For years, financial markets were conditioned to believe that central banks controlled the price of money. Today, that assumption is becoming much less comfortable. Two developments in the US bond market are telling the same story from different angles. First, Wall Street is beginning to struggle with the sheer quantity of debt required to finance the artificial-intelligence investment boom. Second, the US Treasury increasingly appears concerned that long-term government yields are becoming too high and too volatile. The connection between the two is important. The AI investment cycle is no longer being financed only through corporate cash flows. The largest technology companies, infrastructure investors, and data-centre operators are increasingly turning to debt markets. Amazon, Alphabet, Nvidia, Meta, Oracle, SpaceX and others have already raised more than USD 200 billion in dollar bonds this year, compared with just USD 13 billion over the same period last year. And there is more coming. Banks expect another USD 50–60 billion of hyperscaler debt issuance after the US Labour Day holiday, while several large data-centre financing transactions are already being prepared.

At first sight, this appears perfectly rational. AI infrastructure requires extraordinary amounts of capital. Data centres, electricity grids, memory, processors and cooling systems all require investment before the economic returns materialise. But capital markets are beginning to show signs of indigestion. Several recent technology bond issues have traded below their issue prices shortly after launch. Nvidia, Amazon and SpaceX all experienced weaker secondary-market performance, forcing investment banks to rethink how new deals are structured and distributed. BlackRock’s recent USD 12.5 billion financing for a Meta data centre illustrates the change. Banks deliberately favoured pension funds and insurance companies — investors more likely to hold the bonds over the long term — rather than short-term accounts seeking quick profits. Even then, the financing required a yield of around 7.5% to attract sufficient demand.

This is a significant development. The AI boom has so far been analysed mainly through equity markets: valuations, earnings expectations and the sustainability of capital expenditure. But increasingly, the stress may appear first in credit markets. Debt investors are asking a simpler question than equity investors: will the cash flows eventually justify the borrowing? And unlike equity investors, they do not need a spectacular upside scenario. They simply need to be compensated adequately for duration, credit and liquidity risk. The problem is that corporate issuers are competing with an enormous borrower already dominating the market: the US government.

Washington continues to finance annual fiscal deficits approaching USD 2 trillion. The result is a persistent supply of Treasury securities at exactly the same moment that technology companies are asking investors to finance hundreds of billions of dollars of new infrastructure. This creates a basic market problem. There are many borrowers. There is only one pool of capital. The consequence is that long-term yields must rise sufficiently to attract buyers. This is why the behaviour of US Treasury Secretary Scott Bessent deserves attention. Several recent decisions suggest that the Treasury is increasingly uncomfortable with the level of long-term yields. Bessent participated in the first US currency intervention to support the Japanese yen since 1998, reducing the risk that Japan would need to sell US Treasury holdings to finance its own currency intervention. He has also promoted greater use of the Federal Reserve’s facility allowing foreign central banks to obtain dollars against Treasury collateral rather than selling those securities outright. The objective is understandable. Japan remains one of the world’s largest foreign holders of US government debt. If defending the yen required large-scale Treasury sales, the resulting upward pressure on American yields could become uncomfortable very quickly. The Treasury has also subtly changed its language regarding future bond issuance. Previously, officials spoke about potential future “increases” in coupon auction sizes. The latest guidance instead referred to potential “changes”, prompting investors to speculate that long-dated issuance could actually be reduced. That linguistic adjustment may appear insignificant. Bond markets do not think so. Around 61% of clients surveyed by BMO now expect the next adjustment to 30-year Treasury auction sizes to be a reduction rather than an increase. This is effectively a form of debt-management intervention.

If the market does not want enough long-duration debt at acceptable yields, the Treasury can issue more short-term bills and fewer long-dated bonds. But this does not solve the underlying fiscal problem. It merely changes where the debt sits on the curve. And this is where the contradiction in current US policy becomes increasingly visible. The administration wants lower long-term interest rates. It also wants significant tax cuts, large defence spending, aggressive AI investment and an industrial policy requiring enormous amounts of private capital. At the same time, inflation remains above the Federal Reserve’s target and geopolitical shocks continue to generate upward pressure on energy costs. All of these forces increase the demand for capital. Yet the objective is somehow to reduce its price. Markets may eventually refuse to accommodate both. The recent behaviour of 30-year Treasury yields is already sending a warning. Investors increasingly require higher compensation to hold long-term US government debt because they are uncertain about future inflation, fiscal deficits and supply. The Treasury can modify auction sizes. The Federal Reserve can reduce communication. Washington can encourage foreign central banks not to sell. But none of these measures eliminates the fundamental equation. If the supply of debt rises faster than investors’ willingness to absorb it, yields must adjust.

Today, the behaviour of bond markets matters more than almost any individual Federal Reserve statement. The central bank controls overnight interest rates. It does not fully control the 10-year or 30-year cost of capital. The market does. For nearly fifteen years, that distinction barely mattered because central banks were enormous buyers of government bonds and inflation remained low. That regime is disappearing. Investors are rediscovering that long-term yields are ultimately determined by supply, inflation risk and confidence in fiscal policy. AI financing is now adding another powerful source of demand for capital to an already crowded market. The result is a world where the bond market increasingly sets the limits of economic policy. For investors, this is perhaps the most important structural shift. Governments, technology companies and infrastructure projects can all promise extraordinary future returns. But before those returns arrive, they still need somebody to lend them the money. And lenders are beginning to ask for a higher price.