Is it time for duration?

No one has wanted to own bonds, and duration has hurt. For most of the past decade, owning long-duration U.S. Treasuries has been an unusually unpleasant experience. Inflation returned, the Federal Reserve lifted short-term rates from near zero, and long-term yields repriced sharply higher. Bond prices moved in the opposite direction. These headwinds have been bracing. Ten-year Treasuries’ rolling 10-year annualized total return had fallen to roughly zero as of August 2026 — the weakest reading in data back to the 1930s.

For those of us who favor mean-reversion, a knee-jerk reaction is to conclude that long Treasuries are due for a rebound. There is some logic to the idea: after a painful adjustment, yields are much higher and prospective returns are plainly more attractive than they were when long-term government bonds yielded 1% or 2%. But, let’s look a little more closely!

The earlier lows on the chart occurred around 1959 and 1981, and those were different times. The early 1980s did mark the beginning of one of the greatest bond bull markets in history as inflation ultimately subsided and interest rates declined. The 1959 episode did not. Long-term interest rates continued rising, in fits and starts, for another two decades.

The math is better.

Long-term bonds have an important characteristic that can be easy to forget during periods of very low interest rates: the yield at which they are purchased matters enormously. That’s pretty much the truth about any bond, holding credit constant, if you hold it to maturity.

As of September 4, the 10-year Treasury yielded approximately 4.78% and the 30-year approximately 5.24%. Those yields do not protect an investor from further price declines if rates rise, but they provide substantially more ongoing income than investors received for most of the post-financial-crisis period.

We can contrast that with a decade ago; then investors buying long Treasuries were accepting very little income while assuming substantial duration risk. Today, the same duration comes with considerably more compensation. If inflation slows, economic growth slows or investors become more willing to own long-term government debt, falling yields could add capital appreciation to that income. If inflation proves stubborn or the market demands a larger premium for holding long-dated securities, however, prices could fall more.

That distinction matters if you are trying to plot a course for the future, because the chart below is backward-looking.

Chart 1

Long-Term Treasuries Have Experienced Their Weakest Decade in Modern History

U.S. 10-year Treasuries: 10-year rolling annualized total returns, 1936–August 2026

The 0.1% reading in August 2026 is below the previous troughs of 1959 and 1981. The outcomes of those two periods were very different. Source: Robert J. Shiller, monthly 10-year Treasury total returns; FRED (GS10); Belvest calculations.

The owners of the Treasury market have changed.

There is another reason that looking at the early 1980s in isolation can be misleading. For decades, a meaningful portion of Treasury demand came from foreign governments and central banks accumulating dollar reserves. China became the most visible example. At the end of 2013, mainland China held roughly $1.27 trillion of Treasury securities. By June 2026, that figure had fallen to approximately $633 billion—even as the amount of Treasury debt outstanding had increased enormously. China’s important, but not as important as it was.

But there is an important qualification. China’s decline has not meant that foreign investors as a group have abandoned Treasuries. Total foreign Treasury holdings have continued to grow as investors in Japan, the United Kingdom and numerous other jurisdictions have absorbed additional securities. At the end of 2025, foreigners still owned roughly $9.2 trillion of Treasury debt. (For Japan, see our other recent commentary piece.)

It’s probably more accurate to describe this as a shift in the composition of demand rather than the disappearance of foreign demand.

At the same time, hedge funds have become much larger participants. The Office of Financial Research estimates that hedge funds held approximately $2 trillion of cash Treasuries at the end of 2025, almost three times their holdings five years earlier and equivalent to roughly 7% of marketable Treasury debt. That sounds like a simple replacement of “patient” foreign capital with “fast” leveraged capital. It isn’t quite that simple either.

Chart 2

China Has Become a Much Smaller Treasury Holder While Hedge-Fund Participation Has Increased

China and hedge-fund Treasury holdings as a percentage of U.S. public debt, 2013 versus latest available data

Important note: The two series rely on different underlying measurement approaches, so the chart is best interpreted as illustrating the direction and magnitude of the change rather than as a precise apples-to-apples ownership comparison. Source: U.S. Treasury; Federal Reserve; The Investment Institute by UniCredit.

Hedge funds are important—but they are not simply speculating on bonds.

A significant portion of hedge-fund Treasury activity is connected to relative-value strategies, not thematic macro, and certainly not buy-and-hold.

The best-known is the Treasury “basis trade.” A hedge fund buys a cash Treasury security, finances it in the repurchase—or repo—market, and simultaneously sells a Treasury futures contract. The objective is generally not to predict the direction of interest rates but to capture a small difference between the prices of the two instruments. And, because those differences are small, funds will often apply substantial leverage.

This creates a somewhat counterintuitive situation. Hedge funds can help the Treasury market absorb enormous quantities of securities because they provide a buyer for cash bonds. At the same time, the leverage used to finance those positions can become a source of instability if repo financing becomes expensive, volatility jumps or futures margins rise sharply.

The Federal Reserve estimates that highly leveraged arbitrage strategies—including both Treasury basis trades and swap-spread trades—accounted for nearly half of hedge funds’ approximately $2.4 trillion of long Treasury exposure in September 2025.

