How AI data centre spending is creating hidden debt for tech giants

AI spending by Alphabet, Microsoft, Amazon, Meta and Oracle is driving a surge in hidden debt, with off-balance-sheet commitments reaching $1.65 trillion

The race to dominate artificial intelligence (AI) is creating a new financial challenge for some of the world's biggest technology companies. A study by Nikkei Asia estimates that the combined off-balance-sheet, or "hidden", debt of Alphabet, Microsoft, Amazon, Meta and Oracle has jumped eightfold in about four years to around $1.65 trillion. This is now higher than the roughly $1.35 trillion of debt shown on their balance sheets, making it harder for investors to judge their true financial exposure.

 

The figures may rise further as Alphabet, Microsoft, Amazon and Meta are due to report their latest quarterly earnings this week.

 

 

What is hidden debt?

 

Hidden debt refers to financial commitments that do not appear directly on a company's balance sheet but are disclosed in the notes to its financial statements.

 

These commitments are legal and allowed under accounting rules. However, they can make it difficult for investors, especially retail investors, to judge how much companies have actually committed to spending in the future.

 

In the case of AI companies, much of this hidden debt comes from long-term agreements to lease data centres or buy expensive graphics processing units (GPUs), servers and other computing equipment that have not yet been delivered or started operating.

 

Since these assets are still under development or not yet in use, the related commitments remain off the balance sheet until certain accounting conditions are met.

 

AI infrastructure is driving the surge

 

Technology companies are spending heavily to expand AI capabilities by building data centres and securing advanced graphics processing units (GPUs), servers and other computing equipment.

 

Instead of owning many of these facilities outright, companies often sign long-term lease agreements with data centre operators. In many cases, the operators provide the land, buildings and power infrastructure, while the tech companies commit to using them for years.

 

Since these projects are still under development, many of the related lease obligations remain off the balance sheet until the facilities become operational.

 

Meta and Oracle stand out

 

Among the five companies, Meta has the largest estimated hidden debt at around $420 billion, nearly three times its reported debt.

 

Oracle has seen one of the fastest increases. Its hidden debt reached about $273.3 billion by the end of May, more than 30 times higher than four years ago. The company is expanding its large Stargate AI data centre project with OpenAI through long-term lease arrangements with external data centre operators.

 

Although these commitments follow accounting rules, some investors and analysts are becoming more cautious, the news report said.

 

Morgan Stanley has examined the growing off-balance-sheet obligations in a report for investors, while Moody's has warned that lease commitments for projects that have not yet started are rising rapidly.

 

The companies argue that future earnings from AI and cloud computing will justify these investments. By the end of March, Microsoft, Alphabet and Amazon had a combined cloud services order backlog of around $1.45 trillion, providing visibility into future revenue.

 

New funding models add complexity

 

Institutional investors are also playing a bigger role in financing AI infrastructure.

 

Meta formed a joint venture with investment funds managed by Blue Owl Capital to develop a large data centre in Louisiana. The project was initially valued at $27 billion in 2025, but Meta has since said its total investment is expected to exceed $50 billion.

 

Meta owns a minority stake in the operating company while leasing the facility, allowing it to secure computing capacity without recording the full financial commitment as debt immediately. The agreement also includes protections for investors if the project is no longer needed and the lease is terminated.

 

Why experts are concerned

 

Many technology companies are already raising money through corporate bonds and share sales because AI investments are outpacing their earnings.

 

The additional use of institutional funding has raised concerns that spending on AI infrastructure may be growing too quickly, Nikkei Asia reported.

 

Economists at the Bank for International Settlements described this approach as "shadow borrowing", where companies increase financial commitments without adding debt directly to their balance sheets.

 

Experts cited by Nikkei Asia warned that if AI demand slows or data centres remain underused, companies could face large losses once these obligations move onto their balance sheets.

 

AI investment continues to accelerate

 

Despite these concerns, global spending on AI continues to rise.

 

According to SenseAI Ventures' 'State of AI Report 2026', global AI investment reached $800 billion in 2025. Venture capital funding almost doubled to $226 billion, showing that the industry is moving from experimentation to large-scale commercial deployment.

 

The report found that 79 per cent of AI funding came through mega-rounds of more than $100 million, indicating that investors are making larger, high-conviction bets on companies with clear business demand.

 

It also said the focus is shifting away from building AI models and infrastructure alone towards deploying AI in products, business operations and revenue-generating applications, with the application layer emerging as the biggest driver of commercial value.