The artificial intelligence (AI) arms race is transforming the global data centre market. AI is driving tech debt to record levels; however, an increasing share of that borrowing is being quietly shifted off corporate balance sheets.
Recently, there has been a fundamental change in how AI infrastructure is financed. This raises questions for data centre operators, investors, and lawmakers about risk, transparency, and sustainability.
Debt Becomes the Default Funding Model
Since 2000, Big Tech has been defined by fortress balance sheets: exceptionally strong and healthy financial positions. Companies like Meta, Google, Microsoft, and Oracle generated enormous revenues and carried relatively little debt, funding expansion largely from operating income.
That model is changing because of AI spending. According to figures from Dealogic, global tech companies issued a record $428.3bn in bonds through early December 2025, as spending on AI data centres surged.
This is a major shift from past practice. Instead of funding expansions using cash reserves, tech companies are increasingly turning to debt markets to finance AI infrastructure.
Reuters analysis of around 1,000 tech firms with market capitalisations above $1 billion (£741 million) shows how quickly leverage is rising. Median debt-to-EBITDA ratios reached 0.4 by the end of September 2025, nearly double the level seen during the pandemic-era debt spike in 2020.
Michelle Connell, president at Portia Capital Management, described AI data centres as a structural shift for the sector. Companies are forced to continually reinvest in AI infrastructure because the technology ages quickly, and accelerator chips must be replaced frequently. Unlike standard enterprise IT, AI infrastructure must be refreshed every few years to stay competitive.
An analysis by the Financial Times found that tech companies have shifted more than $120 billion (£88.9 billion) of AI data centre spending off their balance sheets through special-purpose vehicles (SPVs) funded by investors. SPVs allow companies (notably Meta, Oracle, and xAI) to secure financing that does not appear on corporate balance sheets. This helps these companies maintain high credit ratings and better financial indicators tied to their balance sheets.
In practice, the model works by creating SPVs that own the physical assets of AI data centres, like the land, buildings, power infrastructure, and even processing chips. Institutional investors, including Pimco, BlackRock, Apollo, Blue Owl Capital, and banks such as JPMorgan, provide debt and equity to the SPV, as the tech firms sign long-term lease agreements for capacity.
Meta’s $30 billion (£22 billion) Hyperion data centre project shows how the SPV model has evolved. Completed in October, the deal created an SPV called Beignet Investor, backed by Blue Owl Capital. Around $27 billion (£20 billion) came from loans provided by investors, including Pimco, BlackRock, and Apollo, with an additional $3 billion (£2.2 billion) in equity from Blue Owl.
This $30 billion (£22 billion) did not appear as debt on Meta’s balance sheet, allowing the company to raise a further $30 billion (£22 billion) in the corporate bond market shortly afterwards. Meta does retain exposure: it owns 20% of the SPV and must cover losses if the facility’s value falls below a certain level and the lease is not renewed.
Similar structures are being pursued by xAI, which is raising $20 billion (£14 billion) through an SPV that will buy Nvidia GPUs and lease them back to the company, and by CoreWeave, which created SPVs to finance its $11.9 billion (£8.8 billion) computing contracts with OpenAI.
Private Credit Steps Into the Spotlight
The rise of SPVs has pulled private credit markets deep into the AI infrastructure boom. UBS estimates that tech companies had borrowed around $450 billion (£333 billion) from private funds by early 2025, $100 billion (£74.1 billion) more than in the previous 12 months.
Using private credit to finance AI infrastructure is attractive to both sides. Tech companies gain flexible, large-scale funding without balance sheet impact, while investors gain exposure to what they hope will be long-term, infrastructure-like cash flows anchored by blue-chip tenants. But is it sustainable?
There is a possibility that AI demand falls short of expectations. If it does, the risk is concentrated among a relatively small number of customers.
Scepticism about AI revenue forecasts is already showing up in equity markets. Shares in some companies heavily exposed to AI data centre investment, including Oracle, have been under pressure as analysts question when (or whether) AI workloads will generate returns commensurate with their capital intensity.