That is a very large number, but it is also an important correction to the idea that hedge-fund Treasury ownership is “overwhelmingly” a basis trade. A substantial portion is not. The Federal Reserve nevertheless considers hedge-fund leverage a meaningful financial-stability vulnerability, and leverage remained near record levels in 2026.

So the Treasury market has not simply exchanged reliable investors for unreliable ones. It has become larger, more diverse and, in some places, more dependent upon leveraged financial intermediation.

There is also a surprisingly large new supplier of duration.

One of the more remarkable developments in fixed-income markets has little to do with Washington.

The enormous investment required to build artificial-intelligence infrastructure is increasingly being financed with debt. Five large hyperscalers, together with NVIDIA and associated special-purpose financing vehicles, issued less than roughly $50 billion annually for most of the period before 2025. Issuance rose to approximately $180 billion in 2025 and had approached $300 billion by August 2026.

J.P. Morgan has attempted to put that number into Treasury-market terms by converting the debt into 10-year Treasury-equivalent duration. On that basis, hyperscaler and NVIDIA issuance has increased from less than 10% of comparable Treasury bond issuance several years ago to approximately 30% in 2025 and nearly 70% in 2026. That is really an extraordinary change, even if the calculations are an approximation. Still, let’s describe this carefully.

Chart 3

AI-Related Borrowing Has Become a Major Source of Long-Duration Bond Supply

Hyperscaler + NVIDIA debt issuance, including selected SPVs, as a percentage of Treasury bond issuance

The comparison converts issuance into 10-year Treasury-equivalent duration. The roughly 70% figure therefore does not mean technology companies are issuing debt equal to 70% of all Treasury borrowing. It compares the incremental amount of interest-rate duration being supplied by these borrowers with a comparable measure of long-term Treasury issuance. Source: Bloomberg; J.P. Morgan Asset Management, August 31, 2026.

Competition for capital is real, but “crowding out” is harder to prove.

The distinction is important because it would be easy to take the 70% figure too far: the Treasury remains a vastly larger borrower when bills, notes, bonds and refinancing activity are considered together. The hyperscaler comparison is interesting because it isolates the long-duration portion of the market, where insurers, pension funds, asset managers and other investors make decisions about how much interest-rate and credit risk they want to own. And AI-related borrowers are now offering those investors considerably more corporate debt.

All else equal, additional supply should require investors to be compensated. Treasury bonds, corporate bonds, mortgages and other long-duration assets compete to some degree for the same pools of capital, and relative yields and credit spreads influence where that capital goes. But there is not yet much evidence of literal crowding out.

For instance, the Bank of England noted in July that the five major AI hyperscalers represented only about 3% of outstanding U.S. investment-grade debt at the end of 2025, even though they accounted for more than 15% of U.S.-dollar investment-grade issuance early in 2026. It also found that the new supply had, to that point, been readily absorbed and that there was little evidence AI borrowing was preventing governments or other companies from accessing capital markets.

That could change. The sums involved are growing quickly, and AI infrastructure increasingly relies on debt, private credit and structured financing. But a large new source of bond supply is not the same thing as proof that technology companies are already forcing Treasury yields higher.

There are two different risks to watch.

There is a final temptation to connect these stories too tightly. Yes, hedge funds use leverage to intermediate portions of the Treasury market. AI infrastructure is generating rapidly growing quantities of corporate and structured debt. Dealers and large financial institutions participate in both markets. That creates possible channels through which stress could spread.

But it does not mean an AI credit problem would automatically produce a Treasury selloff. In a conventional credit shock, investors often do exactly the opposite: corporate credit spreads widen while money moves into Treasuries, pushing government bond yields lower. A severe financial-market disruption could certainly strain dealer balance sheets, repo markets and Treasury liquidity at the same time. But that is a broader financial-stability issue, not yet evidence of a direct mechanical link between hyperscaler debt and leveraged Treasury trades.

A more conservative conclusion is simply that there is considerably more duration for global investors to absorb, while an important part of Treasury-market intermediation now relies on leveraged investors whose behavior can change quickly during periods of stress.

It doesn’t need to be 1981 all over again!

One must be really cautious about using any of this for investment positioning.

On one hand, long Treasuries have suffered enough that their starting yields are substantially more attractive than they have been for much of the past 15 years. That alone improves their potential role in diversified portfolios. They can generate meaningful income and, in a disinflationary slowdown or recession, could once again provide capital appreciation when risk assets are struggling. But there are still those headwinds.

We’ve highlighted the fiscal story: Federal borrowing requirements are large. The Treasury buyer base has become more price-sensitive. Hedge-fund leverage adds another potential source of volatility. AI investment is creating a formidable new supply of long-dated corporate paper. And, most importantly, inflation and the term premium can still move against us.

Getting back to my earlier admonition: be careful about extrapolating mean reversion, but for investors who have spent the past several years minimizing duration, the calculation may be different. At yields above 5% at the long end, investors are finally being paid meaningfully to assume interest-rate risk. Whether that compensation is sufficient will depend on inflation, fiscal policy, economic growth and each portfolio’s particular objectives.

Perhaps that is the real message in these charts? The long Treasury market does not look like 1981, and it does not have to. After one of the worst periods for long bonds in modern history, the price of owning duration has changed. Long Treasuries are neither an obvious bargain nor an obvious trap. They do deserve our attention.

Important disclosures