The Fed Hiked. The Long End Still Climbed.

The Federal Reserve moved its policy rate up a quarter point on September 16, to a range of 3.75% to 4%. Six trading days later, the 10-year Treasury yield had climbed from 5.01% to 5.18%, its highest level since 2007. The real yield on the 10-year inflation-protected note rose from 2.68% to 2.85%, while the 30-year bond reached 5.47%. The Fed tightened. The long end did the opposite of what a textbook transmission mechanism promises.

At the press conference, Neil Irwin of Axios asked Kevin Warsh what the bond market was telling the Fed. Warsh called such moves “overdetermined”, listing a stronger economy, geopolitics and a scramble for capital in which “the so-called hyperscalers are out in the market raising funding”. The funding explanation matters most. The buildout of data centers, chips and power has moved from an equity story into the price of every new dollar borrowed.

The long end is not ignoring the Fed. It is repricing the financing regime beside it. Growth is running hot, federal borrowing is far larger than before the financial crisis, and the buyers who once took duration without haggling are less reliable. The first two forces are visible in the data; the third is real but less decisive than its neatest version suggests. That is why the rate hike failed to tame long yields, and why the AI boom now has a bond-market bill.

The calm is real

Begin with what the bond bulls and fiscal hawks share: this is not a market in panic. The September 10 sale of 30-year bonds drew bids for 2.61 times the amount offered. Dealers were left holding 2.2%, the smallest share in the 14 years of records TSCS could pull. The MOVE index, the options market’s price for insurance against bond swings, ended that week at 81, against 135 the last time the 10-year touched 5%, in October 2023.

Matthew C. Klein called a 10-year below 5% while nominal incomes accelerate “the opposite of a disorderly market”. Michael Howell saw the same calm in the premium investors pay to hold Treasuries rather than comparable AAA bonds. That premium firmed rather than faded, while term premia, the extra yield for tying money up in long bonds, stayed broadly steady. A market that is functioning is not the same as a market that is cheap.

The composition of the move also points away from an inflation panic. From the late-February low, the 10-year climbed 104 basis points: 96 in real yield and 8 in expected inflation. The market’s guess at a decade of inflation barely moved, from 2.25% to 2.33%, even as Brent traded above $100 and consumer prices rose 3.4%. TSCS warns that thin trading in inflation-protected bonds makes the split less precise; a Fed staff model that strips out that illiquidity puts it closer to 70/30 than 90/10.

Nor can the Fed’s short end pin the long end down. A quarter point on the funds rate is small beside federal receipts and mostly reaches households as interest income they spend. Treasury buybacks of long bonds, $4 billion and then $6 billion, held yields down for one day, Steve Eisman points out. Howell’s image is a beach ball pushed under water: pressure forced out of one part of the curve surfaces somewhere else.

Three prices for 5%

The first price is growth. Michael Howell estimates nominal output is expanding at roughly 7% to 8% and says a 10-year yield near 5% remains about two percentage points too low. Since late June, his decomposition puts 29 basis points of the rise in term premia and 51 in expected policy rates. Matthew C. Klein reaches a similar conclusion from prices: long yields look low against any sensible path for growth and inflation, and look too high only if both are about to slow sharply. Kris, who writes Potential Multibaggers, adds a third-quarter growth estimate of 5.1% to inflation and gets nominal growth above 8%, comfortably above the Fed’s new rate. Potential Multibaggers makes the arithmetic explicit.

The second price is debt. Federal debt held by the public rose from 35% of GDP in 2005 to 99%, while the deficit reached $1.97 trillion through August. George Noble, speaking with Steve Eisman, starts from roughly $40 trillion of federal debt and concludes that no Treasury secretary can talk rates down against that weight. Eisman likens the effort to hunting a whale with a BB gun. Stephen Clapham’s portfolio answer is close to zero exposure to Treasuries, except for long bonds issued at pandemic-era coupons that now trade near half their face value.

The third price is a missing buyer. Michael W. Green says central banks running quantitative easing and Japanese life insurers, whose home market paid nothing, once absorbed long bonds without asking the price. The marginal buyer now borrows, finances positions and cuts risk when volatility rises, so it charges more to warehouse duration. In that telling, the term premium is the fee set by whoever must hold the supply left after Japan, the Fed and index funds take their share.

There is evidence for that story, but it has been tested. A Tokyo insurer earns about 2% on a 5% Treasury hedged back to yen for one year and about 3% for a full decade, roughly what a Japanese 10-year pays at home. TSCS found Japan’s $135 billion fall in Treasury holdings between February and July was mostly bills and falling prices. American insurers increased notes and bonds from $119 billion at the end of 2024 to $152 billion by mid-2026; the Fed has reinvested maturing Treasuries since December. The cleaner conclusion is less selling, and not much buying either.

The fiscal explanation has its own problem. Howell measures the supply-linked term premium at minus 27 basis points. If investors were plainly charging for deficit risk, that is where the charge should show. TSCS says the Fed staff’s three models disagree. The evidence is split.

