Meta Platforms, Inc. (META)

News Summary

• Meta's Muse agentic AI reached 10+ million U.S./Canada setups in roughly six weeks with a 25% daily active rate among registered users; shopping is the leading use case with average purchase size above $200, and Zuckerberg confirmed compute capacity — not demand — is the binding constraint on global rollout, making inference infrastructure the critical path to user growth.
• Meta disclosed that migrating content understanding models to its first-party Muse model family cut per-unit content processing costs by more than 80% across Instagram, Threads, Facebook, and ads; separately, management committed to a greater-than-10x improvement in Muse inference cost efficiency within one year via model distillation, off-peak capacity utilization, and inference optimization — a direct signal of the scale and urgency of the inference workload Meta is trying to solve.
• Full-year 2026 capex guidance was narrowed and raised to $135–145B (from $130–145B prior), with management explicitly flagging 'aggressive' continued AI infrastructure investment into 2027; the four major hyperscalers combined are tracking roughly $730B in capex this year, with analyst consensus projecting the group approaching $1T in 2027, led by Alphabet at over $310B — confirming a sustained, multi-year AI hardware demand cycle.
• Q3 revenue of $64.6B rose 26% YoY and beat consensus by roughly $1.3B; adjusted EPS of $7.40 beat the FactSet consensus of $6.54 by $0.86, though GAAP EPS of $3.83 was depressed by a $9.9B legal charge; Q3 capex was $34.9B and free cash flow was negative $2.1B; Q4 revenue guidance of $73–76B ($74.5B midpoint) came in approximately 1.2% above consensus, and operating margin is expected to recover as the one-time legal charge does not recur.
• META rose approximately 3–5% in after-hours trading following the print; retail sentiment on Stocktwits was bullish with high message volume; Baird noted increased positive investor sentiment around Muse and said capex raises could now be received more positively given clearer AI ROI, while Bank of America and Morgan Stanley both flagged that Muse monetization remains a multi-year build and will consume compute capacity without near-term revenue contribution.

Financial Highlights

Earnings Call Highlights

Meta's Q3 2026 call was dominated by the launch and early traction of Muse, its personal AI agent, which management positioned as the company's defining strategic bet and a direct embodiment of its 'personal superintelligence' mission. Zuckerberg was visibly energized and confident, framing Muse not merely as a product but as the nucleus of a future agent network at billion-person scale, while Susan Li provided the financial scaffolding around compute costs, monetization sequencing, and capital structure flexibility. Analyst questions were sharply focused on the unit economics of serving Muse at scale, the friction points in agentic commerce, and how quickly Meta can bend the cost curve, signaling that the Street is treating this as a credible but capital-intensive inflection. The tone was bullish but grounded in acknowledged constraints around compute intensity, global rollout pacing, and the need to build monetization in parallel with adoption. CapEx guidance of $135–145B for 2026 and explicit commentary about 'aggressive' 2027 investment underscored that Meta is in full infrastructure build-out mode.

• Meta guided full-year 2026 CapEx to $135–145B (narrowed from $130–145B), driven by servers, data centers, and network infrastructure, with management explicitly stating they expect to 'continue investing aggressively' in AI infrastructure into 2027, signaling sustained hyperscaler-level demand for AI compute.
• Muse is described as 'very compute intensive to run,' but Meta committed to improving cost efficiency by more than 10x within one year through better utilization of off-peak capacity, more efficient models, and inference optimization — a direct signal of the inference optimization workload and the scale of GPU utilization challenges they are actively solving.
• Zuckerberg confirmed that compute capacity constraints, not demand, are the primary limiter on Muse's global rollout: 'We have not launched it globally yet... partially that we wanna make sure that we do this efficiently,' implying that inference infrastructure scaling is the critical path to user growth.
• Meta disclosed it is moving content understanding models onto its first-party Muse model family, with offline evaluations showing more than 80% reduction in per-unit content processing costs across Instagram, Threads, Facebook, and ads — a concrete data point on the cost leverage of owning the full model stack.
• Zuckerberg articulated a clear model efficiency roadmap: 'I just expect that we will be able to deliver the same intelligence that we have now in MuseWonder... at a fraction of the cost in a smaller model a year from today,' framing continued distillation of frontier models into efficient product models as a core strategic priority alongside scaling training runs.

Sources

• Meta Platforms Q3 Earnings Call Highlights - Yahoo Finance
• Meta Platforms, Inc. Q3 2026 Earnings: Live Updates of $META Earnings ...