--- title: "META (Trans): Muse’s Long-Term Goal Is a Billion Users; Burn compute first, Monetize later" type: "Topics" locale: "en" url: "https://longbridge.com/en/dolphin/post/44471684.md" description: "Muse has amassed over 10 mn registered users in the US and Canada roughly six weeks post launch. DAU penetration is ~25%, and Q4 revenue guidance is $73–76 bn.However, Muse's compute costs remain elevated, and management targets >10x improvement in compute efficiency over the next year to reach profitability as revenue scales. The company is pivoting its focus to personal superintelligence and plans to monetize earlier than in prior cycles." datetime: "2026-10-21T23:12:13.000Z" locales: - [en](https://longbridge.com/en/dolphin/post/44471684.md) - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/44471684.md) - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/44471684.md) author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)" generator: "portal-rs" --- # META (Trans): Muse’s Long-Term Goal Is a Billion Users; Burn compute first, Monetize later **Dolphin Research compiled the following** $Meta Platforms(META.US) **FY26 Q3 earnings call transcript** **I. Results recap: key takeaways** 1\. Shareholder returns and capital structure: Meta will create a new class of non-voting shares and distribute them as a stock dividend, one new share for each existing share. This is equivalent to a 2-for-1 split, with trading expected to begin in late Jan 2027. 2\. Guidance. Details follow. a. Q4: total revenue of $73–76 bn. FX at current rates is a ~1% drag on YoY growth, and the effective tax rate is 15%–17%. b. FY26: total expenses raised to $175–179 bn to include $9.9 bn of legal expense recognized in Q3. Ex-legal, operating income is still expected to be higher in 2026 vs. 2025. c. FY26: capex (incl. finance lease principal) narrowed to $135–145 bn (prior $130–145 bn). Reality Labs (RL) losses to narrow to $18–19 bn. d. 2027: no full-year revenue guide, but management expects strong revenue growth. Meta will continue to invest aggressively in core biz. and new AI initiatives, with budget and capacity plans to be disclosed on the Q4 call; RL losses to narrow to ~$15 bn and trend toward profitability thereafter. 3\. Key financial metrics this quarter. See below. a. Revenue: total revenue of $64.6 bn (+26% YoY). Family of Apps revenue was $64.2 bn, incl. ads at $63.1 bn (+26% YoY, constant FX similar). b. Other and RL: other revenue $1.1 bn (+59%, driven by WhatsApp paid messaging and subscriptions). RL revenue $443 mn (−6%, weaker Quest unit sales partly offset by AI glasses). c. Volume and pricing: ad impressions +10%, average price per ad +14%. Gains came from better ad efficacy and macro, partly offset by faster impression growth in lower-monetizing regions/surfaces. d. Expenses: total expenses $51.6 bn (+68%), incl. $9.9 bn related to the Aug 2026 multi-state AG settlement. Additional cost drivers were legal, infra and third-party AI token costs. e. Profit: OP $13.0 bn (−37% YoY; OPM 20%); ex-legal, OP +11% YoY. Tax rate was 24%, above guidance as a portion of the legal expense was non-deductible; NP $9.9 bn and EPS $3.83. f. Cash and headcount: capex $34.9 bn; FCF −$2.1 bn. Cash and marketable securities at period-end were $82.4 bn, debt $83.7 bn; headcount exceeded 66k, down ~12% QoQ post-May 2026 layoffs. 4\. Risk: scrutiny over teen-related issues continues in multiple markets. Several teen-related cases in the U.S. go to trial in coming months, which could have a material impact on biz. and financials. **II. Earnings call details** **2.1 Exec remarks: core messages** 1\. Users and content recommendations. Highlights below. a. Family DAP reached 3.64 bn; engagement hit new highs on both Facebook and Instagram. Threads MAU has surpassed X, with time spent per user up over 20% YoY. b. Instagram time spent per user hit a record with double-digit YoY growth. Facebook Reels time spent grew double-digit YoY globally and in the U.S. c. Relevance improvements to Instagram Reels lifted time spent by over 4%, with the new architecture ranking by sequences rather than single items. Nearly 45% of recommended Reels were posted within the past day, up over 11 ppt QoQ. d. Facebook completed the global rollout of its generative recall architecture, covering ~1/4 of non-friend Feed browsing and half of Reels browsing, lifting Reels watch time by 1%. Ranking models doubled usable history, adding another 1% to U.S. Reels watch time. e. Content understanding models migrated to the in-house Muse model, which outperformed the prior model in offline evals. Unit content processing cost fell by over 80%. 2\. Ads and monetization. Progress continues. a. The Meta generative recommender launched generative recall in Facebook Feed, improving target user conversions by nearly 2%. GEM expanded to more surfaces, lifting offsite conversions by 1.7% and app ads conversions by 9.6%; Andromeda optimizations boosted Facebook conversions by 1.4%. b. Advertisers adopting Meta AI business assistant recommendations saw nearly 5% more conversions at the same CPA, and spent 9% more. This underscores early ROI gains. c. Nearly 10 mn advertisers used gen-AI creative tools in Q3, with video generation adoption up 25% QoQ. Adoption momentum remains solid. d. Partner ads grew over 97% YoY, with annualized revenue topping $15 bn. Creator Marketing Hub launched globally, with more than 10 mn creators searchable. e. The unified subscription service Meta One is expanding globally with 50+ features, and combined subs and trials surpassed 15 mn. Core experiences remain free. 3\. Muse personal Agent. Scaling underway. a. In roughly six weeks, over 10 mn users in the U.S. and Canada set up Muse, and