The largest price-insensitive buyer in the US bond market is a rebalancing formula. It shows up with 38 cents of every dollar the AI build-out is issuing, and 17 cents of the duration.
Target-date funds held $4.855 trillion at the end of 2025. They rebalance to a fixed stock-and-bond mix without asking anyone. With US stocks up 13.07% this year through August and the broad bond index down 0.58%, that formula has already forced roughly $136 billion out of equities, of which about $113 billion lands in bonds. It covers 38% of 2026’s AI-related investment-grade supply measured in dollars, and 17% measured in duration, because the money buys an index with a 5.5-year clock while data-center paper is being written at 30 and 40 years. The automatic buyer takes the dollars and leaves the duration.
Government bond yields hit multi-decade highs this week and The New York Times published the sensible thing to publish: mostly, do nothing, and consider rebalancing. Millions of American workers had already rebalanced. They did not read the article, they did not log in, and they did not decide. A formula written into their plan documents did it for them, and it has been doing it all year.
That formula is now one of the largest standing bids in fixed income. It is also the wrong shape for what the market is trying to sell.
What Happened
Investors sold sovereign debt hard on Tuesday, September 1. The US 10-year note reached about 4.78%, its highest since January 2025. The 30-year pushed past 5.3%, a level last seen in 2007. Japan’s 10-year touched 3% for the first time since 1996. UK 10-year gilts cleared 5.2% and the 30-year gilt reached 5.89%, the worst since 1998. Bloomberg’s global sovereign gauge rose for a fourth straight session to 3.72%, the highest since the middle of 2008. The Treasury responded by at least doubling the maximum size of its long-dated buybacks from $2 billion to $4 billion, running September through November, which pulled the 30-year back to 5.19%.
The Times ran the reader-facing version on September 2: what does this mean for your money? The answer given was reasonable and, on its own terms, correct. Most people own bonds inside a broad fund, that fund has not lost much, and higher yields make future returns better. Vanguard’s Matt Wrzesniewsky supplied the line the piece was built around, calling income itself a powerful diversifier.
The reassurance holds, but it rests on a chosen window. Vanguard Total Bond Market Index Admiral returned 2.63% over the trailing year. Year to date it is down 0.58%, and over three months down 0.75%. Look back twelve months and the fund is fine. Look back eight and it is not. Both statements are true, and only one of them was in the article.
None of which is the interesting part. The interesting part is what the same funds were doing to the market while their holders were being told to sit still.
The Backstory
The Pension Protection Act of 2006 made target-date funds an approved default for automatically enrolled workers. That single safe-harbor provision turned a niche product into the plumbing of American retirement. Assets went from under $8 billion in 2000 to $4.855 trillion at the end of 2025, per the Investment Company Institute, up 20.3% in a year on Morningstar’s count. Sway Research puts the total near $5.2 trillion once custom strategies are included. Five firms hold 80% of it. Vanguard alone runs about $1.8 trillion, a 37% share.
Vanguard’s own How America Saves, now in its 25th edition and drawn from roughly five million participants, shows how thoroughly the decision has been removed from the saver. At the end of 2025, 96% of its plans offered target-date funds and 98% of auto-enrollment plans used one as the default. Of participants offered them, 84% used them, and 73% of those held a single fund matching their retirement year. Sixty-nine percent of all participants had their entire balance in a professionally managed allocation, against 9% in 2005.
Then the number that matters for the bond market. Only 5% of non-advised participants made any trade in 2025. Among participants invested purely in a target-date fund, the figure was 1%.
That is the technical definition of a price-insensitive buyer. Ninety-nine percent of the largest pool of managed retirement money in the world does not respond to price, does not respond to yield, and does not respond to headlines. It responds to a payroll cycle and a glide path.
The Plan
Nobody at these funds is deciding to buy bonds this week. The buying is an identity.
Start a fund at equity weight w. Let equities return R and bonds return B. To get back to w, the fund must sell exactly w(1-w)(R-B) per dollar of starting assets. That is the whole mechanism, and it is why the trade is largest for the balanced funds in the middle of the glide path and smallest at the extremes.
The ICI publishes the aggregate allocation. Across all target-date mutual funds at year-end 2025: 42% domestic equity, 29% world equity, 24% domestic bonds, 5% other. Equity weight, 71%. Through August 31, the S&P 500 returned 13.07% including dividends while the broad bond index returned minus 0.58%, a gap of 13.65 points.
Run it. The rebalancing trade is 2.81% of assets. On the ICI’s $4.855 trillion base, that is $136 billion leaving equities. Since domestic bonds are 24 of the 29 non-equity points, roughly $113 billion of it lands in bonds proper. On Sway’s wider $5.2 trillion base the equity sale is $146 billion. Both figures use the year-end 2025 asset count, so both understate the pool that exists today.
Now hold that against supply. Wall Street desks put AI-related investment-grade issuance near $300 billion for 2026, and near $360 billion once converted into 10-year duration equivalents, because data centers, substations and power contracts amortize over decades and the bonds are written to match. The target-date bid of $113 billion covers 38% of the dollars. Multiply it by the Aggregate index duration of 5.5 years and it covers $62 billion of 10-year equivalents, or 17% of the duration.
