Financing the AI boom – In Conversation with Imène Rahmouni-Rousseau

Imène Rahmouni-Rousseau · European Central Bank

Hyperscalers' roughly $1 trillion a year AI capex exceeds internal cash flow, so they must raise external financing.

Expected AI capex is about $1 trillion a year until 2030, roughly 3% of US GDP and comparable to the GDP of several medium-sized European economies. Big tech companies can no longer rely on internal financing to fund this, so even the largest need external financing. They are going to private debt markets and public markets, and bond markets now at scale, which makes AI a bond-market story as well as an equity one. She also flags hidden leverage, off-balance-sheet financing and circular financing as risks in this funding chain.

“Even the biggest one, they need to resort to external financing to fund this arms race in AI infrastructure.”at 0:26

Imène Rahmouni-Rousseau · European Central Bank

Part of hyperscalers' AI investment is likely to stay in Europe, where they already earn a quarter of revenue.

A quarter of hyperscaler revenues come from Europe today, and they are already present in the European data-centre landscape. Europe is therefore a credible destination for AI funding raised for infrastructure. Some of the money can still flow back to the US through the swap market, and the data is not good enough to say how much stays in Europe. Market intelligence from talks with issuers and banks suggests part of it does stay.

“a quarter of the revenues of hyperscalers come from Europe today... And I think that this is a quite credible destination for this money.”at 15:22

Imène Rahmouni-Rousseau · European Central Bank

US hyperscalers' large euro bond deals show Europe's corporate bond market is deep enough for big issuance.

US hyperscalers are issuing reverse Yankee bonds in Europe, a US company issuing in euros. They are using the depth of the euro market, which the ECB's corporate bond purchase programme helped deepen. They issue in large sizes, at longer maturities of 10 to 50 years, and often at high ratings, with some rated AAA. Amazon's roughly €14.5 billion sale was a first of its kind, showing that large single-company issuance is possible in Europe. She sees this as a vote of confidence in the European corporate bond market. It also opens the door for European companies to issue at similar size, and gives euro-area investors and new foreign money access to AI-linked credit. Some hyperscalers may also swap euro proceeds back into dollars, which carries currency risk.

“these US companies are opening the door for our European companies to do the same. Yeah, in the future.”at 9:42

Imène Rahmouni-Rousseau · European Central Bank

Hyperscaler bond issuance crowds out other issuers only temporarily, though hyperscaler spreads have risen.

The ECB tested whether hyperscaler issuance crowds out the private sector, reversing the usual emerging-market idea that sovereign issuance crowds out private bonds. On days when hyperscalers come to market, other issuers face higher new-issue premia, but the effect is temporary and fades. Other issuers have adapted by avoiding hyperscaler issuance windows. The spread on hyperscaler bonds, the gap to the risk-free rate, has risen as their issuance has continued, which shows investors now demand a premium to hold their debt. The rise has been very limited for other sectors. The overall message is limited crowding out but not a fully benign one.

“issuers from other sectors have learned to adapt. Yeah? They don't come in the issuance windows of hyperscalers. They adjust their timing.”at 13:17

Imène Rahmouni-Rousseau · European Central Bank

European equity markets are missing out on the AI rally, despite ASML's presence in the Netherlands.

She contrasts AI-driven markets such as South Korea, where Samsung and SK Hynix are a poster child, with Europe, whose equity markets are largely missing out on AI. ASML in the Netherlands is the main European exposure she names. She argues the game is not over, since European companies will adopt AI and it will require large investment. Her argument is largely qualitative and is weaker than her credit-market analysis.

“I think our equity markets on AI is a little bit missing out.”at 3:08

Imène Rahmouni-Rousseau · European Central Bank

An AI correction could spill into euro-area households, pension funds and European firms' financing, and GDP.

Euro-area households hold hyperscalers equal to about 7% of their equity portfolios. New bond deals add to these exposures, and the exposures will grow mechanically. Pension funds, insurers and investment funds where households save are therefore increasingly exposed to AI. If AI earnings fail to match the huge investment plans, weakness in one part of the tightly connected ecosystem (chipmakers, data-centre builders, cloud providers) can spread across it. A correction could be broader than in equity markets alone, especially where positions are financed by leverage, including leveraged ETFs on AI stocks that magnify losses. European companies would find financing harder, and an AI correction could weigh on overall GDP growth. She does not take a view on whether AI is a bubble.

“Euro Area investors households are also heavily invested in U.S. Tech... their holdings of hyperscalers amount to 7% of their equity portfolios.”at 16:57