it has ranked near the top of the App Store almost daily. About 25% of set-up users use it daily, more than 2x Threads at a similar stage, with retention improving as time spent rises. b. Shopping is the top use case, with over 25% of WAU participating and an Avg. order value above $200. Stripe increased credit limits in week one; productivity and creation are the next-largest use cases. c. Muse Secure VM is available, and Muse Confidential VM is coming soon, which prevents Meta from accessing user content. Muse Wonder further improves quality, engagement and retention. d. Entering scale-up mode, management set three priorities: expand product, grow revenue and cut compute cost. The goal is to serve billions and build the Muse network. e. Biz. model: most users will be free, with merchants paying based on transactions facilitated. Premium models/extra tokens, merchant value-added features and enterprise offerings will be subscription-based, and can plug into content and ad personalization; current Muse users in the U.S. generate Family ARPU of several hundred dollars annually, and Muse is expected to be higher long term. f. Compute: over the next few years Meta will build capacity to serve 1 bn+ users. Over the next year, compute cost efficiency is expected to improve by over 10x via idle-capacity utilization, more efficient models and inference optimizations; at that point, revenue scale could drive profitability, though total service costs will still rise. g. At least 500 mn self-identified small businesses use the Family apps. Muse can help them acquire customers and improve efficiency. 4\. Models and enterprise services. Roadmap and go-to-market. a. Muse Wonder is the first frontier model; since the Apr launch of the new models, per-user engagement with Meta AI has more than doubled. Meta has released Muse Spark 1.2/1.3, Muse Code, open-weight Muse Glimmer and Muse Spark Open, and Muse Voice. b. The path is to push the frontier while distilling to smaller models for personal Agents. Larger models are training, while small models will deliver Muse Wonder-level intelligence at higher efficiency. c. Training is a large fixed cost, so Meta formed the enterprise platform led by CJ Desai to capture high-margin inference demand via APIs and Muse Code. This aligns cost structure with monetization. d. Meta is using Muse Wonder to run evals that let models self-improve the recommendation and ad systems. This should accelerate future model improvements. 5\. Reality Labs and new products. Strategy updates. a. Muse will arrive on Meta glasses next month. Hardware will not be priced for high margins, with monetization to come from software and superintelligence services on-device; glasses shipments continue to grow rapidly, and paid subs are expected to increase. b. In VR, Meta will launch only upon major breakthroughs (e.g., this year's Meta VR Glasses) and slow investment in that area. This reflects a more focused cadence. c. Edits exceeded 50 mn active users within a year; Muse went from concept to product in about eight months. Muse Charm progressed from concept to hardware in under a year and is about to launch. **2.2 Q&A** **Q: What are the main constraints on Muse user growth?** A: **The biggest constraint is compute cost, and Muse has not launched globally as efficiency must improve first.** It is currently live only in a handful of markets such as the U.S. and Canada, with market-specific tuning and connector work also ongoing; if rolled out broadly, user numbers would be well above today. The company needs to ensure enough capacity to serve a projected 1 bn+ users, contingent on hitting defined efficiency targets, which is one constraint over the next year; monetization must progress in tandem with cost-down. Product value will improve as models iterate: the free tier defaults to Muse Spark, while Muse Wonder is currently paywalled and performs meaningfully better, and a more efficient Wonder is in the works for wider rollout. **Q: What is the vision and core value of Meta's enterprise services?** A: **Near term, the focus is model APIs, Muse Code and merchant Agents, and all three are growing quickly.** Longer term, Meta aims to build a Muse commercial network: automate rote tasks for SMBs, create a Muse version that can operate in enterprise environments, and enable personal Agents to transact directly with merchant Agents. The CFO added: AI is upending enterprise software, and incumbents have not locked in the market, with some current leaders not even existing a few years ago. Meta brings models, infra, global distribution and relationships with 500 mn merchants, and aims to build recurring revenue across subs, usage-based API and messaging, scaling investment as demand and PMF are validated over a multi-year opportunity. **Q: How will Meta address friction where aggregators or suppliers are reluctant to let Agents connect for agentic commerce?** A: **The vast majority of merchants are eager to integrate, with many inbound, and results are above expectations.** Different partners move at different speeds, and Meta will collaborate on terms they can accept. The CFO added: some merchants remain cautious to protect customer relationships; the market likely evolves from Agents emulating human web browsing to authorized, structured interactions, including authenticated APIs, permissioned integrations or direct personal-Agent-to-merchant-Agent links. Meta previewed an open standard for personal Agent protocols co-developed with Sierra, Walmart, Stripe, Shopify and others this month; adoption will be uneven and take time, and most near-term experiences will be AI-assisted rather than fully autonomous. **Q: Why create a non-voting share class, and how does it relate to capital sources and structure?