Meta

You extended server useful lives to 5.5 years at exactly the moment Amazon shortened a subset citing AI obsolescence. Who's right?

Amazon's FY2025 10-K explicitly names "artificial intelligence and machine learning" as the reason for its subset shortening, citing "increased pace of technology development." META's FY2025 10-K does not explain why the extension to 5.5 years is appropriate versus the prior four-to-five-year range — it simply calls it a completed assessment. The two disclosures are irreconcilable; they cannot both be right for the same hardware generation at the same time. Amazon's explanation is more specific and attributable; META's is a change in estimate without stated justification. (Sources: Amazon FY2025 10-K; Meta FY2024 10-K; Meta FY2025 10-K — all PRIMARY.)

Reality Labs lost $19.19 billion in FY2025 and you expect $19 billion again in FY2026. At what point does the RL investment become too expensive to continue?

The FY2025 10-K states: "we expect our 2026 Reality Labs operating losses to remain similar to 2025." META does not provide a profitability timeline or breakeven target for RL in its filings. Total RL operating losses from FY2023 to FY2025 are ~$52.4B ($16.12B + $17.73B + $19.19B), with $2.21B in cumulative RL revenue over that period — a 96% loss ratio. (Source: Meta FY2024 10-K, Meta FY2025 10-K — PRIMARY.)

ARPP grew 27% in Q1 2026. How much of that is AI-driven ad improvement vs. the cyclical ad market recovery?

META's filed documents disclose that ad impressions grew 12% and average price per ad grew 9% in FY2025 (source: FY2025 10-K — PRIMARY). The price increase is cited as being "due to increases in ad impressions delivered and average price per ad." The specific contribution of AI ad targeting vs. cyclical recovery is NOT DISCLOSED and NOT SOURCED. Architect must flag that ARPP growth cannot be disaggregated from filings alone.

The $14.72 billion contingent cloud capacity commitment as of March 2026 — who is the cloud provider and why does a company spending $69 billion on its own data centers need to buy capacity from someone else?

META's Q1 2026 10-Q discloses "$14.72 billion of cloud capacity over a five-year period" as a contingent obligation but does not name the provider (source: META Q1 2026 10-Q — PRIMARY). META's FY2025 10-K directs approximately 82% of capex to FoA and 18% to RL — the external cloud capacity may represent AI training overflow or specific model workloads not suited to META's infrastructure. Provider identity: NOT SOURCED.

Your FY2025 reported net income was $60.46 billion — below FY2024's $62.36 billion despite 22% revenue growth. If you remove the $2.59 billion useful-life extension benefit, net income falls to approximately $57.87 billion — a 7.2% decline YoY. Is META actually in an earnings compression cycle masked by accounting?

The math is correct from primary filings: FY2025 net income ($60.46B per XBRL) minus $2.59B extension benefit = $57.87B underlying — versus $62.36B in FY2024. That is a 7.2% decline in underlying net income vs. reported 2.4% decline. R&D grew $13.5B (+31%) in FY2025; RL losses grew $1.46B. The accounting extension absorbed approximately 18% of the earnings compression. (Sources: Meta FY2025 10-K, XBRL CIK 0001326801 — PRIMARY.)