AI meets the rate

Warsh’s funding point brings the bond move into the Buildout. Howell says an economy running hot on Treasury financing and the AI capital-spending boom was pushing the 10-year toward 6%, a pace last seen in the mid-1980s. Deficits near 6% of GDP and extraordinary AI spending are increasingly financed through short-term debt rather than internal cash, building a refinancing problem. The risks are a sharp peak in capital spending in 2027 and energy costs feeding inflation.

JunkBondInvestor shows who is already paying. Two days after the Fed moved, CleanSpark, a bitcoin miner building a data center for Meta, sold $2.3 billion of five-year notes at 8.25% and drew about $10 billion in orders. CoreWeave priced a convertible at a 2.875% coupon, up from 1.75% in April, and made the deal bigger. These borrowers chose to pay more and kept building. The borrower with no choice is the one on a floating-rate loan: in the second-quarter sample of public loan issuers, 22% already had cash-flow coverage below 1.5 times. JunkBondInvestor’s credit account separates a higher hurdle from an immediate stop.

Equity markets are less forgiving than the bond syndicate. Ren, who remains fully invested in the AI trade, says a 10-year yield near 5% is the hurdle every long-duration growth company must clear. Daniel Romero says Treasuries above 5% weigh on financing and stocks more heavily than at any point in 15 years, pointing to a semiconductor index fall of 30% while oil rallied 35%. Apollo, relayed by DeepValue Capital, assumes hyperscaler operating cash flow triples from $600 billion to $2 trillion, with OpenAI and Anthropic accounting for more than half of a $2.3 trillion backlog. The semiconductor comparison is set out here.

The price of duration is being set by the marginal buyer, not by the Fed’s policy rate.

The strongest case against this reading is that expensive money is a symptom of strength. Auctions draw bids, dealers are not warehousing paper, real yields account for most of the increase, and AI borrowers are still raising billions. If nominal growth exceeds 8%, a 5% 10-year yield may be late restraint, not an early threat. Higher financing costs can expose weak projects while leaving profitable technology spending intact. On this view, the Fed has finally stopped underestimating growth and inflation.

That case deserves more weight than a crisis narrative. Yet a financing boom can continue while its bill grows. Andrew Walker of Yet Another Value Blog notes that companies locked fixed-rate debt between 2021 and 2023, flattering today’s cash flow. It comes due between 2028 and 2031. The relief is a calendar effect, not proof that capital has ceased to matter.

What settles it

The Fed is trying to steer a price it may no longer control. TSCS describes a trap: raise rates and the central bank feeds an interest bill approaching $1 trillion a year; cut with inflation at 3.4% and it loses the long end. Howell says bill-heavy issuance holds the 10-year about 37 basis points lower. Green says some Treasury demand follows index rules, so the signal arrives filtered.

Inflation is the hinge. Klein reads supercore PCE as about two percentage points faster than before the pandemic, with more than half of components rising at least 3% a year, versus roughly a third from 2000 to 2020. DeepValue Capital reads inflation below 2% on a three-month annualized basis in August and calls oil from the closed Strait of Hormuz a supply shock no rate rise can fix. Hugh Vuillier of Variant Perception says trimmed-mean PCE is still trending lower. Variant Perception’s account shows why the Fed can look late to one analyst and reckless to another.

The Federal Reserve’s next test is October 28. As of September 23, futures priced about a 70% chance of a second hike. Another increase supports Klein’s view that officials are catching up. No increase strengthens Jim Paulson’s case that the first move relied on backward-looking data while slow hiring, the lowest core CPI since 2021 and soft retail sales pointed elsewhere. The Fed’s September statement records the move; the press-conference transcript records the reasons.

The market has more immediate tripwires. Howell turned buyer when nominal yields rose above 5% and real yields neared 3%, calling it the best entry point in years and seeing 10-year Treasuries returning 10% to 15% over 12 months. He says 5.5% may be roughly 50 basis points from a major peak. Katsenelson would still not buy at 6%. TSCS’s kill switch is a 3.15% real 10-year yield. It stood at 2.85% on September 24. FRED’s Treasury series keeps the levels honest.

Foreign holdings, the term premium and payrolls will decide which story survives. Treasury data due in mid-October will show whether Japan resumes buying despite the poor hedge. Green says a term premium rising toward its historical 1.2% to 1.4% range would point to normalization rather than a buyer shortage; Howell says it is falling. The 10-year/5-year spread narrowed from about 45 basis points at the start of the year to about 15, the shape of a late cycle. $32 billion of leveraged loans comes due before end-2027, with a broader refinancing wave in 2028-31. A negative payrolls print beside an inverted curve would, in DeepValue Capital’s reading, put recession six to 24 months out.

We would be wrong if long yields fell without a slowdown in growth, inflation or AI capital spending, proving that buyer demand and Fed control mattered more than the financing arithmetic.

The AI buildout has treated capital as a quiet input. The bond market has started charging for the noise.