AI Data Center Debt: Big Tech's Hidden $3 Trillion
$3T AI Datacenter Debt 80% Off The Books Watch on YouTube, 4:25The full breakdown runs 4:25. The written version below covers the 4 numbers.
There is roughly $3 trillion of committed AI spending across Alphabet, Amazon, Meta, and Microsoft, and about 80% of it does not appear on any balance sheet. The Wall Street Journal put the number on it this morning. This page walks the 4 numbers from smallest to biggest, the same way the video does.
Committed AI spending
Off balance sheet
In the footnotes
The 4 numbers
- $248B in lease liabilities, on the books. Data centers they lease and currently operate.
- $356B in long-term debt, on the books. Actual bonds. With the leases, roughly $600B that Wall Street can see and model.
- $904B in signed leases for data centers that are not operating yet, off the books. A lease does not count until the building starts operating, so a 15-year commitment for a data center that opens in 2028 is a footnote today.
- $1.5T in purchase commitments for chips, compute, and energy, off the books. For scale, the entire 2007 subprime mortgage market was about $1.3 trillion. This one line item is bigger.
Why the richest companies on earth keep it off the books
None of this is fraud. It is structure. Keep the commitments in the footnotes and earnings stay clean, credit ratings stay pristine, and the AI growth story keeps its multiple. Meta already showed the playbook: its biggest data center is financed through a joint venture, so the debt lives somewhere else. The spending is real and the obligations are binding, but the balance sheet stays photogenic.
Who eats it if AI revenue disappoints
The risk moved rather than vanished: to the landlords, the joint ventures, and the private credit funds writing these deals. If AI revenue shows up, everyone grows into it. If it disappoints, the commitments surface as impairments, renegotiations, and somebody's bad quarter, and the place they surface first is the credit market. The one number to watch is high yield spreads: the standing rule on this channel is that a break above 400 gets a same-day video. The mechanics are in The AI Credit Crunch: High Yield Spreads Explained.
Source: The Wall Street Journal, "Why Big Tech's AI Spending Is $3 Trillion Higher Than It Seems," Aug 17, 2026. Figures for Alphabet, Amazon, Meta, and Microsoft as reported. Full sourcing in the video description.