AI Data Centre Race Builds US$1 Trillion Lease Burden for Big Tech
iTnews
ENRICHED
Details
- Date Published
- 4 Aug 2026
- Priority Score
- 2
- Australian
- Yes
- Created
- 4 Aug 2026, 10:00 pm
Description
Oracle has the largest apparent concentration risk.
Summary
This report highlights a massive US$1.09 trillion financial commitment by major technology firms to expand the infrastructure required for frontier AI models. The rapid build-out of data centers reflects a significant shift in corporate risk as companies lock in long-term capacity to support advanced AI capabilities. These infrastructure investments underscore the race to achieve massive scale in compute power, which is a primary driver of modern AI progress and its associated safety and control challenges. While the article focuses on financial leverage, the scale of this physical infrastructure expansion is a critical indicator of the speed and intensity of global frontier AI development.
Body
Microsoft, Meta Platforms, Oracle, Amazon and Alphabet have committed about US$1.09 trillion ($1.55 trillion) in future payments under leases that have not yet begun, mostly for data centres needed to power the artificial intelligence boom.
The commitments show that a substantial part of Big Tech's AI spending spree has already been locked in, without yet appearing as debt-like lease liabilities on company balance sheets.
If demand for AI computing continues to surge, the facilities will underpin the next phase of cloud growth.
If it does not, the companies could be left paying for vast amounts of costly, long-lived capacity that is difficult to shed.
The total is nearly four times the roughly US$285 billion of lease liabilities already recognised on the companies' balance sheets, according to company filings compiled by Reuters.
The gap reflects accounting treatment. Signed leases generally are not recorded as liabilities until a facility is available for use. Until then, companies disclose the future payments in notes to their financial statements.
The commitments are not hidden, and rating agencies may already account for some of them. But their scale reveals how much of the AI buildout has yet to enter reported lease liabilities, fixed charges and reported leverage measures.
The US$1.09 trillion cannot simply be added to debt. Uncommenced lease commitments are generally undiscounted payments spread over many years, whereas recognised lease liabilities reflect their present value.
Oracle has the largest apparent concentration risk. It disclosed US$260 billion of uncommenced commitments, nearly seven times its US$37.89 billion of recognised lease liabilities.
The commitments are substantially for data centres, expected to begin between fiscal 2027 and fiscal 2029, and generally run for 15 to 19 years.
Oracle has warned that the duration, renewal terms and pricing of its data-centre leases may not align with customer contracts, leaving it exposed if customers do not renew or cannot perform.
Its borrowings equalled about 4.4 times trailing EBITDA at the end of May, according to a Reuters analysis of LSEG data and company filings.
Including recognised operating and finance lease liabilities lifted that ratio to about 5.7 times.
S&P Global Ratings said it incorporated Oracle's US$260 billion of uncommenced leases into its adjusted-debt forecast and expected leverage to be around 4.4 in fiscal 2027.
Microsoft had the largest disclosed pipeline, at US$329.1 billion, against US$88.52 billion of recognised lease liabilities.
Meta disclosed US$278.99 billion of uncommenced operating and finance lease payments, then signed a further US$68 billion of data-centre leases in July, lifting the five companies' known pipeline to about US$1.16 trillion, including those later agreements.
Alphabet reported US$85.2 billion of uncommenced leases, while Amazon disclosed US$137.21 billion.
Amazon's figure is less directly comparable because its lease portfolio also includes warehouses, offices, aircraft and vehicles.