My key takeaways from the $META (Meta Platforms) Q3 2026 earnings call. 1. Muse's cost per user is undisclosed. "We don't have specific numbers to share around the cost structure of Muse today." The free tier runs on the smaller Muse Spark model, and Muse Wonder, the better frontier model, sits behind the paid tier because it costs more to serve. The entire profitability case rests on a 10x improvement in compute cost efficiency over the next year (off-peak capacity, smaller models, inference optimization; inference meaning the cost of running the model each time someone uses it). That 10x is a target with no baseline attached. 2. Free cash flow went negative. Capex was $34.9B in the quarter and free cash flow (cash left after capital spending) was -$2.1B. Cash and securities sit at $82.4B. Full-year 2026 capex is now $135B to $145B, with the low end raised $5B. Total expenses grew 68% to $51.6B, driven by legal costs, infrastructure, third-party cloud, and third-party AI token costs. No 2027 capex or expense numbers until the Q4 call. 3. The new share class is a funding tool. Meta is creating non-voting stock, distributed one-for-one to existing holders (effectively a 2-for-1 split), trading from late January 2027. Stated uses: capital raising, employee compensation, acquisitions. No specific financing plans yet. My read: a company with negative free cash flow just built itself a currency to sell equity or pay for deals without diluting Zuckerberg's voting control. 4. Core ads carried the quarter. Ad revenue $63.1B, up 26%. Impressions up 10%, average price per ad up 14%, with part of the pricing lift credited to easier macro conditions versus Q3 last year. Reported operating income fell 37% to $13B (20% margin) because of a $9.9B multistate Attorneys General settlement charge. Strip that out and operating income grew 11%. Headcount is 66,000, down 12% from Q2 after the May layoffs. 5. Muse demand is strong, and global rollout is gated by compute. 10M+ people in the U.S. and Canada set it up in about six weeks. Roughly 25% of them use it on a given day. Shopping is the top use case at over 25% of weekly actives, average purchase size above $200 (travel included), and Stripe had to raise credit limits in week one. The caveat: these are set-ups and engagement, not paying users, and only two countries have it because serving more people costs too much right now. 6. Monetization starts earlier than the usual Meta playbook. Historically Meta grew a surface first and monetized years later. Muse costs too much to wait. The model: free for most users, a fee on transactions paid by businesses, subscriptions for top models and extra tokens, and paid business tiers. The enterprise side (model API, Muse Code, business agents) is growing "quite quickly" with zero revenue disclosed. New AI businesses begin contributing in 2027, but the meaningful scaling is pushed further out. 7. The quiet story is inside the ad machine. Partnership ads (brands running ads with creators) grew 97% YoY to a $15B+ annual revenue run rate, meaning the current quarter's pace annualized. Moving content-understanding work onto in-house Muse models cut per-unit processing costs by 80%+ in offline tests. That is the cleanest proof so far that owning the models pays: lower cost inside a business already printing money. Family of Apps other revenue grew 59% to $1.1B on WhatsApp paid messaging and subscriptions. 8. Reality Labs is being rationed. Revenue fell 6% to $443M on weaker Quest sales, partly offset by AI glasses growth. New VR devices now ship only on major breakthroughs. Losses guided to $18B to $19B in 2026 and about $15B in 2027, with budget reallocated primarily to AI. A cut, framed as focus. 9. Legal risk is still open. Beyond the $9.9B charge, youth-related trials are scheduled in the U.S. in the coming months, and scrutiny is ongoing in several markets. The quarter's 24% tax rate came in above outlook because part of the legal expense isn't deductible. Bottom line: Management's story is that Muse is Meta's next billion-user platform and the ad business pays for the buildout. Half of that is proven: ads grew 26% with underlying operating income up 11%, and Muse has real early demand. The other half is not: no cost per user, no Muse revenue number, no 2027 capex, and the 10x efficiency gain is a forecast. Negative free cash flow plus a new non-voting share class moves the question from whether Meta can afford this to how it will pay for it. Owning $META here means betting the cost curve bends before the funding gap forces dilution.