"Meta Earnings, Meta's Timing Problems, The Financial Tail" — @BenThompson

Why this is in the vault

Direct capex-anchor update for the RDCO chip-fab/memory capital-cycle thesis: Thompson turns from bull ("A Script for Mark Zuckerberg," 2026-07-07) to alarmed, flagging that Meta's committed-spend math ($145B this-year AI capex, ~$700B total committed, $349.3B non-cancelable contracts, $347B in leases-not-yet-started with $68B added in July alone) is now outrunning any credible near-term monetization story.

The core argument

Three linked pieces in this Update (Thompson just back from vacation, catching up on an earnings backlog).

Meta earnings. Q3 guide disappointed and the stock fell about 8% to $539.03. Thompson's recurring diagnostic — impressions growth vs. price-per-ad growth both rising together — held last quarter but is less extraordinary this quarter, while expenses rose 55% against 28% revenue growth (with capex depreciation not even fully phased in, and some expense growth possibly non-recurring legal charges). He came away "a bit alarmed."

Meta's timing problems. Three distinct mismatches: (1) a cash-vs-accounting timing gap — capex is paid now, expensed later via depreciation, and Meta is layering on enormous lease liabilities to bridge it, effectively double-paying (renting third-party compute while building its own future data centers) with no cloud business yet to absorb the excess; (2) a strategic timing problem — Zuckerberg isn't worried about ads, he's worried about existential AI risk, and argues Meta's full-stack ownership (data centers, chips, infra) is required for "model sovereignty," a framing Thompson has historically endorsed; (3) a competitive timing problem — Anthropic and OpenAI likely hold a widening structural inference-cost advantage (per Thompson's earlier "Who's Afraid of Chinese Models"), so Meta's future-payoff story rests on infrastructure the frontier labs already built years ago. Thompson also skewers Zuckerberg's "data flywheel" argument as an echo of Microsoft's "billion Windows PCs" complacency right before mobile disruption — trajectory, not scale, is what matters.

The financial tail. Thompson reads Zuckerberg's and CFO Susan Li's pitch of enterprise APIs, business agents, and compute resale as "the financial tail wagging the dog" — the committed spend is forcing management to invent monetization narratives, several of which don't hold up, since Meta has no enterprise go-to-market track record and (per Thompson) won't be a credible enterprise player for at least a decade, comparing unfavorably to how long Google's cloud pivot took. He notes the irony that Zuckerberg's own consumer-scale framing undercuts the enterprise pitch. Closes with a genuine recommendation of Zuckerberg's WSJ op-ed "The AI Future is for Everyone."

Mapping against Ray Data Co

This directly updates the memory/chip-fab capital-cycle thesis tracked in investing-edgar-watch and the smart-money surveillance loop: Thompson is now flagging demand-side fragility at exactly the hyperscaler whose capex commitment ($700B total, $349.3B non-cancelable) has been read as evidence the buildout is durable rather than bubble-shaped. This is the same author reversing tone in one month — his 2026-07-07 piece built the bull case for Meta's capex as structurally defensible via ad-inventory expansion; this piece flags that the accounting math (expenses +55% vs. revenue +28%, cash-vs-depreciation lag, lease liabilities front-loading) is now the visible strain point. Treat as a data point for the Phase 2→3 capital-cycle transition, not a reversal of the thesis — worth cross-referencing against the next hyperscaler-capex quarterly pulse for whether Amazon/Google/Microsoft show the same expense/revenue divergence or whether Meta is an outlier.

Related

• [[2026-07-07-stratechery-meta-earnings-script]] — the bull-case predecessor this piece implicitly walks back from
• [[2026-05-04-stratechery-google-meta-earnings]] — prior joint earnings read, useful baseline for Meta's expense trajectory

Sponsorship

No paid sponsor in this issue. Self-promo only: Sharp Tech and Dithering episode plugs in the intro, and a standard closing cross-promo block for the Stratechery Plus podcast stable (Sharp Tech, Dithering, Sharp China, Greatest Of All Talk, Asianometry) plus a team/group-subscription pitch. No bias implication beyond ordinary in-house cross-promotion.