** A: **The new class broadens financing tools, with no specific fundraising plan at this time.** It can be used for capital raising, employee comp and M&A, while preserving A/B shareholder voting rights and the founder-led structure; the goal is to pre-build the capital structure to access diversified funding for AI-driven investment opportunities. **Q: What could be unlocked as the market shifts to authorized, Agent-to-Agent transactions?** A: **Hard to predict precisely and the shift will be gradual and uneven, but it should enable more proactive transactions.** For example, when Muse places a call on behalf of a user, a merchant Agent answering would make it smoother; Muse is inherently proactive, finding alternate paths when blocked, running persistent searches, issuing periodic updates and recommending, say, 10 new books each month, so demand is not limited to latent intent surfacing, and merchant Agents behave similarly, creating incremental activity. **Q: Can you frame 2027 expenses and capex?** A: **No, budgets and capacity plans for 2027 are still early and will be updated on the Q4 call.** Management is evaluating many investment proposals with varied payback periods and aims to support as many as possible within overall budget and financial targets. **Q: How much headroom remains for ad system improvement from the new models?** A: **There is meaningful runway across three tracks, and the standard ranking/recommendation pipelines remain healthy.** First, the conventional ML optimization path supports continued core biz. growth through 2027; second, research shifting to LLM-based architectures to reason more deeply about user preferences will benefit both organic and ad recommendations; third, gen-AI ad tools, largely via Advantage+, will expand image-to-video, audio generation and other capabilities to make ads more personalized. **Q: Will strong 2027 revenue growth come from ads or new AI revenue?** A: **Primarily from core ads and engagement investments, with new AI revenue starting to contribute in 2027.** New AI businesses include subscriptions, transactions, ads and enterprise services, contributing incrementally from 2027 but scaling more meaningfully over the long term; most are still early, and the near-term focus is product-market fit at scale, with AI becoming an important source of revenue diversification over time. **Q: After Connect, has Meta reorganized around Muse? Any changes in biz. goals?** A: **Management reiterated qualitatively that a personal superintelligence remains the North Star, without specific goal changes disclosed.** The direction was set over a year ago, and the MSL reorg was to build in-house models for personal Agents and deliver end-to-end experiences, with a similar approach for glasses; these are long-term bets, and strong execution should translate into results. **Q: What does Muse need to do to serve small businesses?** A: **Muse for Small Business launched in late Sep as a set of new skills and connectors inside existing apps.** In a few steps it can connect Instagram account analytics, Facebook Pages, ad accounts and dozens of third-party tools like Shopify, QuickBooks and Canva, learn a merchant's catalog, brand voice and customer needs, analyze ad/content performance, draft marketing and customer service responses, and flag issues. Merchants use it to optimize ads, triage DMs, expand sales, update websites based on inventory and schedule staff, and Meta is iterating features accordingly. With touchpoints to 500 mn self-identified SMBs and multiple global channels, consumer-side investments are reusable, and AI opens the enterprise market to Meta. **Q: What does Muse's per-user cost structure look like, and how will the cost curve come down?** A: **No specific cost data disclosed, but compute is high and efficiency is the key to scale.** Near-term levers include shifting some tasks to idle/off-peak capacity, using smaller models and multiple inference efficiency optimizations, with GPU utilization as the core metric; longer term, as LLMs scale, tasks can complete with fewer tokens, and AI can optimize the software running the models, creating further cost-down. **Q: Will Muse monetize earlier than the traditional 'grow first, monetize later' approach?** A: **Yes, monetization will start earlier than usual.** The cost and opportunity are both large, so monetization is needed to support scale, and much user behavior is inherently commerce, allowing Meta to take a cut of value created to fund compute buildout. On costs, within about a year smaller models could deliver Muse Wonder or current frontier-level intelligence at a fraction of today's cost; users may then expect more than today's level, so revenue, models and efficiency must advance together. Meta will not put a paywall in front of everyone to build the Muse network quickly, as scale drives collaboration value. **Q: Do the ~25% user penetration and $200+ order value commerce data include travel?** A: **Yes, travel is included.** Travel planning is also among Muse's top use cases. \ **Risk disclosure and statements:**[**Dolphin Research disclaimer and general disclosures**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [RL.US](https://longbridge.com/en/quote/RL.US.md) - [META.US](https://longbridge.com/en/quote/META.US.md) - [FBL.US](https://longbridge.com/en/quote/FBL.US.md) - [METU.US](https://longbridge.com/en/quote/METU.US.md) - [METD.US](https://longbridge.com/en/quote/METD.US.md) - [FBY.US](https://longbridge.com/en/quote/FBY.US.md) - [METW.US](https://longbridge.com/en/quote/METW.US.md) --- > **Disclaimer: This article is for reference only and does not constitute any investment advice.**