Goldman finds $4.1T in AI commitments Big Tech keeps off its books

Goldman Sachs just put a price tag on Big Tech's AI building spree, and it's a jaw-dropper. A report from credit strategist Amanda Lynam estimates hyperscalers are sitting on $3.6 trillion in lease and purchase commitments. Fold in Nvidia, Broadcom and SpaceX, and the total off-balance-sheet pile climbs to about $4.11 trillion.

The money splits into two buckets. Roughly $1.7 trillion rides on lease obligations, while about $1.9 trillion covers other purchase commitments. The five hyperscalers in the crosshairs are Alphabet, Amazon, Meta, Microsoft and Oracle. More than 75% of those leases haven't begun yet.

Lease obligations across those five have ballooned eightfold since 2023. Morgan Stanley pegged the broader off-balance-sheet figure near $3.12 trillion at the end of August 2026, so Goldman's newer tally implies about $992 billion piled on in just a few months. Meanwhile, the debt and lease liabilities actually sitting on the balance sheets total around $1 trillion — under a quarter of what's tucked out of sight.

Nobody is accusing these companies of hiding anything. Under standard accounting, commitments don't register as liabilities until they kick in. Goldman flags where that could bite: future cash flows, leverage metrics and how ratings firms might start grading these companies.

The backdrop is a financing boom. AI-related debt issuance in 2026 has already blown past earlier full-year forecasts, according to the research. Private markets are funding a bigger slice of data center and chip projects, with special purpose vehicles doing more of the lifting. Syndicated bond deals arrive with public paperwork; private arrangements are far harder for outsiders to track.

So keep an eye on that 75%. As those idle leases commence, they migrate onto balance sheets and into quarterly filings, making the mountain of commitments much harder to miss. Nvidia and Broadcom landing in the tally is a reminder that this exposure runs in every direction across the supply chain.