.png)
Long-end government bond yields are at levels most investors in this market have never worked through. The US 30-year briefly cleared 5.33% this week, a 19-year high, and last week’s auction stopped at 5.216%, making it the most expensive 30-year money the Treasury has raised since 2001. France is at post-2008 highs, Germany at 2011 levels, 30-year gilts approaching 6%, and 30-year JGBs close to their all-time high.
So what is actually going on? Three explanations are well rehearsed. The fourth, in our view, is the one that matters.
The three familiar answers
Energy and the war. Brent is around $90, up roughly a quarter from pre-conflict levels. With Hormuz transits still constrained and the EIA assuming Middle East production does not normalise until early 2027, the inflation impulse is persistent enough to keep inflation expectations elevated.
A Fed being tested. Fed funds have been held at 3.50–3.75% for five meetings, the June dots put the 2026 median at 3.8% with nine of eighteen pencilling in a hike, and the new Chair does not intend to guide markets. Less guidance means more rate volatility, which is itself a term-premium tax. At 4.18%, the 2-year is not pricing only a rate hike cycle, it is also pricing the uncertainty around a reaction function it does not yet know.
Elevated nominal growth. This deserves less handwringing than it gets. S&P 500 revenues grew 15% in Q2 and consensus has 2026 earnings up 30%. If nominal GDP is running hot, nominal yields belong higher. A 5% long bond alongside double-digit nominal growth is arithmetic, not distress. and is compatible with equities performing well.
None of that explains the shape of the move. Yields have refused to fall on softer data, and the curve has steepened from the long end rather than the short end and long dated breakevens have stayed anchored: the repricing has come almost entirely through real yields. For the time being, this is neither a structural inflation scare nor a long-term credibility problem.
Rates are increasingly a supply problem, not a central bank discipline problem.
Two claimants, one pool of savings
The world is being asked to absorb two structural waves of long-duration paper at once, and they compete with each other.
The first is sovereign. OECD central governments issued around $17tn in 2025 and are on track for roughly $18tn this year. US interest expense has reached $1.17tn ten months into fiscal 2026, up 15% year-on-year, against an average rate on outstanding debt of 3.44% — every refinancing is a step up. The refinancing wall is also unusually large: around $6.1tn of US government debt is scheduled to mature in 2027–28 and will need to be rolled over, much of it at higher rates. Treasury has so far held coupon sizes flat and pushed the burden into bills, but TBAC estimates that, if current auction/coupon sizes remain unchanged, it could still face roughly a $1.45tn financing gap over FY2027–28. The maturity wall is the gross refinancing challenge; the financing gap is the additional supply that existing auction sizes cannot absorb. Deferred duration supply, not absent duration supply.
The second is the AI build-out, and it is new. Goldman Sachs’ credit team counts roughly $489bn of AI-related supply so far this year, already well beyond their full-year 2025 estimate of $322bn. The five largest hyperscalers issued around $159bn of US corporate bonds in the first five months alone, against a 2020–24 average of $28bn a year. Goldman Sachs Asset Management puts hyperscaler issuance at 16–23% of gross year-to-date supply across the USD investment-grade, high-yield and leveraged loan markets. Big-five capex is tracking above $600bn, increasingly debt-funded as free cash flow stops covering it.
This is long-duration paper by design: the assets are long-lived and their economics do not vary with the cycle, so borrowers want fixed-rate liabilities far out the curve. The Dallas Fed has made the point explicitly. AI issuance is now a material source of duration supply, and it partly substitutes for, or crowds out, other investment-grade borrowers.
And the visible bond market is the smaller half of the story: private credit to AI borrowers has gone from near zero to over $200bn, and the big five carry some $662bn of data-centre leases not yet commenced, off balance sheet. The claim on global savings is larger than the league tables show.
Look across the 2026 rally and it appears well diversified: equities at record highs, credit spreads through the floor, volatility subdued, gold firm above $4,400, EM carry performing. Different assets, different geographies, different risk factors.
They are not different trades. They are one trade expressed five ways, each a claim that the long-run real cost of capital is going to fall from here.
The market is trading the destination — lower real rates — and not the starting point, which is higher real rates.
The numbers make the point uncomfortable. US high yield entered August around 271–281bp, in the richest decile of its history against a long-run median near 450bp; investment grade sits at 81bp. The S&P 500 printed a record 7,815 in mid-August on a forward multiple of 20x, with VIX around 15. Against all of that, July’s 10-year TIPS auction cleared at 2.438%, the highest at auction since October 2008, and the 30-year real yield is near 2.9%.
The bond market is pricing a materially higher equilibrium cost of capital than the equity and credit markets are. One of them is wrong.
