The bond market isn't broken–it's simply charging for uncertainty again.
KEY TAKEAWAYS
The term premium is the additional yield investors require for holding long-term Treasury securities because the future is uncertain. It cannot be directly observed and must be estimated using economic models.
Today's increase in the term premium reflects a return toward historical norms rather than an unprecedented spike. After years of negative term premiums driven by quantitative easing, investors are once again demanding compensation for long-term risk.
Multiple forces are pushing long-term Treasury yields higher, including persistent government deficits, changing investor demand, inflation uncertainty, and policy uncertainty. These factors operate largely outside the Federal Reserve's direct control.
Massive AI infrastructure spending is now influencing Treasury markets through unprecedented corporate bond issuance. High-grade technology companies compete with Treasury securities for investor capital, lifting borrowing costs across the economy.
The recent rise in long-term yields does not necessarily indicate that the Federal Reserve has lost credibility. Instead, markets are repricing risk after an extended period in which central bank intervention suppressed normal market behavior.
MY HOT TAKES
The financial media frequently oversimplifies complex bond market concepts, leading investors to misunderstand what is actually driving interest rates. Precision matters because inaccurate narratives produce poor investment decisions.
The Federal Reserve's influence over long-term interest rates has always been more limited than commonly believed. Quantitative easing temporarily masked that reality rather than eliminating it.
Artificial intelligence is affecting the economy in ways that extend well beyond technology stocks. Capital markets increasingly connect AI investment directly to borrowing costs for households and businesses.
Investors should spend less time trying to predict every Fed meeting and more time understanding structural changes in bond markets. Long-term capital flows often matter more than short-term policy announcements.
A bond market that demands compensation for uncertainty is healthier than one distorted by extraordinary intervention. Normal risk pricing is an important feature of functioning financial markets rather than a sign of crisis.
You can quote me: "AI data centers in Virginia are helping determine mortgage rates in Indiana."
A glimpse into the future. A glimpse into the future. Here's a question I've been asking people for thirty years, and nobody has ever gotten it wrong. If your best friend asks to borrow twenty bucks until Friday, you simply hand it over. If he asks to borrow twenty bucks until 2056–well,, you start asking questions, and you start wanting something extra for the trouble. That "something extra" has a name on Wall Street. It's called the term premium–you might have heard it tossed about in the media–it's the most important number in the bond market that you cannot find on any screen, and right now it is doing something the Federal Reserve would very much prefer it stopped doing.
Let's start with what the thing actually is, because the Street has done a magnificent job of making it sound like calculus when it's really just arithmetic with a shrug attached. A Treasury yield is not a price so much as an average of every one-year decision between now and maturity. If you could see the future perfectly–if you knew precisely what the Fed was going to do every single year between now and 2056–the 30-year bond would simply yield the average of all those short-term rates strung end to end. But nobody can see the future (a fact that has never once stopped anyone on Wall Street from charging for the attempt 😉). So the market tacks something on top. That something is the term premium: the extra yield investors demand for the risk that the next three decades don't go according to the script. And note what happens as you stretch the loan out. Ten years of not knowing is uncomfortable. Thirty years of not knowing is a different animal entirely.
Now here's the part that makes it genuinely strange. You cannot look it up. There is no ticker, no quote, no bid and ask. It is a RESIDUAL–what's left over after you subtract the expected path of short rates from the yield you can actually observe. And for roughly eight years after the financial crisis, that leftover was NEGATIVE. 😯 Investors accepted less compensation than the expected path of short rates for the privilege of owning long bonds. Which sounds like a clinical diagnosis until you remember that the Fed was buying trillions of them and was not, shall we say, price sensitive. On the ten-year, the record low was around negative 1.36% in the summer of 2020. The median going back to 1961 is roughly 2.4%. Today we're somewhere around 0.7 to 0.8. So when someone tells you the term premium is high, what they actually mean is that it finally stopped being weird.
Which brings us to the measurement problem, and this is where I want you to pay attention, because it's the tell. There are three ways to count something you can't see. You can fit a statistical model to the entire Treasury curve and back out the residual–that's the Adrian-Crump-Moench model out of the New York Fed, the one Fed chairs and presidents quote from the podium, and it's the closest thing we have to an official number. You can run a different model with different assumptions–that's Kim-Wright, published by the Board of Governors, currently landing in the same neighborhood. Or you can just ask people: take the primary dealer survey's forecast for average fed funds over ten years and subtract it from the observed yield. Three methods, three answers, and they generally rise and fall together. The Fed's own researchers have put the correlation of monthly changes between the two main models at about 0.7, which is comforting on direction (they are highly correlated) but still completely useless on level. The same research notes that the models can, at times, disagree on the SIGN. Not the magnitude. The sign. Two Federal Reserve models, looking at the same curve, occasionally can't agree on whether the number is positive or negative.
And now file this one away, because it's the detail that should make you sit up. Both of those models stop at ten years. The New York Fed publishes term premium estimates from one year all the way out to ten, and then the series simply ends. There is no official thirty-year print. Not because nobody wants one–because the further out you go, the less anyone trusts the model. So the maturity where the compensation is largest, where the pain is most visible, and where every mortgage in America takes its cue is precisely the one nobody will put an official number on. Everything you read about the "term premium" is a ten-year number standing in for a thirty-year problem.
