Why Are Global Bond Yields Rising?

Stephen Kirchner recently surveyed a number of explanations for the sharp rise in US bond yields. His account draws partly on published commentary and partly on private correspondence, including an argument from macroeconomist Dimitris Valatsas. Kirchner starts with five contending theories.

• US fiscal deficits / Treasury supply: large government borrowing increases the supply of Treasuries and the demand for global saving.

• Foreign abandonment of Treasuries: foreign investors, particularly official holders, may be reducing their willingness to hold US government debt.

• The AI investment boom: large AI-related capital expenditure creates additional private demand for financing and competes for global saving.

• Fed/Treasury communication and policy expectations: changing expectations about monetary policy and official policy signals may have pushed longer-term yields higher.

• A higher r*: the return on technology capital has risen substantially, increasing the demand for capital and pushing up the natural rate of interest, r*.

Kirchner notes that Robin Brooks is sceptical that foreign investors are abandoning Treasuries and equally sceptical that the AI investment boom is responsible. Brooks emphasises the US fiscal deficit instead, arguing that government dissaving is the principal source of pressure on yields. I think Brooks dismisses the AI explanation too readily. I find Valatsas’s argument regarding r* persuasive, up to a point: where I part company is confidence rather than mechanism.

Kirchner is focused on the recent rise in US Treasury yields. I want to start with a broader question. How much of the movement is American and how much is part of a global repricing? Bond yields have risen substantially in Australia, the UK, Germany and Japan as well as the United States. That matters for identification. An explanation specific to the United States may explain the additional rise in Treasury yields, but by itself it cannot explain a common movement across global bond markets. The starting point here is therefore the global component, before turning to what is specifically American.

And there is something missing from the explanations of that global movement. Today’s yields are being compared, implicitly or explicitly, with a post-GFC world in which central banks became enormous, price-insensitive buyers of government bonds. That source of demand has largely disappeared. Governments remain large borrowers, a major new private investment cycle has emerged, and the global environment has become substantially more uncertain.

The global repricing

The above chart makes the identification point clear. Ten-year yields are now around 5.2 per cent in the United States, 5.4 per cent in Australia and 5.5 per cent in the UK. Germany, where yields were negative for much of the QE period, is around 3.5 per cent, while Japan has moved from yields around zero to more than 3 per cent. Whatever is driving the additional recent move in Treasuries, the larger repricing is plainly not confined to the United States.

That observation provides a useful first test of the competing explanations. A specific foreign abandonment of Treasuries does not naturally explain a rise in yields across other major sovereign bond markets. US fiscal deficits are more relevant because Treasury borrowing competes for global saving and can therefore affect yields elsewhere, but the global movement makes it harder to attribute the broader repricing principally to US fiscal policy. The same qualification applies to Fed and Treasury communication. US monetary-policy expectations can spill into global bond markets, but an explanation based on US policy needs to explain that international transmission rather than simply the movement in Treasuries. The global evidence does not rule out forces originating in the United States, but it raises the bar for explanations that are specifically American.

Much of the initial rise from the lows of 2020 and 2021 is straightforward. Inflation returned and central banks tightened monetary policy. Expected future short rates rose, pulling longer-term yields with them. Japan followed a different timetable because the Bank of Japan retained negative rates and yield-curve control long after other central banks had begun tightening. Fiscal policy also became more expansionary across many advanced economies, increasing government demands on global saving.

The global environment has since become considerably more uncertain. The Iran war and the broader instability in the Middle East have increased uncertainty around oil and refined-product supplies, shipping and the persistence of inflation. More generally, war, trade restrictions and geopolitical fragmentation have made the outlook for inflation and monetary policy less predictable. The effect on sovereign yields is not mechanical. A conventional flight to safety can lower them. But an energy shock can raise the expected path of real policy rates if central banks are expected to maintain tighter policy to contain its inflationary consequences. Greater uncertainty about the future path of inflation and short rates can also increase the term premium investors require to hold long-duration nominal bonds. Neither mechanism requires a rise in measured inflation compensation.

But that still leaves the question of where the price of capital settles once the initial inflation shock and monetary normalisation have passed.

The missing buyer

For more than a decade, the world’s major central banks were extraordinary, price-insensitive buyers of government bonds. QE removed duration from private portfolios and was explicitly intended to lower longer-term interest rates. Whatever was happening to the underlying balance between saving and investment, private investors did not have to absorb the full supply of sovereign bonds because central banks were continually adding to their holdings.

My suspicion is that QE explains a substantial part of why global bond yields became so extraordinarily low. Something remarkable happened to global bond yields from the early 2000s through the post-GFC period. The global saving glut was already pushing yields lower before the crisis, while demographics, weaker potential growth and greater demand for safe assets also contributed. But these seem to me partial explanations for the extraordinary levels ultimately reached. The most plausible ringleader is QE. The major central banks became enormous, price-insensitive buyers of sovereign bonds, deliberately removing duration from private portfolios for the express purpose of lowering long-term yields. The direction of the effect is not controversial, and the empirical literature generally finds that QE lowered yields. The difficult question is how much. We cannot observe the counterfactual path of bond yields without QE, so it would be too strong to attribute the entire decline to central-bank purchases. But I find it equally difficult to believe that an intervention of that scale was merely incidental to the extraordinarily low yields observed while it was in place.