The composition gap runs the same direction. Technology was slightly more than a third of US equity market capitalization at the end of 2025, but technology bonds were under 10% of US corporate supply, on Vanguard’s own reading of Bloomberg data. So the rebalance sells an index that is one-third AI and buys an index that is one-tenth AI, at a fraction of the maturity the issuers need. The trade is a systematic seller of the AI equity trade and only an incidental buyer of the AI credit trade.
This is why the belly of the curve keeps finding sponsorship while the long end keeps clearing wide. There is a very large automatic buyer. It buys five and a half years.
The Business Model Angle
A regulatory default is the cheapest distribution channel ever built. The asset-weighted target-date expense ratio is 0.27%, down from 0.67% in 2008, which sounds like brutal fee compression until you multiply it by $4.855 trillion and get roughly $13 billion a year in management fees. Customer acquisition cost is a line in a plan document. Churn among pure holders is 1%. Three-quarters of these funds hold underlying funds from their own firm, so the sponsor collects at the wrapper and again at the component. Fee compression is survivable when the shelf space is statutory.
Contrarian by construction beats contrarian by conviction. BMA’s Camp Kotok piece documented professional allocators who privately called AI the most likely credit event and stayed long anyway, because underweighting the benchmark is financed out of their own fee stream. Target-date funds sold the AI equity trade every month this year and nobody had to defend it in a client meeting. When a discipline is expensive to hold, the durable version of it is the one nobody is asked to hold.
The unbuilt product is duration for retirement money. The issuers need a 30-year buyer. The largest pool of long-dated liabilities in the country is currently benchmarked to a 5.5-year index because that index is what the glide path specifies. Insurers and pension funds already sell duration-matched products at real spreads. Whoever brings a genuinely long-duration sleeve inside the qualified default, and gets it past a fiduciary committee, is selling into demand that already exists on both sides.
The Risk
The single largest weakness in the arithmetic is that funds may not complete the trade. Parker, Schoar and Sun, writing in the Journal of Finance, found target-date funds rebalance back to target within one or two months, which supports the flow. The same work also found the fixed-income leg of that rebalancing is weaker and slower than the equity leg, and that funds use cash and derivatives to manage it. Treat $113 billion as an upper bound on bonds actually bought, not a settled figure.
Second, this is a flow, not a trade. Target-date funds take continuous contributions, and much of the rebalancing is done by directing new money rather than by selling anything. Spread across eight months against a global bond market the OECD sizes near $109 trillion, with governments and companies expected to borrow a record $29 trillion in 2026, the target-date bid is meaningful but not decisive. It is one buyer among many, and bond funds have taken large discretionary inflows this year on top of it.
Third, the proxies are rough. The S&P stands in for an equity sleeve that is 29 points international out of 71, and the year-end 2025 base ignores 2026 growth. If AI-related supply lands closer to the $400 billion Vanguard’s own strategists have floated rather than $300 billion, dollar coverage falls to 28% and the gap widens rather than narrows.
The strongest counterargument is that none of this is a problem. Parker and his co-authors concluded that target-date rebalancing dampens volatility and stabilizes markets, which is a good outcome. Rising yields mean the automatic buyer is accumulating at better prices every month, exactly as intended. The mismatch described here is a mismatch between what the product promises a saver and what the Treasury market happens to need this quarter. Those were never supposed to be the same thing.
Quick Questions
Does this mean my retirement account is funding AI data centers? Partly, and passively, through the index. The sharper point is directional: the same account is selling AI equities every month it outperforms.
Should savers do anything? The article that prompted this one had it right. Nothing here argues for action by an individual. It argues that the aggregate of individual inaction is now a market force.
Why does duration matter more than dollars? Because a bond buyer clears years of interest-rate risk, not dollars of it. Two buyers writing identical checks absorb wildly different amounts of supply if one buys five-year paper and the other buys thirty.
Who buys the long end, then? Pensions, insurers, sovereign holders and price-sensitive traders. Net foreign private demand for Treasuries came in near $16.6 billion in June, which is part of why the Treasury is buying back its own long bonds.
The Business Model Analyst Take
BMA argued in August that the bond market stopped pricing the Fed and started pricing duration. This is the other side of that ledger. The demand side has a giant, reliable, utterly indifferent buyer, created by a 2006 safe harbor rather than by any investment thesis, and it will keep buying through every headline about soaring yields because 99% of its holders will never look. That is an enviable position for an asset manager and a limited one for a Treasury auction.
The transferable lesson sits one level up, and it is not about bonds. When your buyer is a formula, a default setting or a procurement rule, you get volume you never have to sell for, and you lose the ability to change what they buy. The glide path does not read the news. It does not care that the paper on offer runs thirty years instead of five. It will show up next month with the same check, in the same maturity, no matter what the market needs.
Every business with a captive channel eventually meets this. Demand that arrives automatically is demand you cannot redirect. Find out which of your revenue nobody is choosing, and then ask what happens the day you need those customers to buy something slightly different.
