BofA Just Called 5% Treasuries a 'Generational Entry.' The Mortgage and AI Capex Markets Have to Backfill Demand for That Call to Work

Bank of America Securities, in Michael Hartnett's weekly \"Flow Show\" published around September 4, 2026, declared that long-dated U.S. Treasuries — accessed through the iShares 20+ Year Treasury Bond ETF and the PIMCO 25+ Year Zero Coupon U.S. Treasury Index ETF — had fallen to their worst 10-year rolling return in a century at negative 2% annualized, framing that setup as a \"generational buying opportunity.\" The trade the note argued for is buying duration at exactly the moment BofA's own institutional clients are most underweight it.

The Flow Show also spelled out the political catalyst Bank of America is watching. A Democratic sweep of the November midterms — which would shift policy toward what Hartnett called \"populist socialism\" — would lower growth assumptions, drag long yields down, weaken the dollar through a narrower rate differential, and turn bonds into the fourth-quarter contrarian play. The trade only requires that the \"Anything But Bonds\" era that dominated 2022–2024 reverses.

The September2026 BofA Global Fund Manager Survey, with fieldwork from September 4 to 10 and covering 190 panelists managing $512 billion, is the institutional verdict on whether BofA's own long-bond call will work. The numbers show the survey respondents are not buying it. Cash jumped to 3.9% of AUM from 3.5% — the largest monthly rise since March — and the broad sentiment gauge fell to 7.0 from 8.0, the least bullish reading in three months.

When the same bank whose strategists are calling for a long-bond bottom is also reporting that nearly half of the managers it surveys are underweight the asset class, the call is by construction contrarian. The setup resembles the 2009 vintage: maximum pessimism on rolling returns, crowded shorts, and an institution willing to put a bullish label on a trade that almost no surveyed participant is positioned for.

Two structural drags sit on top of the long bond. The first is issuance. The U.S. Treasury on August 19, 2026 announced it was increasing the maximum size of its long-end liquidity support buyback operations on 10-year-to-20-year and 20-year-to-30-year nominal coupon securities to at least $4 billion per operation from the prior $2 billion cap, effective September 9, 2026. On September 9 the Treasury also announced an additional buyback operation of up to $6 billion for the following day — a roughly three-times-normal operation. Hartnett had already characterized the doubling of buybacks as \"quasi\" quantitative easing, conditional on whether Treasury Secretary Scott Bessent can drag the 30-year yield below 5%.

The second drag is foreign demand softening. TIC data for June 2026 showed foreign holdings of U.S. Treasuries at $9.29 trillion, with Japan cutting its stash to $1.116 trillion (down 2.3% month-over-month), the UK at $940 billion (down 19.4% from a 2025 peak), and mainland China at $633 billion. When the three largest official holders are all trimming at once, marginal demand for incremental long-end supply has to come from price-sensitive domestic buyers — pension funds, insurance balance sheets, and household flows into Treasury funds — exactly the channels whose allocations respond to the very yield levels BofA is now arguing are too high.

The single largest non-government source of long-dated U.S. dollar credit creation in 2026 is AI capex financing by hyperscalers. Vanguard tallied the five largest hyperscalers — Amazon, Alphabet, Meta Platforms, Microsoft, and Oracle — at roughly $35 billion of annual bond issuance on average from 2020 through 2024. That pace tripled to $93 billion in 2025. By July 31, 2026, the same five had already issued $132 billion of bonds, including a single $53 billion multi-tranche deal and a rare 100-year maturity bond.

The volume matters because AI debt absorbs duration. Hyperscaler capex is approaching $800 billion in 2026 and is forecast to exceed $1 trillion annually from 2027 through 2030, with a substantial share of data-center financing routed through private credit and off-balance-sheet structures that do not show up cleanly in public bond indexes. Every additional $100 billion of long-dated corporate credit competes with U.S. Treasury issuance for the same institutional buyers. That is the structural backdrop Bank of America's \"generational entry\" call has to push against — and the reason the FMS's 42% reading on AI capex as the most likely source of a systemic credit event is the most-watched variable for anyone sizing the long bond today.

The mortgage market is the cleanest transmission test of Bank of America's thesis. The 30-year fixed mortgage rate finished the week of September 10, 2026 at 6.79% — the highest weekly print since July 2025 — and sits just 1.98 percentage points above the 10-year Treasury yield. That spread has been stuck near two percentage points since September 2022, and the only way it widens meaningfully is for Treasury yields to fall faster than mortgage rates do. A 50 basis-point rally in the 10-year would, holding the spread constant, put the 30-year mortgage in the 6.3% area — the threshold housing economists have flagged as the level at which affordability pressure on first-time buyers begins to ease.

Equity duration is the second leg. The iShares 20+ Year Treasury Bond ETF is already showing what a duration rally would deliver — the fund closed September 12, 2026 with a 30-day SEC yield of 5.29% and total assets of $47 billion, meaning each50 basis points of yield compression translates to roughly $5 of price gain per $100 of face. If BofA's call is right and the long bond re-rates, the same discount-rate shift lifts the present value of long-duration cash flows at Microsoft, Alphabet, Meta Platforms, Amazon, and NVIDIA. The reverse is also true: if yields stay pinned at 5% because the AI debt wave absorbs duration, the equity-duration benefit never arrives, and the mortgage market keeps the existing6.79% handle into year-end.