The artificial intelligence race is no longer defined only by headline investments in chips and models. It is increasingly being shaped by long-term commitments to the infrastructure required to power them.
According to a Reuters analysis of company filings, Microsoft, Meta, Oracle, Amazon and Alphabet have committed about $1.09 trillion in future payments under leases that have not yet commenced, largely for AI-focused data centres. The figure highlights how much of Big Tech's infrastructure expansion has already been contractually committed, even though much of it has yet to appear as lease liabilities on corporate balance sheets.
The total is nearly four times the roughly $285 billion of lease liabilities currently recognised by the five companies, based on filings compiled by Reuters.
Accounting rules mask the full scale of AI commitments
The difference between recognised lease liabilities and future commitments stems from accounting treatment rather than undisclosed obligations.
Signed leases are generally recognised as liabilities only when the leased facility becomes available for use. Until then, companies disclose future payment obligations in the notes accompanying their financial statements.
Reuters noted these commitments are publicly disclosed and may already be considered by credit rating agencies. However, they remain absent from reported lease liabilities, fixed charges and commonly used leverage measures until the facilities become operational.
The publication also pointed out the $1.09 trillion should not be treated as equivalent to debt because the commitments represent undiscounted lease payments spread across many years, while recognised lease liabilities reflect their present value.
Cloud demand will determine the payoff
The unprecedented commitments reflect confidence that demand for AI computing capacity will continue rising.
If cloud and AI workloads expand as expected, the new facilities are likely to support the next phase of hyperscale cloud growth.
If demand weakens, however, companies could find themselves committed to paying for expensive, long-duration infrastructure with limited flexibility to scale back, Reuters reported.
Microsoft leads the pipeline as Oracle faces the biggest concentration risk
Among the five companies, Microsoft reported the largest disclosed pipeline of future lease commitments.
According to Reuters, disclosed uncommenced lease commitments include:
- Microsoft: $329.1 billion, compared with $88.52 billion in recognised lease liabilities.
- Meta: $278.99 billion in uncommenced lease payments, followed by an additional $68 billion in data centre leases signed in July.
- Oracle: $260 billion, against $37.89 billion in recognised lease liabilities.
- Amazon: $137.21 billion, although Reuters noted the figure also includes warehouses, offices, aircraft and vehicles alongside data centres.
- Alphabet: $85.2 billion.
Following Meta's additional leasing agreements in July, Reuters estimated the combined known pipeline across the five companies had risen to approximately $1.16 trillion.
Oracle's exposure stands out
While Microsoft disclosed the largest overall commitment, Reuters identified Oracle as carrying the greatest apparent concentration risk.
The company disclosed $260 billion of uncommenced lease commitments, nearly seven times its recognised lease liabilities.
Oracle said the commitments relate substantially to data centres expected to become operational between fiscal 2027 and fiscal 2029, with lease terms generally lasting 15 to 19 years.
The company has also cautioned investors that lease durations, pricing and renewal terms may not always align with customer contracts, creating potential exposure if customers fail to renew agreements or cannot fulfil contractual obligations.
Reuters' analysis of LSEG data and company filings found Oracle's borrowings were equivalent to around 4.4 times trailing EBITDA at the end of May. Including recognised operating and finance lease liabilities increased the ratio to approximately 5.7 times.
S&P Global Ratings told Reuters it had already incorporated Oracle's $260 billion of uncommenced lease obligations into its adjusted debt forecast and continues to expect leverage of around 4.4 times in fiscal 2027.
AI infrastructure spending is entering a new phase
The figures illustrate how competition among hyperscale technology companies has shifted beyond AI models to the physical infrastructure supporting them.
Data centres have become one of the largest capital commitments in the AI economy, requiring long-term investments in power, land, networking and computing capacity well before customer demand fully materialises.
Reuters' analysis suggests much of this next wave of investment has already been contractually secured, even though a significant portion has yet to appear in traditional balance sheet metrics.
As AI adoption accelerates across enterprises and cloud providers, the pace of infrastructure deployment will determine whether these long-term commitments translate into sustained growth or become costly excess capacity.