AI data center leases signed by five of the world’s largest technology companies now add up to roughly 1.09 trillion dollars in future payments, according to a Reuters analysis of company filings published this week, and almost none of that figure currently shows up as debt on any of their balance sheets. Microsoft, Meta, Oracle, Amazon, and Alphabet have all committed to leases for data center capacity that has not yet been built, locking in AI infrastructure spending years before the facilities exist while accounting rules let the obligation sit quietly in the footnotes rather than the balance sheet itself.
Why AI Data Center Leases Don’t Show Up as Debt Yet
Reuters’ analysis found the 1.09 trillion dollar figure is nearly four times the roughly 285 billion dollars in lease liabilities the same five companies have already recognised on their balance sheets. The gap comes down to a specific accounting rule: a signed lease is only recorded as a liability once the leased facility is actually available for use. Until a data center is built and operational, the future payments a company has already committed to sit only in the notes accompanying its financial statements, disclosed but not counted toward the debt-like leverage ratios investors and rating agencies typically watch.
That is not a loophole companies are hiding, exactly. Rating agencies can and do factor disclosed future commitments into their own analysis, and Reuters notes the 1.09 trillion dollar figure cannot simply be added to reported debt, since it represents undiscounted payments spread across many years rather than the present-value calculation used for recognised lease liabilities. But the scale of what remains outside standard leverage measures is large enough that it materially understates how financially committed these companies already are to the AI infrastructure buildout, regardless of how AI demand actually develops from here.
Oracle Carries the Most Concentrated Risk
Oracle’s position stands out from the other four companies specifically because of how large its uncommenced commitments are relative to its existing balance sheet. The company has disclosed 260 billion dollars in future lease payments for data centers that have not yet begun, nearly seven times its 37.89 billion dollars in already-recognised lease liabilities, with most of the new capacity expected to come online between fiscal 2027 and fiscal 2029 under leases running 15 to 19 years.
Oracle’s own filings warn that the timing, renewal terms, and pricing of these data center leases may not line up neatly with the customer contracts the capacity is meant to serve, leaving the company exposed if a customer does not renew or cannot perform on its side of the arrangement.
Oracle’s borrowings already stood at about 4.4 times trailing earnings before interest, taxes, depreciation, and amortisation as of the end of May, a ratio that climbs to roughly 5.7 times once recognised lease liabilities are folded in. S&P Global Ratings has already begun incorporating Oracle’s 260 billion dollars in uncommenced leases into its own adjusted-debt forecasting, projecting leverage around 4.4 times in fiscal 2027 once that adjustment is made, a clear signal that credit analysts are not waiting for the accounting rules to catch up before treating these commitments as real financial exposure.
Microsoft and Meta Lead the Pack in Absolute Terms
Microsoft holds the largest single pipeline of any of the five companies at 329.1 billion dollars in future lease commitments, against 88.52 billion dollars already recognised on its balance sheet. Meta disclosed 278.99 billion dollars in uncommenced operating and finance lease payments in its most recent filing, then signed a further 68 billion dollars in new data center leases in July alone, a single month’s addition large enough to push the five companies’ combined known pipeline to roughly 1.16 trillion dollars once that later agreement is included.
Alphabet’s disclosed uncommenced leases total 85.2 billion dollars, the smallest of the five, while Amazon’s 137.21 billion dollar figure is harder to compare directly since its broader lease portfolio also includes warehouses, offices, aircraft, and vehicles alongside data centers specifically.
This Isn’t a New Warning, Just a Bigger Number
Credit analysts flagged a version of this exact risk six months before Reuters’ latest tally. Moody’s Ratings reported in February that the same five hyperscalers had already amassed 662 billion dollars in uncommenced lease obligations, a figure Moody’s analysts calculated was equivalent to 113 percent of the companies’ most recent adjusted debt. Moody’s traced part of the mechanism to residual value guarantees, contractual backstops that let a landlord recover the difference if a data center’s market value falls short after a company declines to renew a lease, which under current accounting rules do not need to be recorded as liabilities unless renewal is deemed probable rather than merely likely.
Meta’s own filings illustrate how much that distinction can hide. The company disclosed data center leases commencing in 2029 with an initial commitment of roughly 12.3 billion dollars, alongside a residual value guarantee with an aggregate threshold of 28 billion dollars that was not recorded as a liability at all because Meta deemed a payout under that guarantee not probable. The jump from Moody’s 662 billion dollar figure in February to Reuters’ 1.09 trillion dollar tally in August, in roughly five months, shows how quickly this category of off-balance-sheet commitment is growing even as the underlying accounting treatment stays exactly the same.
Part of a Much Larger Global Infrastructure Race
These lease commitments are the financial complement to a physical buildout LiveAIWire has tracked extensively this year. Our coverage of the consolidation among AI compute infrastructure providers found the same underlying pressure driving both stories: demand for AI training and inference capacity is expanding faster than the traditional capital cycle for building data centers can comfortably absorb, pushing hyperscalers toward exactly the kind of accelerated, long-duration lease structures that Reuters and Moody’s are now flagging as a financial risk.
The same dynamic is visible in LiveAIWire’s reporting on the EU’s own AI gigafactories tender, where governments are racing to build sovereign compute capacity on a similarly compressed timeline, treating speed of construction as a competitive necessity that leaves less room for the conventional financial caution a slower buildout would allow.
The physical and financial pressures converge most visibly in communities where the buildout is actually happening. LiveAIWire’s coverage of Google’s contested data center project in Visakhapatnam found local water scarcity concerns colliding directly with a facility built to serve exactly the kind of long-term AI infrastructure commitment now showing up in Reuters’ 1.09 trillion dollar figure, a reminder that these lease obligations are not abstract accounting entries but commitments to build specific, resource-intensive facilities in specific places, whether or not local communities were meaningfully consulted before the lease was signed.
What This Means for You
If you hold shares in any of these five companies, or in a fund with meaningful exposure to them, the practical takeaway from this reporting is that headline debt and leverage figures currently understate each company’s real financial commitment to the AI buildout, and Oracle’s position in particular is worth watching closely given how large its uncommenced obligations are relative to its existing balance sheet. Rating agencies are already adjusting their own models to account for this gap, which means credit downgrades or tighter borrowing terms could arrive well before these leases formally convert into recognised liabilities on a quarterly earnings report.
For everyone else, the number is a useful corrective to the assumption that the AI infrastructure race is being funded entirely out of these companies’ enormous cash reserves. A meaningful share of it is being financed through long-duration lease commitments that function economically like debt while sitting outside the accounting categories investors and analysts have traditionally used to measure it, a gap that will close automatically as facilities come online between now and 2029, whether or not AI demand grows enough by then to justify the commitment.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.