The important change today is not quantitative tightening itself. QT returned some duration to private markets, but that was the mechanical unwinding of part of what came before, and in the United States the Federal Reserve has already ended balance-sheet runoff. The more fundamental change is the cessation of QE. The global bond market once again has to clear without central banks continually adding large, price-insensitive purchases of government securities. If QE pushed long-term yields materially below where they would otherwise have been, some of the subsequent rise need not represent a new force pushing interest rates above normal. It may instead reflect the disappearance of a force that had previously pushed them below it. In that sense, part of the rise in yields may simply be a return towards normal.

There is also a potential problem here for estimates of r*. The natural rate is not observed. It is inferred from the behaviour of output, inflation and interest rates using models with considerable uncertainty. If QE persistently compressed market interest rates, it is possible that some empirical estimates interpreted part of that compression as evidence that the underlying equilibrium rate had fallen. That does not mean the estimated decline in r* was simply an artefact of QE. But it does make me wary of treating the extraordinarily low market rates of the 2010s as a clean observation of the natural price of capital.

The end of QE by itself does not explain why yields should continue rising. It tells us why one source of downward pressure has disappeared. What makes the present environment more interesting is what has happened to the demand for capital at the same time.

The AI investment boom

Brooks dismisses AI as an important explanation for higher yields. Part of his argument is that today’s AI buildout is less capital-intensive than the internet boom, which required the physical construction of communications networks. More importantly, he points to the US saving-investment accounts. During the dot-com boom, non-financial corporations became net borrowers. Today they remain net savers, while government dissaving accounts for the overwhelming part of the US external financing requirement. On that basis, he concludes that the fiscal deficit, rather than AI investment, is responsible for the pressure on yields.

I don’t think the accounting establishes that conclusion. A corporate sector can remain a net saver while substantially increasing investment if corporate saving and profits are also high. Sectoral balances describe the net positions that emerge at prevailing prices. They do not identify the counterfactual equilibrium interest rate in the absence of the increase in investment. The question is not whether AI investment explains the US current-account deficit. It is whether the investment boom has increased the demand for capital and therefore affected the price at which global saving and investment are brought into balance.

The scale of the financing is difficult to dismiss. Morgan Stanley estimates that global AI-related debt issuance reached roughly US$466 billion in the first nine months of 2026. That is only the debt component. The buildout is also being financed through equity, private credit, project finance, bank lending and internal cash flow. Whatever one thinks of the eventual productivity payoff from AI, building it requires semiconductors, servers, data centres, transmission capacity, power generation and associated infrastructure. This is not a capital-light development.

The microeconomics is straightforward. Firms perceive a new set of investment opportunities with unusually high expected returns, shifting desired investment outward. It is a gold rush. The discovery of what may be an enormously valuable new technology has triggered a race to build the infrastructure needed to exploit it. Some investments will produce extraordinary returns. Others will prove to have been mistakes. That does not matter for the present argument because investment decisions depend on expected returns, not the returns subsequently realised. What matters for the bond market today is that firms and their financiers are willing to commit enormous amounts of capital because they believe the returns are there.

The price of capital then does the rationing. The supply of saving is not infinitely elastic, so the increased demand for investment funds puts upward pressure on interest rates. Higher prospective returns attract additional saving and alter portfolio allocation, while higher financing costs make marginal projects uneconomic. Projects that make sense at a 4 per cent cost of capital may not make sense at 7 per cent. Morgan Stanley reported a sharp slowdown in AI-related issuance in September as rising rates forced some project financing to be renegotiated, describing it as a pause rather than a retreat. That is not evidence against the investment-demand explanation. If projects are being renegotiated or delayed as financing costs rise, the higher price of capital is already constraining demand.

The financing is also distributed across the risk spectrum. Some of the world’s strongest technology companies can issue debt offering a meaningful yield premium over Treasuries for relatively modest additional credit risk. Other AI-related investment is financed through project debt, private credit or weaker borrowers offering much higher prospective returns. The important change is not the banal observation that corporate debt yields more than government debt. It is the scale of the additional claims now being made on capital. Nor does any of this require foreign investors to abandon the United States. Treasury issuance and private investment simply have to be financed simultaneously at prices that clear the market.

Valatsas and r*

Valatsas takes the investment argument one step further. As reported by Kirchner from their private correspondence, his argument is that the expected marginal return on technology capital has risen substantially. If so, desired investment should rise with it. If the shift is sufficiently large and persistent relative to global saving, the equilibrium real interest rate should also rise. Rising technology equity values and rising bond yields need not therefore be contradictory signals. They can both reflect unusually high expected returns to new capital.

