Non-US Capital Is Funding European AI Infrastructure to Bypass Dollar Dependency

The capital stack for European AI is already non-American, and Brussels has not noticed

AtlasEdge closed a €1.2 billion debt facility last week with a syndicate that included Mizuho and MUFG alongside ABN Amro and KfW. The non-obvious point is not that Asian banks are lending into European AI infrastructure. It is that they are doing so via debt, on commercial terms, at exactly the moment Brussels is debating whether to screen non-EU equity owners out of the sector entirely. The governance regime being designed and the capital actually arriving are looking at different problems.

The political conversation in Brussels assumes that European AI sovereignty is a fight between European champions and US hyperscalers. The bond market disagrees. Amazon raised €14.5 billion in a single euro deal in March, the largest corporate euro bond ever issued, with proceeds explicitly earmarked for AI capex and held in euros rather than swapped back to dollars. [1] Alphabet's ¥576.5 billion yen issuance was the largest ever by a foreign borrower in that market, and a single round of sterling issuance lifted it to fourth-largest borrower in the ICE BofA sterling corporate index. [2] Non-dollar issuance now accounts for roughly 30% of hyperscaler bond funding, double the prior share, according to Bank of America figures cited by BNP Paribas and JPMorgan syndicate heads. [3]

Morgan Stanley expects total hyperscaler borrowing in euro debt to reach €50 billion in 2026, a flow that could push US firms past France as the eurozone's largest source of corporate debt issuance. [4] Read that sentence twice. The largest corporate borrower in Europe's own debt markets, funding the most strategically sensitive infrastructure on the continent, will shortly be American. The capital is European in currency and partly Asian in its lender base. The operators and the eventual economic owners are American. This is the actual sovereignty question, and nothing in the draft Cloud and AI Development Act, scheduled for June 3, addresses it. [5]

Brussels is built to screen ownership. The Dutch government blocked the Kyndryl acquisition of Solvinity on infrastructure-control grounds, the template most European capitals expect to use. [6] EU industry chief Stéphane Séjourné wants a larger role for European firms in cloud procurement; defence chief Andrius Kubilius wants European-player preferences for military workloads; tech chief Henna Virkkunen, generally judged the likely winner of the internal fight, wants rules applied uniformly to all operators. [7] Every one of these positions is about who runs the facility and who owns the equity. None addresses who holds the debt.

This matters because covenants in a €1.2 billion syndicated facility carry operational rights that an equity screen never sees. Acceleration triggers, change-of-control provisions, information rights, and security over assets all sit at the lender level. AtlasEdge's syndicate spans European, Japanese, and US banks; the facility was oversubscribed. [8] Whether any of those lenders has data-access or audit rights over the underlying sites is not public, and the EU regime as currently drafted gives Brussels no mechanism to ask. The €1 trillion investment gap the Commission itself identifies between US and European digital infrastructure capacity will close, if it closes at all, via debt markets that no European regulator currently maps. [9]

The Asian capital story: real but narrower than the headlines

A separate set of claims circulates about SoftBank funding a €75 billion French AI facility and Hong Kong launching an offshore yuan vehicle for AI infrastructure. Neither is verifiable from primary sources and neither should anchor a strategy. SoftBank's 2017 commitment to invest £100 billion in the UK is a useful reminder that headline pledges from Asian capital to Western tech are routinely a multiple of what is deployed.

What is verifiable is narrower and more useful. Japanese banks are now regular co-lenders in European data-centre debt. Singapore captured roughly 99% of disclosed Southeast Asian AI infrastructure equity between 2019 and mid-2026, a total of about US$1.2 billion. [10] MiniMax's January 2026 Hong Kong listing was the only AI infrastructure IPO recorded in that dataset. Regional early-stage funding fell 48% year-on-year in 2025. [11] The Asian capital story for AI infrastructure is not an equity wave. It is a bank-debt and bond-investor presence, concentrated in a few financial centres, showing up as syndicate participation rather than as ownership.

For an enterprise sourcing AI capacity in Europe over the next eighteen months, that distinction is the whole game. The operator you contract with is overwhelmingly likely to be a US hyperscaler or a Europe-domiciled colocation business. The capital structure underneath that operator is euro-denominated and partly Asian-lent. If a European political backlash arrives, it will hit the operator layer first while leaving the capital layer intact and unscreened.

Counter-case: currency optimisation, not a sovereignty shift

The strongest counter-argument is that hyperscalers borrow in euros and yen because the curves are cheaper and because they need natural hedges against the euro and yen revenue their European and Japanese subsidiaries generate. There is no geopolitical signal in any of it. David Zahn at Franklin Templeton has flagged the obvious concentration risk: pulling AI-correlated credit into European fixed income simply imports US tech volatility into European pension portfolios without producing anything resembling sovereignty. [12] On this reading, the AtlasEdge syndicate is a straightforward commercial transaction, MUFG and Mizuho are simply doing what Japanese megabanks have done in European infrastructure for two decades, and the right framing is portfolio diversification by global lenders rather than capital realignment.

This is partly right, and where it is right, it makes the strategic problem worse rather than better. If the flows are commercial and currency-driven, they are also durable and insensitive to political signalling. Brussels cannot wish them away with a procurement preference. They will continue regardless of whether the Cloud and AI Development Act favours European operators, because the lenders are not making a bet on operators; they are making a bet on euro-denominated AI capex as an asset class. The erosion of sovereignty is happening through a channel that European industrial policy was not designed to see, let alone control. BNP Paribas's investment-grade finance desk has put the trajectory plainly: within twelve months, some of these hyperscalers will be among the largest issuers in any currency, globally. [13] That is a capital-markets structure story, and it has already moved past where the regulatory debate sits.

The second-order implication for enterprise finance and strategy teams is straightforward. Counterparty risk on European AI capacity now requires reading the debt stack, not the cap table. Any procurement decision running longer than three years should require disclosure of senior lender identity, change-of-control covenants, and any security granted over physical sites. Most large enterprises do not currently demand this, because most large enterprises still assume that ownership equals control. In European AI infrastructure over the next eighteen months, that assumption will not hold.

What to watch

1. The final text of the EU Cloud and AI Development Act and whether it contains any provision addressing debt holders, security interests, or lender disclosure in critical infrastructure. If the published Act, expected from the Commission on or shortly after June 3, 2026, addresses only operators and equity ownership, the regulatory gap identified here is locked in for at least one legislative cycle. [14]

2. Whether Morgan Stanley's €50 billion hyperscaler euro-debt forecast for 2026 is met or exceeded by year-end. A figure at or above €50 billion confirms that US firms have become the largest corporate borrowers in the eurozone and validates the structural-shift thesis. A material shortfall would suggest the March Amazon deal was a one-off and the counter-case is correct. [15]

3. Whether any EU member state replicates the Dutch Kyndryl/Solvinity block against a transaction involving Asian lender syndicates rather than Asian equity owners. A blocked or conditioned debt facility, as opposed to a blocked acquisition, would be the first concrete signal that European screening regimes have caught up with the channel through which capital is actually arriving. [16]