completes a frame every carrier executive should now be using: AI is three exposures, not one. Operations (the models running your business), underwriting (the risk in your insured portfolio), and investments (the risk in your investment portfolio). Most carriers are governing the first, bragging about the second on earnings calls, and haven't inventoried the third at all.
The composition is the tell. Seventy percent of the tagged exposure — $32.09 billion — sits in just 20 names. The top of the list reads like a roll call of the AI buildout itself: a $5.29 billion stake in an Intel chip-fabrication joint venture, Oracle ($3.26B), Amazon ($2.48B), the SoftBank Vision Fund ($2.35B), plus data-center ventures and the power companies that feed them — Energy Transfer, Dominion, QTS, Vantage, Compass, Switch. Carriers are not just selling to the AI economy. They are financing it.
The poster's sharpest claim — attributed, not asserted — concerns Oracle at number two. His deep dive concludes the company is not actually investment grade, and that credit-default-swap spreads agree with him. His argument: a downgrade would force insurers holding the paper to sell and would raise capital charges. You can quibble with his rating judgment — he himself asks whether IBM, Dominion, and Cisco really belong on an "AI exposure" list — but the mechanism is standard carrier economics. Investment-grade ratings are load-bearing walls in insurance balance sheets. If the poster is right, the question isn't whether Oracle counts as AI exposure; it's how many carriers would discover, in the same week, that their infrastructure-debt book repriced at once.
Investment-grade ratings are load-bearing walls in insurance balance sheets.
What I'd do from the operator's chair
If I were in the CIO or CRO seat this morning, three moves:
One: ask the investment team the three questions the underwriting team just got. Where does AI sit in our portfolio, are we being paid for it, and what happens to capital charges if a top issuer gets downgraded. Berkley gave the industry the script for the liability side; run it verbatim on the asset side. The investment committee and the underwriting committee should be reading the same exposure inventory.
Two: stress the concentration, not the average. Three percent of assets is a rounding error; 70 percent of the exposure in 20 names is a correlation story. The Apollo-linked holdings (AP Grange, SoftBank Vision Fund) are, per the poster, mostly contained in Athene — which means a small number of carriers hold a large share of the risk. Ask your risk team to model the correlated exit: what breaks if the AI-infrastructure credit story cracks and several insurers reach for the same door?
Three: give the three exposures one owner. Operations, underwriting, investments — one named executive who can answer the board's question: how much AI exposure do we carry, in total, across all three? Right now most carriers would hand the board three decks from three committees. The board doesn't have an AI exposure problem in three places. It has one exposure problem with three doors.
The real signal
The number that matters isn't $45.92 billion. It's that nobody inside the carriers knew it before a LinkedIn post produced it. An outside vendor tagged their portfolios before they did. That is the governance gap in one sentence: the industry is buying the AI trade at scale without a ledger that says so.
Berkley priced the exposure. This data inventories it. The carriers that close the loop between those two verbs — priced and inventoried — are the ones whose balance sheets survive the AI cycle. The rest are flying instruments they haven't built yet.
— Krishna