Analysis separates hyperscaler data-center commitments from reported debt

A financial analysis of disclosures through mid-2026 for Alphabet, Amazon, Meta, Microsoft and Oracle has challenged the practice of treating more than $1 trillion of future infrastructure obligations as debt already carried by the companies.

The analysis divides the obligations into four categories: non-cancellable contractual commitments; financing and construction commitments; leases that have not yet commenced; and off-balance-sheet special-purpose-vehicle financing. The first three are legally binding future payment obligations, but are not equivalent to current debt, the analysis says. In particular, the disclosed figures are not discounted for the timing of payments.

Across the five companies, non-cancellable contractual commitments exceeded $1.5 trillion, with the source linking most of that amount to AI data-center construction. Alphabet alone accounted for more than $800 billion, while Meta’s commitments were cited at about $350 billion. Financing and construction commitments were a smaller category, at about $100 billion in aggregate.

Uncommenced leases across the five companies exceeded $1 trillion, compared with current aggregate debt including leases of about $700 billion. Meta and Microsoft were identified as having the largest volumes of uncommenced leases, at more than $300 billion each. The source separately cited Meta at about $350 billion of leases scheduled to commence by 2036.

The analysis identifies Meta’s data-center financing structure as the least transparent element. According to the secondary-source account, Meta raised multibillion-dollar financing for data centers through SPVs without recording the associated debt on its consolidated balance sheet, while committing to compensate bondholders for losses in the event of non-payment. A $27.3 billion figure for Meta data-center-related debt appeared in the source headline, although the available text does not provide the methodology, reporting date or detailed composition of that measure.

The distinction matters as hyperscaler AI build-outs increasingly rely on a mix of direct capital expenditure, long-term supply and construction contracts, future lease commitments, and structured financing. Large commitments can signal expected demand for data-center infrastructure, but they should not be read as evidence that capacity has been delivered or that the corresponding CAPEX has already been incurred. Conversely, focusing only on recognised balance-sheet debt may understate exposure to long-dated commitments and contingent guarantees.

For market participants, the analysis suggests that credit and demand assessments should separately evaluate recognised debt, committed future payments, lease obligations not yet commenced, and off-balance-sheet guarantees. The data available does not support combining these categories into a single measure of current leverage.