Share
Goldman Sachs has put a number on Big Tech’s AI building spree, and it is a big one. The bank estimates that hyperscalers carry $3.6 trillion in lease and purchase commitments.
Add Nvidia, Broadcom and SpaceX to the tally, and total off-balance-sheet commitments climb to about $4.11 trillion. Most of that money has not been spent yet, which is exactly why credit analysts are paying attention.
What Goldman actually counted
The estimate comes from a report by Goldman Sachs credit strategist Amanda Lynam. It focuses on five hyperscale tech companies: Alphabet, Amazon, Meta, Microsoft and Oracle.
Those commitments split into two buckets. Roughly $1.7 trillion is tied to lease obligations, and about $1.9 trillion covers other purchase commitments.
More than 75% of the hyperscalers’ lease obligations have not yet begun, according to the report.
The leases mostly cover data centers, infrastructure and equipment that the AI industry needs to run. The purchase commitments, meanwhile, include obligations due by the end of 2027 that total in the hundreds of billions.
The growth rate is the eye-catching part. Lease obligations for the five hyperscalers have increased eightfold since 2023.
For comparison, a Morgan Stanley estimate put the broader off-balance-sheet figure at roughly $3.12 trillion at the end of August 2026. Goldman’s newer $4.11 trillion tally implies an increase of about $992 billion in just a few months.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
We respect your privacy. Unsubscribe anytime.
The gap between what’s owed and what’s reported
The most important number in the report may be the smallest one. On-balance-sheet debt and lease liabilities for these companies total around $1 trillion.
Put differently, the obligations investors can see on the balance sheet are less than a quarter of the ones they can’t. Off-balance-sheet exposures exceed reported leverage by more than four times.
None of this means the companies are hiding anything. Commitments that have not started yet simply do not show up as liabilities under standard accounting until they kick in.
Goldman’s research flags several areas where this could matter: future cash flows, leverage metrics and potential shifts in how ratings firms evaluate these companies.
How AI got this expensive
The commitments sit inside a wider surge in AI financing. AI-related debt issuance in 2026 has already surpassed earlier full-year forecasts, according to the research.
Where that money comes from is shifting too. Private markets are financing a larger share of data center and chip projects, with special purpose vehicles playing a more prominent role.
That changes the transparency equation. Syndicated bond deals come with public disclosures, while private financing arrangements tend to be harder for outside investors to track.
What this means for investors and lenders
The 75% of leases that have yet to commence is the figure to watch. As those obligations start, they will migrate onto balance sheets and into reported liabilities, making the commitments far more visible in quarterly filings.
Nvidia and Broadcom appearing in the broader tally shows the commitments flow in multiple directions across the supply chain.
The growing role of private credit and special purpose vehicles adds another layer. These structures can spread risk across more investors, but they can also make it harder to see where exposure is concentrated if conditions change.
