Microsoft, Meta, Oracle, Amazon, and Alphabet have committed approximately $1.09 trillion in future payments for lease contracts that have not yet started, largely linked to the enormous data center capacity needed for artificial intelligence and cloud services. The figure, compiled by Reuters from regulatory documents and Goldman Sachs analysis, nearly quadruples the approximately $285 billion these companies currently acknowledge as lease liabilities.
The key points about data center leases for AI in 30 seconds
- Microsoft has about $329.1 billion in leases not yet started, and Oracle has another $260 billion.
- Meta reported $278.99 billion and later signed additional agreements worth around $68 billion.
- Amazon reaches $137.21 billion, and Alphabet $85.2 billion, although not all of these contracts are exclusively for data centers.
- Much of these obligations are not yet listed as liabilities on balance sheets because the leases haven’t started.
- The risk arises if future demand for AI compute falls short of the contracted capacity.
These figures shed light on a less visible aspect of the race to build AI infrastructure. When Microsoft, Meta, or Google announce billions in investments in data centers, the focus tends to be on capital expenditures, NVIDIA servers, electricity consumption, or new campuses.
But tech companies are not building all that infrastructure themselves.
They are also reserving capacity from third parties through contracts that can extend over more than a decade.
This results in huge financial commitments that do not yet appear as liabilities on the balance sheet in the same way as bank debt or bonds.
It doesn’t mean companies are hiding $1.09 trillion in debt. They are different accounting categories. But it does mean they have contractual obligations that will influence their costs for many years to come.
Microsoft Already Has $329.1 Billion Committed
Microsoft currently reports the largest known figure.
As of the end of its fiscal year on June 30, the company declared $329.1 billion in lease contracts not yet started, mostly related to data centers.
This growth has been extraordinarily rapid.
Three months earlier, on March 31, 2026, Microsoft reported $196.6 billion. In just one quarter, it added over $130 billion in new commitments.
The company sees this growth as a response to expected demand in the coming years.
It makes operational sense. Building an AI data center isn’t a decision that can be made once the client appears.
It requires securing land, power connections, transformers, cooling, networking, and servers. In some markets, energy availability is also becoming a major constraint to increasing capacity.
Pre-booking infrastructure years in advance ensures it.
But it also means committing to pay for it.
Oracle: $260 Billion for Contracts Up to 19 Years
Oracle’s case is even more striking when compared to its balance sheet size.
The company itself informed the U.S. Securities and Exchange Commission (SEC) that, as of May 31, 2026, it had $260 billion in additional lease commitments, almost all related to data centers.
These contracts will generally commence between fiscal years 2027 and 2029.
An important detail: they will have durations of 15 to 19 years.
They are not capacity reservations for a few months.
Oracle is committing infrastructure that could continue costing them well into the 2040s.
Meanwhile, its balance sheet recognized about $37.9 billion in current lease liabilities. The remaining $260 billion is nearly seven times that amount.
The risk isn’t necessarily that Oracle can’t utilize this infrastructure.
The company is rapidly expanding its cloud business and needs capacity for large AI contracts. The problem lies in the timing gap between committing the infrastructure and knowing what the actual demand will be in ten or fifteen years.
Meta Also Surpasses the $250 Billion Mark
Meta follows a similarly aggressive strategy.
At the end of Q2, it reported roughly $278.99 billion in obligations for leases not yet started, relating to data centers, colocation facilities, and network infrastructure.
Contracts can start as late as 2036 and last up to 30 years.
Then came another move.
In July, after closing the quarter, Meta signed new data center contracts adding approximately $68 billion in commitments, with periods of 18 to 20 years.
This illustrates how fast these figures are growing.
A financial statement can become outdated in just weeks.
Amazon and Alphabet Reach the $1 Trillion Total
Reuters estimates Amazon’s commitments at around $137.21 billion and Alphabet’s at $85.2 billion.
Here, it’s important to note a major distinction.
Not all companies report their figures using the same criteria.
