Alphabet, Meta, Microsoft, Amazon and Oracle had about $3.2 trillion in AI-related off-balance-sheet commitments at the end of September 2026, according to Nikkei. The reported total rose 90% from the end of March, showing how quickly the companies are promising future payments to secure computing capacity, equipment and electricity.
Nikkei’s report, published November 4 in Japan, or November 3 UTC, puts those commitments at roughly 80% of the five companies’ combined assets. That is a consequential measure of the AI infrastructure buildout, but it needs careful interpretation: future contractual commitments are not interchangeable with money already borrowed.
The accessible portion of Nikkei’s article confirms the headline findings. The detailed composition below comes from BusinessKorea’s attributed account. Nikkei’s tally includes estimates, and changes in Alphabet’s calculation methodology complicate comparisons over time. Those qualifications matter when interpreting the reported 90% increase.
What the $3.2 Trillion Includes
BusinessKorea says Nikkei grouped two broad categories of AI-related obligations: long-term purchasing agreements and leases that have not yet commenced.
| Category | Reported share | Approximate value implied by the total | What it covers |
|---|---|---|---|
| Long-term purchase agreements | 60% | $1.9 trillion | Semiconductors, servers and electricity |
| Equipment and facility leases not yet commenced | 40% | $1.3 trillion | Future use of equipment and facilities |
The dollar amounts in the table are calculations using the reported $3.2 trillion total and the 60/40 split, rounded to avoid suggesting more precision than the underlying reporting provides.
The distinction between the two categories is useful. A semiconductor order secures equipment. An electricity agreement secures an input needed to operate that equipment. A facility lease secures space or infrastructure. All can support AI capacity, but their payment schedules, commercial purposes and exposure to changing demand differ.
Combining them gives a broad view of future cash demands. It does not produce a single, standardized debt instrument with one interest rate or repayment date.
These agreements also explain how competition in the AI industry reaches beyond model development. Companies are competing for the physical resources needed to train and serve models. Committing early can help secure supply and give infrastructure providers enough confidence to expand. The financial tradeoff is that spending decisions become less flexible before the resulting capacity starts earning revenue.
My view is that this is the tally’s most useful contribution: it shows how much future capacity the companies are committing to, beyond what they have already paid to build.
Why These Obligations Are Off the Balance Sheet
BusinessKorea describes the commitments as future payment obligations disclosed in financial-statement notes, with recognition generally following later as purchases are fulfilled or leases commence. Its summary is an explanation of the accounting treatment, not a substitute for the individual contracts or accounting policies.
The basic distinction is between promising a future transaction and already receiving financing, goods or the use of an asset.
A long-term purchase agreement can be commercially important from the moment it is signed. Yet an agreement to buy servers in the future is different from a loan that has already delivered cash to the borrower. Similarly, signing a lease for a facility that is not yet available for use creates a future commitment before the lease begins.
“Off balance sheet” therefore describes where and when an obligation is recognized. It does not mean the obligation is undisclosed or economically irrelevant.
For investors, two readings would be misleading:
- Treating the entire $3.2 trillion as conventional funded debt would blur important accounting and contractual differences.
- Ignoring it because it sits outside headline balance-sheet liabilities would overlook potentially substantial future cash outflows.
The practical task is to connect the footnotes to the cash-flow outlook. When are payments due? What has to happen before they become payable? What flexibility does the buyer retain? The aggregate number cannot answer those questions on its own.
Alphabet and Meta Account for a Large Share
Alphabet had approximately $1.05 trillion in commitments at the end of September, while Meta had approximately $810 billion, according to BusinessKorea’s summary of Nikkei. Together, those amounts represent about $1.86 trillion, or roughly 58% of the five-company total, based on straightforward arithmetic.
Alphabet’s reported amount was about 2.6 times its March level, subject to the methodology caveat. Meta’s commitments reportedly increased 90% over the same period.
BusinessKorea links Meta’s obligations partly to a data-center leasing joint venture with BlackRock and large capacity-purchase agreements with emerging cloud providers. It also reports a roughly $21 billion agreement with CoreWeave, signed in the spring, for cloud capacity through 2032.
That contract horizon illustrates why these figures should be read differently from quarterly spending. A commitment can stretch across years, while a capital-expenditure figure measures spending during a particular period.
The reported comparison with combined assets also needs context. Obligations equal to 80% of assets are a striking scale indicator, but that is not a conventional debt-to-assets ratio. It compares future commitments with a balance-sheet snapshot. Payment timing, contract terms and future cash generation all affect the burden.
I would use the 80% figure as a prompt to examine the commitments more closely, not as a standalone measure of solvency.
How the Financing Chain Can Concentrate Risk
Long-term contracts help infrastructure providers raise money. BusinessKorea reports that emerging cloud companies, often called neocloud providers, have obtained financing on the strength of agreements with major technology buyers.
The mechanism is straightforward: a lender assessing a new data-center project can take comfort from a large customer’s promise to purchase capacity. That contract helps support the provider’s financing, which funds infrastructure needed to fulfill the contract.
A simplified chain looks like this:
- A major technology company commits to buying future capacity.
- An infrastructure provider uses that expected business to support fundraising.
- Lenders and investors supply money for facilities and equipment.
- Suppliers expand or reserve production against expected demand.
Frequently Asked Questions
3 questions
1Is the $3.2 Trillion All Borrowed Money?
No. Nikkei’s tally combines future purchase agreements and leases that have not yet commenced. These can require substantial payments, but they differ from funded debt.
