The five biggest US tech companies now carry about $3.2 trillion in AI-related commitments that do not appear as debt on their balance sheets, according to a Nikkei Asia analysis published on 4 November 2026. That is nearly double the roughly $1.65 trillion Nikkei counted in July. The figure is an estimate, and it measures promises to pay in the future rather than money already spent, but it explains why investors have become nervous about the scale of the AI build-out.
What the number covers
Nikkei's tally covers Alphabet, Microsoft, Amazon, Meta and Oracle as of the end of September. According to the report, about 60% is long-term purchase commitments for chips, servers, power and data-centre capacity, and about 40% is lease obligations for data centres and other facilities that have not yet begun. These commitments are disclosed in the footnotes of the companies' filings with the US Securities and Exchange Commission, not on the balance sheet that many investors read first. Under accounting rules, a lease generally only appears as a liability once it starts, even when the company has signed a binding agreement. One summary of Nikkei's work says the total equals roughly 80% of the five companies' combined assets.
A second estimate says the same
Goldman Sachs credit strategist Amanda Lynam put out a note on 3 November reaching a similar total of about $3.2 trillion. Her breakdown, as reported by Yahoo Finance and others, is roughly $1.3 trillion in leases not yet started and about $1.9 trillion in other purchase commitments. Lynam said some of these liabilities may already be reflected in analysts' spending forecasts, but that the category is growing fast and represents an additional use of cash.
Caveats worth knowing
- These are compiled estimates, not a single disclosed number, and Nikkei has not published a precise methodology.
- Other tallies differ. One firm, Kalkine, counts all off-balance-sheet commitments and gets a higher figure, and company-by-company numbers vary between sources, probably because of different definitions rather than errors.
- Commitments are not the same as cash already spent. They are payments owed in future years, some stretching over decades.
- Off-balance-sheet does not mean improper. Purchase and lease commitments are ordinary, disclosed business arrangements, and comparisons to past accounting scandals do not fit the structure.
Why it matters
The obligations must be paid whether or not the AI revenue they are meant to support arrives. That is the concern analysts keep returning to. It sits alongside other signs of strain, including Meta's $25 billion bond sale this week and the sell-off in AI infrastructure shares after the US midterm elections. Supporters of the spending argue demand for AI computing is real and growing, and that the companies' cash flows can carry the load. Sceptics say the sums are now so large that a slowdown in demand could leave the companies with expensive capacity they do not need.
For readers outside the US, including governments and businesses across Africa that buy cloud and AI services from these firms, the practical point is that the prices and availability of AI services depend on whether these companies can keep financing their build-out. This is a news report and not investment advice.
Sources: Nikkei Asia (4 November 2026); Goldman Sachs note via Yahoo Finance and TokenPost; Telecompaper; Newsquawk; AI Weekly; DEV Community; AI Daily Digest.
