Norway’s $106 billion hint

In their recent letter to Norway’s Ministry of Finance, Norges Bank (the manager of Norway’s $2.3tn sovereign wealth fund) wrote: “High government debt has gone from being a distinctive feature of individual countries, particularly Japan and some euro area countries, to becoming a more general characteristic of developed economies.” In the letter, they proposed reducing the weight of government bonds from 70% to 50% of their bond holdings to chase better returns elsewhere. This cuts roughly $106 billion from global government bond allocations; in particular, the US Treasury weight falls from 34% to 22%. Although the proposed cut is small relative to the size of global fixed-income markets, it “offers powerful signals as to the slow-moving but persistent shifts in global capital flows and portfolio allocation currently taking place in markets”, write Shreyas Gopal and George Saravelos at Deutsche Bank. As governments of major economies are beginning to face doubt about their debt, new competitors are targeting the same pool of investors’ money. 

Per Emily Herbert of the Financial Times, investors have lent AI hyperscalers $200 billion this year, and this could reach $1 trillion by 2030. JPMorgan believes the five hyperscalers (Alphabet, Amazon, Meta, Microsoft and Oracle) could approach 10% of America’s investment-grade corporate bond market. Herbert notes that the uncertainty over the volume of upcoming debt is driving up the premiums demanded by investors, though it may not deter hyperscalers. Torsten Slok, chief economist at Apollo, claims that the AI boom has become “rate-insensitive”. Recent rate hikes are slowing purchases such as homes and cars, as textbook economics predicts, yet orders for business equipment are growing at their fastest pace since 2021. The consensus forecast for US capex growth has also been raised from 3.1% to 3.7% over the past year, despite rising rates. Slok argues that these counter-textbook behaviours are due to AI, and that rate hikes are not slowing the economy or inflation as intended. 

This is where the two immovable forces of AI debt and government debt collide. While hyperscalers continue to borrow rate-insensitively, governments across the world are running budget deficits that they also finance by borrowing. This means governments cannot stop competing for capital when borrowing rates rise. The effects are currently showing up at the margin. Bankers told the FT’s Katie Martin that nobody is selling government bonds to buy hyperscaler debt; rather, investors who once put all their new money into government bonds might now “just put 80(%) of it in government bonds and 20(%) of it into this other pot instead.” That shift matters most for long-dated debt, where hyperscalers typically borrow. Consequently, new long-dated government bonds may have to offer higher yields to attract buyers. Without a plan for balancing the budget, this will only add to the more than $1 trillion a year the US already spends on interest, according to the CBO. 

Dr Roald Versteeg, Principal Teaching Fellow in the Department of Finance at Imperial College, notes that US government debt is as high as it has ever been in US history, with no plan to bring the budget back into balance. This raises the risk of fiscal dominance: a situation where concerns about debt sustainability become so large that monetary policy becomes subordinate to the financing needs of the government. Dr Versteeg explains that this means there is “a real risk of inflation trending up, or of other forms of financial repression to keep real interest rates down in order to reduce the US government’s debt burden. As investors become aware of this, term premia are likely to rise, increasing the borrowing costs of governments.” Former Fed Chair Janet Yellen sounded a similar warning at the American Economic Association meeting in January. 

The current state of US government debt creates a scenario where the AI buildout has limited upside in restoring fiscal health, while presenting some concerning downside. “The sheer scale and ramp-up of borrowing behind the AI buildout is quite unlike anything we have seen before and potentially a real worry,” says Dr Versteeg. If a cascading collapse of AI companies’ debt were to begin, governments would be poorly placed to act as a backstop. “With government finances as stretched as they are, there is much less room for the state to step in than in 2008 or 2020,” he explains. Additionally, Yellen, citing a paper by Elmendorf, Hubbard and Liscow, explains that even if AI companies fuel productivity growth by 0.5 percentage points per year above the CBO’s baseline, the boost would have to be sustained for 30 years to “prevent the ratio of debt-to-GDP from spiraling upward.” 

Norges Bank’s proposal signals that even the world’s biggest investor no longer treats government bonds as an automatic safe harbour. Governments are now facing pressure from their own mounting debt and the new borrowers competing for the same pool of money. Even if AI companies deliver productivity growth, it may not solve the mounting debt issue.