META Follow-up Call Trans: Muse biz. value driven by high-value users; shopping is a high-frequency use case

Below is the follow-up call Trans for $Meta Platforms(META.US). For earnings takeaways, see Dolphin Research's reviews (original) and the earnings call recap (original).

I. Key takeaways

Muse currently monetizes only via subscriptions. Transactions/ads and other levers will come after user scale and Connect partner coverage improve, and over 20% of U.S. Muse users are high-value with annual ARPU above $1,000.

Muse connector applications stood at 1,500+ at Connect and have more than doubled since. Roughly 30% of users have connected at least one external service.

The 'hundreds of billions of dollars' equity financing in the 8-K is a theoretical ceiling used to model the point where founders could lose control. It does not reflect current financing needs or plans, and there is no related arrangement at present.

Management's '10x' cost-efficiency goal includes portions that will not reach 10x with what is in hand. The remainder depends on tech progress over the next year, and unit economics hinge on the pace of monetization ramp vs. compute cost deflation.

Internal AI productivity: per-capita software output has more than doubled YoY. AI assists in over 70% of analytical output (near zero at the start of the year), and the share of non-engineering roles (PM/design/data science/research) writing code rose from ~5% a year ago to ~30%.

II. Ops review

Muse user mix: in the U.S., 20% of Muse users have annual ARPU ≥ $1,000. This places Muse among FoA's highest-value users, and many of FoA's top-ARPU users were early Muse adopters.

Muse geographic coverage: live only in the U.S. and Canada. Expansion to more regions is constrained by service costs and must be commercially sustainable.

Muse ecosystem: connector applications increased from 1,500+ at Connect to 3,000+ now (management: 'more than doubled'). About 30% of users have connected at least one service.

Muse use cases: shopping is among the highest-frequency activities in Muse. With strong commercial intent, management sees a relatively direct path to transaction monetization.

Small biz.: roughly 500 mn self-identified small businesses on platform, and Muse for small business went live in late Sep. There are about 3.6 bn consumers on platform.

Internal AI productivity: per-capita software output more than doubled YoY, and analytical output rose by ~10%–20% YoY. >70% of analytical output is AI-assisted, and ~30% of non-engineering staff now write code (vs. ~5% a year ago); since 2Q, AI-driven data infra optimizations saved substantial internal compute and cleaned up 50,000+ redundant data assets.

III. Strategy

Muse monetization path: near term, prioritize scale (growth, penetration, features) and the data flywheel. News-related monetization will focus on subscriptions, token usage, and revenue share, with ads not ruled out; transaction monetization awaits user and connector scale, and the contribution in 2027 will depend on the ramp speed.

Muse complements FoA: FoA drives inspiration and discovery, while Muse captures explicit commercial intent (research → compare → purchase). This is a new wedge into search/intent-driven ads, supporting multiple revenue streams, and the long-term vision is agent-to-agent interactions between consumer and enterprise agents, leveraging 3.6 bn consumers and 500 mn businesses.

On-platform vs. off-platform transactions: like ads, transactions need not close within Meta's properties. What matters is proving ROAS for merchants, and the form factor for Muse's transaction monetization is not yet set, provided the value is measurable.

Compute/CapEx: compute supply remains tight industry-wide, and Meta is also constrained. Past demand was underestimated; the approach is build primarily in-house (optimized for own workloads but longer lead times) and use third-party cloud to bridge near-term gaps, with infra designed to be fungible so capacity can flex; land is reserved and large campuses prepared ahead, and server orders are staggered to control current commitments, leveraging the balance sheet as an advantage in an uncertain funding backdrop.

Large clusters and training: multiple data center campuses are at different stages, some will be very large but most are still under construction. Pre-training relies on large clusters, while post-training can be done on smaller clusters.

Cost efficiency: the levers are, in order: ① distill Muse Wonder capabilities into smaller, more efficient models; ② inference and product-layer efficiency; ③ off-peak compute utilization; ④ task-specific specialist models. A full 10x uplift will require additional technical breakthroughs.

Ad ranking and recommendations: continue rolling LLM-style modeling across the ad stack (Meta Generative Recommender). Ads are selected generatively rather than scored one by one, with conversion value prediction beyond just conversion likelihood; new tech ships in a single service first, then expands to others, and there is further runway in ranking/recommendation.

Muse security and trust: each user's Muse runs in an isolated cloud VM, with a Sentinel Agent gating and auditing all actions at the system level. Muse can propose actions but requires Sentinel authorization; credentials are stored securely so Muse can use but not read them, and sensitive actions (emails, purchases) require user confirmation with full audit trails retained.

IV. Q&A highlights

Q: Does news monetization exclude ads, and does 4Q revenue guidance lean more on pricing or engagement?

A: The priority now is scaling Muse and its data flywheel. News monetization centers on subscriptions, tokens, and revenue share, with ads not excluded.

4Q guidance covers multiple scenarios, with revenue expected to keep growing at a healthy clip. Drivers are ad efficacy investments → advertiser demand, plus engagement growth → more ad inventory, and contributions are not broken out.

Q: Does transaction monetization require more Connect partners, and how does the 8-K 'hundreds of billions' equity figure relate to Class C shares?

