One clearly RED indicator; one AMBER–RED: - Indicator 1 (depreciation, RED): $2.92B useful-life extension = $1.00/diluted share boost in a year when net income declined. Direction is wrong (extended when AI hardware may depreciate faster). Pure earnings-inflation accounting. - Indicator 2 (capex gap, AMBER–RED): $69.69B FY2025 capex (+87% YoY) includes $19B+ RL annual losses with no monetizatio...
Last updated:
Human
Zuckerberg told his own employees the layoffs were about capex, not AI productivity. 8,000 people cut to help fund $135B of compute and $100M-a-head researchers. Did AI do this, or did we? He said it himself: we did.
Meta is the cleanest verdict in the set, because the CEO removed the ambiguity. The public framing was an 'AI efficiency push'; internally, Zuckerberg told staff the ~8,000 layoffs were about capex — funding a $125–145B AI buildout that is collapsing the company's free cash flow. Meanwhile Meta is paying individual AI researchers up to $100M–$1.5B. This dossier tracks a reallocation the company itself admitted: from the many, to the few and the machines.
[Claims][Filings / Data]
The ~8,000 layoffs (~10% of staff) are an 'AI efficiency push.'
— Meta, public framing, April 2026
Zuckerberg told employees in a town hall the layoffs were about CAPEX, not AI productivity — the cuts fund the buildout. 2026 capex is ~$125–145B (roughly double 2025's $72.2B), and free cash flow is projected to fall from $43.6B (2025) to ~$8.5B (2026) — an ~80% collapse — as AI capex absorbs operating cash (~$315–370M/day on infrastructure).
SEC 10-K/10-Q (META) capex + cash flow; TheNextWeb town-hall reporting 2026
Meta is becoming leaner and more efficient with AI.
— Meta restructuring narrative, 2026
Simultaneously: $14.3B for 49% of Scale AI + Alexandr Wang as chief AI officer (June 2025), and individual researcher packages of $100M+ (Ruoming Pang ~$200M from Apple; Andrew Tulloch offered ~$1.5B/6yr; Deitke ~$250M/4yr). The payroll was not saved by AI — it was redirected from 8,000 rank-and-file to a handful of stars and to compute.
Fortune 2025-07-11; Axios 2025-07-03; Entrepreneur 2025
The superintelligence bet is worth the spend.
— Meta / Zuckerberg (supervoting control — answers to no one)
FCF down ~80%; gigawatt-scale data centers (Hyperion, Louisiana — $10B, 5GW; Prometheus, Ohio — 1GW). The FCF collapse IS the Bubble Watch capex-vs-cash-flow read for one of its largest names.
SEC filings (META); see The Report / Bubble Watch
The Three Inversions
The Efficiency Lie Sourced
The rare case where the accused confesses. 'AI efficiency' was the press release; Zuckerberg told his own staff the layoffs were about capex, not AI productivity. The 8,000 who left did not lose their jobs to a model — they lost them to a budget decision to fund $135B of compute and pay individual researchers more than the severance of thousands. The machine is the story; the CEO named the real cause himself.
The Safety Ruse Sourced
Two moves at once. Meta frames open-weight Llama as 'democratizing AI' — which also undercuts rivals' paid-API pricing (the Power map). Meanwhile it spent $4.6M+ lobbying state officials in 2026 (its most ever), ran a public campaign against state AI laws, and backs the federal-preemption push that favors scaled incumbents. Openness as principle; regulatory defeat as practice. Honest nuance: the industry is split — Anthropic argues governments should be able to block dangerous AI. — see full analysis →
The Energy Myth Sourced
Meta pledged net-zero by 2030; its emissions jumped 64% year-over-year, and its purchased-electricity emissions rose nearly 150% from 2019 to 2024. The buildout is gigawatt-scale (Hyperion 5GW, Prometheus 1GW) and, like its peers, increasingly gas-powered (Louisiana). The pledge and the power draw are moving apart — and the grid and water cost lands on communities that never voted for the data center. See The Power Draw. — see full analysis →