The mechanism is persuasive. Indeed, I suspect r* is rising. My reservation is with how confidently we can identify the resulting movement in r*. The natural rate is a latent variable that is difficult to estimate even with decades of data. A rise in long-term real bond yields is consistent with a higher r*, but it is also consistent with a higher expected path of policy rates or a higher real term premium. The recent US increase has been overwhelmingly a real-yield rather than an inflation-compensation move, but that still does not tell us how much, if at all, r* has changed.

The observable part of the story is stronger. Firms are behaving as though the expected return on a large class of investments has increased, and they are attempting to raise and spend very large amounts of capital in response. Standard economics gives us good reason to expect that to put upward pressure on the equilibrium real rate. This does not require AI already to have produced an economy-wide productivity revolution. Investment responds to expected future returns. Firms can therefore drive up the demand for capital today because they expect large future productivity gains even if those gains are not yet visible in aggregate productivity statistics. I am much more confident about the direction of that mechanism than I am about anyone’s ability to estimate the resulting change in r* in real time.

AI cannot explain the initial rise in global yields from their 2020 and 2021 lows. Inflation, monetary tightening and the withdrawal of extraordinary monetary accommodation came first. The dramatic AI financing boom arrived later, after the bond market was already operating without the extraordinary incremental demand created by QE. A large new source of private demand for capital therefore arrived after an exceptional source of demand for government bonds had disappeared.

The US wedge

The global test also helps separate the broader repricing from the latest US move. The latter is substantial and requires explanations that are specific to the United States.

Kirchner notes that the US ten-year Treasury yield rose 52.5 basis points between 1 September and around 9 October. Over approximately the same period, the US yield premium over Germany and Japan widened substantially. The latest move was therefore disproportionately American. But it occurred within a much longer global repricing of sovereign yields away from their post-GFC and QE-era lows.

That puts Brooks’s fiscal argument back into play without requiring it to explain the whole global movement in bond yields. The United States is running very large deficits and issuing enormous quantities of debt. Treasury borrowing is large enough to affect the global demand for saving, while a change in the compensation investors require for US fiscal or policy risk can also widen Treasury yields relative to other sovereigns. US fiscal deterioration can therefore exert some global pressure while also creating a local wedge.

The United States is not the only country where that local wedge may be changing. France and Belgium have also faced growing concerns about their fiscal trajectories, accompanied by relatively large increases in long-term yields. Nor are the global and local repricings necessarily independent. A common rise in the price of capital can coincide with a reassessment of sovereign risk, producing larger yield movements where the market deems the risks greater.

But the spreads identify the existence of a US-specific movement, not its cause. Bill Nelson has noted that the September FOMC minutes did not attribute the inter-meeting rise in Treasury yields to fiscal concerns. That does not establish that fiscal policy was irrelevant, but it qualifies the claim that the recent move is straightforwardly a fiscal event. Changes in expected Federal Reserve policy, Treasury supply, monetary-policy communication and the term premium are all capable of contributing to a US-specific increase in long yields. Brooks’s evidence on foreign holdings is also more convincing than his argument against AI. Foreign demand for Treasuries looks reasonably healthy once movements in the dollar are taken into account, making the simple foreign-abandonment explanation difficult to sustain.

Fiscal policy and AI investment should not be collapsed into the same explanation. A government deficit is dissaving. AI investment is an outward shift in desired productive investment caused by higher expected returns. Both can raise the market-clearing interest rate, and the United States currently has unusually large examples of both, but they are economically different. The AI story has the additional implication that the marginal return to capital itself may have risen, which is why it connects naturally to Valatsas’s argument about r*.

Some of the additional US move may also reflect the term premium. Global uncertainty affects all sovereign bond markets, but uncertainty about the US fiscal trajectory and the future path of Federal Reserve policy can produce an additional Treasury-specific premium. That is another reason not to equate every movement in long real yields with a movement in r*.

The price of capital

The global evidence imposes some discipline on the competing explanations. It makes a general abandonment of Treasuries an implausible explanation for the common rise in yields, and it limits what specifically American fiscal and policy developments can explain without an international transmission mechanism. Inflation and monetary tightening drove much of the initial move away from the pandemic lows. The cessation of QE removed an extraordinary source of demand for sovereign bonds. Large government borrowing is adding to the demand for global saving. AI has created a major new private demand for capital, while the Iran war and wider geopolitical uncertainty have increased uncertainty around energy prices, inflation and monetary policy. The latest widening in US spreads then tells us that there is an additional American component requiring its own explanation.

I think Brooks is too quick to dismiss the AI investment boom. The amounts being raised and invested are simply too large for their effect on the demand for capital to be assumed away. Valatsas provides a plausible interpretation of where that leads. If expected returns to investment have risen, the equilibrium real interest rate should rise as well. I suspect r* is rising, but I have much less confidence that we can identify its movement from observed bond yields. Long-term rates contain expectations of future short rates and term premia as well as information about r*. The mechanism is convincing; the measurement is not.

QE belongs in that story because it changed the starting point. For more than a decade central banks deliberately removed duration from private markets and increased demand for government bonds in order to lower long-term yields. We cannot know where yields would have settled without those purchases, but neither should we treat the resulting post-GFC yield structure as a natural benchmark. That source of demand has now largely disappeared.