A: Before monetizing transactions, user count and partner/connector scale must grow. Shopping is one of Muse's top activities, and high-ARPU users were early adopters, making the monetization path relatively direct once scaled.

'Hundreds of billions' is a theoretical cap used to estimate when control could be triggered, not a financing plan. There is no announcement at this time (no near-term equity dilution risk).

Q: Compute environment and build vs. lease mix, and which KPIs does Muse track?

A: Compute remains tight, and demand will keep rising into 2027; in-house build is primary with third-party cloud bridging near-term shortfalls, and the mix is undisclosed. Infra is designed to be fungible, with land/campuses pre-positioned and server orders placed in tranches, preserving long-term flexibility and limiting near-term commitments.

KPIs focus on core usage/engagement and cohort-based retention curves. Management is satisfied with the curve shape, though no numbers were disclosed; next steps include more countries, more partners/connectors, voice mode, deeper integration with FoA and ads, Buy with Muse, and enterprise use cases.

Q: Long-term monetization for consumer agents, and the ordering of the 10x cost-efficiency drivers?

A: Muse is used for proactive research, recommendations, price comparison, and assisting checkout. This wedges into intent-driven ad markets in a new way, and a 500 mn SMB base plus 3.6 bn consumers supports an agent-to-agent ecosystem.

A: In order: distill Muse Wonder into small models → inference and product efficiency → off-peak compute → specialist models. That is the path to potential 10x.

Q: With revenue growth and 10x+ compute efficiency, will incremental capital per incremental revenue fall, and where is internal AI uplift evident?

A: Unit economics depend on two curves: the monetization ramp for Muse vs. compute cost declines. For now, Muse monetizes only via subscriptions, other engines need time, and the most certain low-hanging efficiency gains are short of 10x, with 10x relying on the next year of tech progress (management did not confirm that capital intensity will fall).

Internal uplift is clear: per-capita software output more than doubled YoY, and analytical output is up ~10%–20% YoY, with >70% AI-assisted and the share of non-engineers writing code up from ~5% to ~30%. Since 2Q, 50,000+ redundant data assets were cleaned up and compute was saved, reflected in R&D efficiency and internal infra costs.

Q: How will you promote Muse to ~4 bn FB/IG users and on what timeline, and what is the status of other large clusters after Prometheus?

A: Muse is only live in the U.S. and Canada, and expansion depends on whether service costs can be sustainably covered, with no timeline given. Once opened up, distribution will leverage deep FoA integration and chat entry points.

Public info on other clusters is limited: multiple campuses are in different phases, some will be very large but most are still under construction. Post-training can happen on smaller clusters, so model iteration does not rely solely on bringing large clusters online.

Q: How does Muse balance first-mover advantage with first-use experience, and how much runway remains for GEM/Andromeda/Lattice rollouts?

A: Product-market fit is strong, non-technical users can onboard easily, and proactivity stands out as a feature. Management believes Meta leads in agents and has unmatched scaling capability.

On ads: recall and ranking stacks keep improving, and Meta Generative Recommender selects ad candidates generatively and predicts conversion value. It launches in a single service first before broader rollout, leaving more upside in ranking and recommendation that is not quantified.

Q: Are Muse users different from avg. users in the U.S./Canada, and does moving upmarket require Muse to complete transactions autonomously rather than just facilitating?

A: In the U.S., over 20% of Muse users have annual ARPU ≥ $1,000. It is unclear whether full autonomy is needed; as with ads, transactions need not close on-platform, and what matters is the value created for merchants vs. other marketing channels.

Q: How do users trust Muse with Gmail/bank connections, and will you allow off-platform checkout from Muse?

A: Security architecture is isolated VM + Sentinel review + unreadable credentials + confirmations for sensitive actions + audit trails. Connector applications have more than doubled from 1,500+, with ~30% of users connecting at least one service; the specific share connecting Gmail/banks was not disclosed, and off-platform checkout will depend on evolution, with ROAS measurability being the core and the monetization form not yet set.

VI. Signals to watch

Muse non-subscription monetization timing: when transaction take-rate, ads, and Buy with Muse go live. The race between monetization ramp and compute cost declines will directly shape CapEx ROIC in 2027.

Muse timeline beyond U.S./Canada: management ties this to per-user service costs. Progress on small-model distillation and off-peak compute utilization are leading indicators.

Whether 10x cost efficiency materializes: management acknowledged low-hanging gains are short of 10x. Track disclosures on Muse per-inference unit costs.

Connector ecosystem: application count (now >3,000) and user connection rate (~30%). Also watch whether large merchants keep resisting integration.

Build vs. lease mix and large-cluster rollout pace: share of third-party cloud contracts and post-Prometheus go-live timing for large clusters. If compute remains constrained, CapEx could be revised up again.

Class C/equity financing: 'hundreds of billions' is a theoretical ceiling. Watch for any actual equity actions.

Ad ranking and recommendation: once Meta Generative Recommender rolls into more services, track its contribution to conversion rates and ad pricing. Those are key proof points.

<end here>

Risk disclosure and statement for this article. Dolphin Research disclaimer and general disclosure