Amazon, in particular, uses leases for numerous assets beyond data centers: logistics warehouses, offices, aircraft, and vehicles, among others. Therefore, its $137.21 billion cannot be considered solely infrastructure for AI.
The aggregate figure of $1.09 trillion should be understood with that caveat.
It shows the scale of commitments made by the five leading tech giants but does not exactly equal $1.09 trillion in data center infrastructure for AI.
Why Not All Is On Their Balance Sheets Yet
The explanation lies in how leases not yet begun are accounted for.
When a company signs a contract today for a data center to operate in two years, that agreement can be binding, but the corresponding right-of-use asset and lease liability are generally not recognized until the facility is available for use.
Until then, the commitment appears in the notes of financial statements.
This is precisely what Oracle explains in its latest annual report: its $260 billion in additional commitments were not reflected on the consolidated balance sheet or the lease maturity table as of May 31.
This does not automatically turn those amounts into hidden debt.
However, rating agencies and credit analysts may incorporate them into their own metrics to evaluate a company’s true financial exposure.
Goldman Sachs estimates that total lease commitments for large hyperscalers reach approximately $1.5 trillion, with about one trillion of that for contracts that have not yet begun.
And leases are just part of the overall bill.
The Actual Commitment to AI Is Even Greater
Tech companies also sign contracts for purchasing chips, servers, electricity, and cloud capacity.
The Financial Times estimates that major players involved in AI infrastructure collectively commit nearly $1.5 trillion in additional purchase obligations.
These sums should not be directly added to lease obligations because they are different types of commitments, and companies use varying accounting standards.
But they highlight how the race for AI is becoming more than just annual spending.
Major tech firms are committing a significant portion of their future investments.
The Wall Street Journal estimates that off-balance-sheet obligations, mainly related to AI infrastructure, could total around $3 trillion, including leases and long-term supply contracts.
The market is also beginning to require more financing. Reuters Breakingviews estimates that Alphabet, Amazon, Meta, Microsoft, and Oracle could collectively spend around $750 billion on data centers in 2026.
The Big Bet Is That GPUs Remain Fully Utilized for Years
All this infrastructure makes sense only if demand continues to grow.
Microsoft needs capacity for Azure and AI services. Meta is building massive clusters for training and inference. Oracle requires infrastructure for OCI and its large AI clients. Amazon and Google face similar needs.
If GPUs and servers stay sufficiently busy, having secured energy and data centers in advance can become a competitive advantage.
The opposite scenario is more complicated.
A slowdown in growth, more efficient models, significant technological changes, or installed capacity surpassing demand could leave facilities underutilized while contracts continue to incur payments.
And a data center contracted for 15 or 20 years cannot be easily downsized like a cloud instance cancellation.
That’s the real significance of that $1 trillion.
The AI race is no longer just about buying the best GPUs. Microsoft, Meta, Oracle, Amazon, and Alphabet are reserving electricity, buildings, and computing capacity that do not yet exist, for demand they expect to have over the next decade.
This bet could give them the infrastructure needed to dominate a much larger industry.
It could also leave them with a long-lasting bill if current AI demand forecasts turn out to be overly optimistic.
Frequently Asked Questions
Are Big Tech truly carrying $1.09 trillion in new debt?
No. They are primarily future commitments from leases not yet started and should not be directly confused with financial debt. However, they represent future contractual obligations that analysts may consider when assessing leverage.
Which company has the most pending leases?
Microsoft leads the known figures with approximately $329.1 billion, followed by Meta and Oracle. Oracle reported $260 billion, almost all related to data centers.
Why do these contracts not appear as liabilities yet?
Because many data centers are not yet operational. Commitments are detailed in financial statement notes, and corresponding lease liabilities are recognized when the contracts start.
What if AI demand decreases?
Companies could end up paying for underutilized capacity for years. Some contracts last between 15 and 30 years, so current decisions could have financial consequences well into the next decade and beyond.
via: LinkedIn