I mention this because you may have heard that the term premium just hit its highest level since 2008. I heard it too. It isn't true, and I'd rather you hear that from me than find out later. The 2008 number belongs to a different, proprietary "fair value" spread that a research shop published back in early 2025, and it has been quietly laundered into a term premium headline ever since. What's actually true is that the measured ten-year term premium is at its highest in roughly a decade–back to levels last seen around 2011 to 2014. What's ALSO true, and probably what got scrambled in the retelling, is that the 30-year Treasury yield is genuinely at its highest since 2007. Both facts are interesting. Only one of them is the one everybody is repeating.
So what moves it? Four things, and they're stacking. First, supply–deficits north of two trillion dollars, and every dollar of it has to find a home at an auction. Second, and more important, WHO is showing up to those auctions. For two decades, the long end (longer maturity Treasuries) was absorbed by what I call patient money: foreign central banks, sovereign wealth funds, pension plans buying duration for reasons that had almost nothing to do with return. Patient money doesn't haggle. Price-sensitive money haggles at every single auction, and price-sensitive money is now the marginal buyer. Third, inflation uncertainty, and note that I said "uncertainty," not inflation. The term premium doesn't compensate you for inflation being high. It compensates you for not knowing. Fourth, and this is the one the Fed can actually do something about, policy uncertainty. There's a whole body of academic work on this, and it lands in the same place every time: when the market can't figure out how the Fed will react to a given piece of data, that confusion has a price, and the price shows up in the term premium. Which is worth sitting with for a second. A committee that says nothing costs you money. A committee that says three different things costs you more. And whatever the market charges for that confusion travels straight down the pipe into what everyone else in the country pays to borrow.
And here is the piece almost nobody has connected for you, so let me do it. Ask a good bond strategist right now why term premiums are rising and you may not get "deficits" as the first answer. You'll get the hyperscalers. Hyperscalers? 🙋 The AI buildout is being financed, in enormous size, in the corporate bond market–Amazon, Meta, Microsoft and the rest issuing debt to fund data centers. In fact, just this morning there is some fresh news that Alphabet is going to the bond market to raise some $25 billion with investment grade bonds with maturities between 2 and 40 years. Here's the mechanism, and it's the important bit: there is a certain amount of interchangeability between a high-quality corporate bond and a government bond. To a big institutional buyer, a AAA-rated tech issue and a Treasury are competing for the same dollar. So when a wave of high-grade issuance hits the market, the yield on BOTH has to rise to clear it. Which means–and I want the Millers to sit down for this one–the cost of building data centers in northern Virginia is a variable in the mortgage quote they get in suburban Indiana. Not by metaphor. By arithmetic. The bond market does not have separate rooms for technology CAPEX and household credit. It has one room.
Now go back and look at that list of four drivers, and notice something about it. Deficits are set by Congress. The buyer base is set by pension actuaries in Sacramento and reserve managers in Tokyo and Riyadh. The issuance calendar for data centers is set in Redmond and Menlo Park. Exactly one of the four–policy uncertainty–sits inside the Eccles Building (that's Fed HQ), and even that one is only partly theirs. Which raises the question everybody eventually gets around to asking, usually in a more alarmed tone of voice than the facts deserve.
So is the Fed losing control of the back end? Here's my answer, and it's less dramatic than the question implies: the Fed never controlled the back end. It RENTED it, from about 2009 to 2022, using a balance sheet the size of a G7 economy. What we're watching now is the lease running out. But there's a wrinkle worth chewing on. The July FOMC held rates steady on a contentious 9-to-3 vote, the committee's posture was distinctly hawkish, and the 30-year promptly went to its highest yield since 2007 anyway. Meanwhile the market has been pricing something close to a coin flip on a rate HIKE in September (58%). Read that twice. The front end is pricing tightening, and the very long end is demanding MORE compensation regardless. Ordinarily hawkishness suppresses the term premium–a credible inflation fighter reduces the variance you're being paid to carry. Getting both at once tells you the market isn't questioning the Fed's willingness. It's questioning whether willingness is the binding constraint anymore.
Which brings me to the part I actually want you to take away from all this, and it's quieter than the headlines want it to be. The rising term premium is not a vote of no confidence in the Fed. It's a vote that the Fed is no longer the biggest variable in the equation–that deficits, buyer composition, and a corporate financing wave nobody at the Eccles Building can vote on are doing more of the work. That's not a crisis. That's a market that has resumed pricing risk after fifteen years of being told not to bother. Honestly? I'd rather have this than the alternative. A bond market that demands to be paid for uncertainty is a bond market that's–well, awake.
So back to your friend and his twenty bucks. Thirty years, no collateral, no covenants. What do you charge him on top of whatever short rates average between now and 2056? Call it a point. Maybe more. Though as we just established, the fellow at the next desk would say sixty basis points, a third model would come back negative, and not one of the three of you could prove it. On twenty dollars it works out to about twenty cents a year regardless, so go ahead and waive it. But do ask him what he needs it for, whether he's got anyone else lending to him, and whether he plans on making a habit of this. Congratulations. You've just run a Treasury auction.