Financing for the infrastructure that powers artificial intelligence — data centres, AI chips and compute capacity — has expanded to unprecedented levels in recent months. According to aggregated reporting by the Financial Times and estimates from Morgan Stanley, about $236 billion of AI-related loans and bonds had been issued worldwide by the end of May 2026, roughly four times the volume a year earlier. Morgan Stanley expects issuance to approach $570 billion by the end of 2026.
Notable transactions
Among the headline deals in the sector are:
- a $35 billion chip financing for Broadcom;
- a $27 billion loan package for the Meta Hyperion data-centre project, financed with involvement from Blue Owl Capital;
- a $3.2 billion bond issue by data-centre developer TeraWulf backed by long-term Google lease guarantees;
- a $3.1 billion loan to CoreWeave to buy and deploy Nvidia AI chips;
- an earlier $8.5 billion chip loan to CoreWeave covered by hyperscaler contracts.
These transactions highlight that financing is now flowing not only to big tech companies’ internal projects but also to suppliers, data-centre builders and specialist cloud providers.
Guarantees and new financing structures
A central feature of the current financing model is that large technology companies — Google, Meta, Amazon and Microsoft — underwrite demand for capacity through long-term lease agreements. Where such contracts secure utilisation of a data centre, financing costs can fall substantially: banks estimate pricing can be roughly halved when long-term take-or-pay style commitments are in place.
As a result, banks have begun to design new instrument types that institutional investors — pension funds, insurers and asset managers — can buy. Morgan Stanley reports it has sold more than $40 billion of these new-style AI-infrastructure bonds since last year, and is taking the model into markets in Europe and Asia.
Scale and emerging risks
Morgan Stanley estimates that building global AI infrastructure could require as much as $10 trillion of investment over the coming years. The bank anticipates that AI-linked bonds could soon become one of the larger segments of the corporate credit market.
At the same time, analysts warn of growing risks. The Financial Times and Moody’s note that the financing model becomes riskier as loans rely less on the creditworthiness of giants like Google or Meta and more on smaller or younger AI developers such as Anthropic or OpenAI. Moody’s observes that this investment cycle is historically large and lacks close precedents, making the long-term systemic implications difficult to quantify at present.
Hungary joins the EU AI Gigafactory programme
Hungary has signed up to the European Union AI Gigafactory programme. Tanács Zoltán, Hungary’s Minister for Science and Technology, has argued that Europe is falling behind the United States and that failure to participate in large AI-infrastructure projects risks a prolonged competitive disadvantage.
The Hungarian government plans a total national commitment of €500 million to the programme over time and will initially contribute €25 million to the Polish-led international consortium. With further participation Hungary could obtain ownership stakes and access rights in an AI centre with more than 100,000 GPUs. The stated goals include cheaper and broader access to large-scale compute for domestic companies, startups, research institutes and public bodies, as well as support for R&D grants and national high-performance computing capability.
Conclusion
The scale of recent deals and the volumes of lending and bond issuance show that AI infrastructure is driving a major reallocation of capital in global credit markets. Big-tech guarantees and long-term contracts lower financing costs and enable new investorable products, but the widening scope of counterparties and the size of the cycle introduce novel risks that market participants and regulators will be watching closely.
