Photo: Enguerrand Artaz © LFDE

By Enguerrand Artaz, Strategist, La Financière de l’Échiquier (LFDE)

France’s OAT[1] above 4% for the first time since 2009, the 30-year US yield at its highest level since 2007, the Japanese 10-year approaching the 3% threshold… and equity markets beginning to show signs of fragility as interest rates surge almost everywhere in the world. The lack of progress on the Iran conflict, after Donald Trump announced he would not extend the ceasefire that expired on 17 August, certainly acted as a short-term catalyst, but the drivers of this move run deeper.

Beyond inflation fears linked to Iran, the bond market in fact stands at the crossroads of issues that are structural for the economy and markets: on the one hand, the deterioration of public finances and the rise in debt-servicing costs in Western countries; on the other, AI and the financing of its colossal capital expenditure. These two issues are feeding into each other. To fund AI development, technology giants — the hyperscalers — have made massive use of the bond markets in recent months. From barely 1% in 2024, bond issuance dedicated to financing AI reached 18% of total global Investment Grade[2] issuance in the first half of 2026. This forceful arrival of hyperscalers in the credit market has driven a broad surge in issuance volumes, up 36% since the start of the year compared with the same period last year.

This flood of bond supply is increasingly difficult for investors to absorb. Or rather, it is leading to reallocations, with long-dated government debt among the first casualties. The reason is simple: the competition is direct. First, hyperscalers mostly fund themselves at long maturities. Around 70% of the debt issued has a maturity of seven years or more, with an average maturity estimated at around 15 to 17 years. A recent Goldman Sachs[3] study estimates that hyperscaler issuance accounts for more than 40% of total global Investment Grade issuance with maturities beyond 15 years.

Second, technology companies are now paying fairly generous spreads versus government bonds in the various currencies in which they issue. In recent days, Alphabet, for example, issued a 20-year bond in Australian dollars at a yield of 6.95%… while the 20-year Australian sovereign yield stands at 5.45%. Finally, the investors likely to buy hyperscaler debt are today largely the same ones who hold government debt. In the United States in particular, as the Fed has reduced the size of its balance sheet and foreign central banks’ holdings of US debt have stagnated, private investors have stepped in. They now hold 73% of US debt, compared with 50% ten years ago. And these are precisely the same investors most likely to buy hyperscaler debt, often issued at an attractive premium, at similar maturities, and without the public-finance deterioration issues that sovereign bonds carry.

In short, bond markets are witnessing a clear crowding-out effect: AI debt flooding the market is being absorbed at the expense of government bonds. And this is likely only the beginning. On the one hand, AI financing needs are rising: several hundred billion dollars of new bond issuance can be expected in the quarters ahead. On the other hand, governments’ financing needs also continue to increase, either because of the ongoing deterioration in public finances (France, the United States, the United Kingdom), or to fund structural investment (Germany, Japan to a lesser extent). Unless one imagines a return to massive asset purchases by central banks — which hardly appears to be on the agenda — this heightened competition can only keep driving yields higher, with, in time, the risks this poses for financing the global economy.

Main changes: tightened a few phrasings (“in fact stands” vs “is in fact standing,” “the rise in debt-servicing costs”), fixed subject-verb agreement issues (“technology companies… are paying,” “foreign central banks’ holdings… have stagnated”), smoothed “the investors likely to buy… are today largely the same ones,” spelled out “ten years ago,” and merged the closing fragment into the final sentence for better flow while keeping your stylistic dashes and ellipses intact.

[1] French government bond

[2] Refers to issuers or bonds rated between AAA and BBB or Aaa and Baa3. These securities carry a low default risk.

[3] IG Credit – Too Much, Too Fast, Jeffrey Papai, 12.07.2026

Read all our Strategic Case articles