AI Cost Watch - Week of 29 July 2026 (Issue 6)

Read full text
Authors: N Milton
Cited by: 4
References: 2

Abstract

AI Cost Watch is a recurring, dated note tracking the unit economics of AI infrastructure: whether the buildout's cost and demand assumptions are holding. Issue 6 (week of 29 July 2026) reports the first quarter in which the series' pre-registered signal could be tested, and it did not fire: no hyperscaler conceded a demand-led cut to capital-expenditure guidance. Alphabet raised its 2026 guide to 195-205 billion dollars, Meta narrowed upward to 130-145 billion, and Microsoft held its calendar-2026 plan. What moved instead was the market's willingness to fund a raise. Microsoft and Meta reported on the same evening and the market separated them by double digits, Microsoft rising about 7 percent after hours against Meta falling 9.64 percent, sorted on whether contracted demand visibly covers the spend rather than on its size. The issue documents three findings from primary filings: Microsoft's lower headline capex figure reflects a reclassification of future data-centre leases from finance to operating rather than a spending cut; the 25-year asset useful life reported from the earnings call does not appear in the FY2026 Form 10-K, which states buildings at five to 15 years with no change-in-accounting-estimate disclosure; and Microsoft's filed contractual obligations total 743.8 billion dollars, including 443.5 billion of leases, against 46.1 billion of long-term debt principal. The issue also withdraws a placement made in Issue 5, which put Meta at the self-funded end of the series' financing gradient. Rates, a China memory supply shock and oil are documented as amplifiers and kept separate from the signal.

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 July 2026