Nobody asked you before your retirement was placed on the AI trade, and nobody asked you before the trade fell over this week. State and local pension funds are holding between $578 billion and $707 billion of AI-related stock on behalf of teachers, firefighters, nurses and clerks who never bought a share of anything — and when Nvidia lost 9% of its value in a single session, the loss did not stay on a trading desk. It moved to a contribution rate, which means a school board, which means a classroom.

Equable Institute, a bipartisan nonprofit that studies public retirement systems, assembled the exposure from the funds’ own disclosures and published it on November 4. The number it settled on: $578.1 billion to $706.6 billion in AI-related companies as of June 30, 2026. That is 9% to 11% of the $6.42 trillion these systems hold in total, and Equable labels the range conservative — a floor assembled from what funds admit to, not a ceiling. (Equable Institute release)

The concentration is the part that should stop a reader cold. The 51 largest publicly traded AI-related companies accounted for 35.1% of all the public equity these funds reported holding. The Magnificent Seven alone made up 63.3% of the AI-related holdings. Nvidia stood at $47.3 billion, Apple at $42.5 billion, Microsoft at $28.2 billion. Three tickers, $118 billion, and not one of those positions was ever put to a vote of the people whose retirements they are.

Anthony Randazzo — The Funds Can See This Exposure, and Most of Them Have Not Shown You

Equable’s executive director, Anthony Randazzo, made a request that sounds almost quaint in a market this confident: every public pension fund should audit its own AI exposure internally and then publish what it finds. His reasoning is not doom. He concedes that a strong economic outcome from AI could meaningfully steady a lot of struggling plans. The problem is the other branch of the tree — his words land close to this: if the bubble bursts, the losses would be very meaningful, and the people who depend on those plans deserve to know the scale before it happens rather than after. (Equable interactive)

Notice what he is not asking for. Not divestment. Not a cap. Transparency. An audit and a PDF. In the year 2026, the radical demand from a pension analyst is that a fund write down what it owns, because right now the answer to “how much of my retirement is an AI bet” is a number that exists inside a consultant’s spreadsheet and nowhere a member can read it.

Nvidia’s $47.3 Billion — No Committee Chose This, a Formula Did

Here is the mechanism nobody explains at the retirement seminar. Most of this exposure was not selected. It arrived. Public funds run enormous cap-weighted mandates, and a cap-weighted mandate buys each company in proportion to its market value. As Nvidia, Apple and Microsoft climbed into the trillions, every passive dollar in the system automatically bought more of them. The bet grew because the bet was winning, which is exactly how a bet becomes a position you cannot exit without moving the price against yourself.

So the largest single AI holding across American public pensions, $47.3 billion in Nvidia, is not a conviction. It is an arithmetic consequence. The fund did not choose the AI trade any more than a tide chooses the moon. It matched an index, and the index quietly became a wager on one idea — that the spending on AI data centers keeps compounding for years. Passive investing did what passive investing does: it bought whatever got biggest. What got biggest was a single thesis wearing seven tickers.

The 9% Session — The Invoice Is Delivered to a School Board

The week the analysis landed, the market answered it. On November 5, Nvidia fell 9.26% to close at $181.60, the Nasdaq-100 dropped roughly 6.3%, Microsoft shed 3.58% and Apple slipped 1.53%. Nvidia finished that session about 25% below a 52-week high of $243.37. The sell-off followed Democratic midterm wins that put data center moratoriums on the table — a policy fight that lands directly on the companies whose growth depends on building more capacity. (24/7 Wall St.)

Now follow the money to the person. Nearly every one of these plans is a defined-benefit promise: a set monthly payment for life, regardless of what the portfolio does. When assets fall short of the promise, the gap gets closed by employers, by taxpayers, and sometimes by workers through higher contributions or a worse tier for the next hire. There is no third door. A pension loss is not absorbed; it is reassigned, and it is reassigned to the least mobile people in the building — the ones who took a below-market salary for twenty-five years because the pension was the deal.

That is the asymmetry Axel keeps coming back to. The upside of the AI trade shows up as a funded ratio in a board packet. The downside shows up as a millage increase, a hiring freeze, a shuttered fire company, or a legislature deciding that the retirement age for new employees should be five years later. Same dollar. Very different audience.

The $130 Billion Nobody Can Itemize — Private Equity Holds the Loss Off the Books

Equable’s upper estimate breaks into three pieces, and the third one should worry any trustee paying attention: about $396 billion in AI-related stock reported directly by 24 of the 25 largest funds, roughly $180 billion in public stock the funds hold but do not itemize by name, and about $130 billion in private equity and corporate debt. (Benzinga)

Public stock reprices every second, so you can watch the damage in real time whether you want to or not. Private holdings are marked on a schedule, by people with an interest in the mark. A decline there does not disappear — it just arrives late, in a quarter nobody is looking at, after the fiscal year has been certified. The lag is not fraud. It is the standard feature of an asset class that values itself quarterly while its public comparables move by the minute, and it means the honest answer to “how much did we lose” is: not yet known, possibly not yet admitted.

The 35.1% — Publish the Number, or the Number Publishes You

Some funds are far past the average. Reporting on this round of disclosures has put at least one large system close to a fifth of its assets in the AI trade, and TheStreet’s framing of the finding — in a piece titled after this exact idea — is the one that should be stapled to every trustee’s forehead: the concentration itself is the risk, not the bubble. (TheStreet) A fund can be perfectly right about AI and still be ruined by the path it takes to being right — a 6% index drawdown is survivable, a 40% one is a decade of higher contributions.

Randazzo’s audit is the cheap version of this conversation. The expensive version happens in a closed session after the loss is booked, when the choices are a contribution spike, a benefit cut for people who have not been hired yet, or both. Every one of those options has a human being at the end of it, and none of them were on the ballot.

The trade was placed by a formula, in the dark, on behalf of people who were told their retirement was safe because it was diversified. It was diversified across seven companies that increasingly move as one. The genie didn’t just escape the bottle. It bought the pension fund.

Sources: Equable Institute, 24/7 Wall St., Benzinga, TheStreet.