Morgan Stanley's Deep Dive into Muse: Six Key Questions on Monetization, Costs, and Valuation
nashnova research
Meta's AI agent app Muse hit 5 million downloads in just 23 days — far faster than ChatGPT or Grok; Morgan Stanley dissects six critical questions spanning monetization, unit costs, compute capacity, enterprise, the Amazon relationship, and valuation, with the core tension being that faster user growth means higher compute costs, and Meta must prove revenue can cover the bill.
How fast is Muse growing?
Muse crossed 5 million downloads in just 23 days — ChatGPT took 56 days to reach the same milestone, Grok took 103 days, and Claude took 492 days.
This means → a dramatically faster new entrant has appeared in the consumer-agent race, with Meta's social-distribution muscle acting as a cold-start accelerator for AI apps.
But growth itself is a burden: more users mean higher back-end compute costs, and Meta must prove out its business model quickly.
How will it make money? How does Meta carve into a $30 trillion consumer market?
Morgan Stanley expects Muse to generate little near-term revenue from subscriptions or take rates. The current priority is scaling users, boosting engagement, and training the habit of completing transactions through an agent.
Large-scale monetization likely arrives around 2027–2028. The long-term model is a real-time bidding agent marketplace — businesses bid based on expected ROI and pay Meta a transaction fee.
The addressable market behind this model reaches $30 trillion, spanning retail, travel, mobility, food delivery, advertising, logistics, and wearables.
In plain terms = Muse is "burning cash to acquire users" right now, betting that once the habit sticks, Meta can clip a fee on every consumer transaction.
How much does each user cost to serve?
Morgan Stanley estimates Muse's serving cost at roughly $3 to $130 per user per month, depending on token consumption. At a baseline of 25% weekly token-quota usage, the monthly cost is about $37 per user.
The estimate assumes Meta pays AWS on-demand pricing for all CPU sandbox capacity — a worst-case rate — so actual costs are likely lower.
This means → running an agent app is not an easy business to replicate by outspending. Meta's core ad business generates roughly $250 billion in revenue, growing 27% year-on-year, with a global average revenue per user of about $70 — enough to absorb heavy upfront AI infrastructure spending.
Is there enough compute?
Morgan Stanley estimates Meta's 2026 CPU purchases total approximately 643 million vCPUs, translating to about 321 million Muse instance slots.
Under assumptions of 3 hours of daily active use, a 2× peak multiplier, and a 1.2× capacity buffer, supporting 1 billion daily active users would require roughly 300 million concurrent instances — existing capacity largely covers this.
The report projects Meta's 2027 capex at roughly $225 billion, implying about $59 billion in depreciation; third-party compute contract spending rises to about $26 billion, up from $17 billion in 2026.
What about enterprise and the Amazon relationship?
Meta is pushing agents into the enterprise market, targeting roughly $25 trillion in global knowledge-worker economic activity. The product lineup includes Muse for SMBs, WhatsApp/Messenger agent tools, enterprise products, Muse Coding, and a future model API codenamed "Watermelon."
A sensitivity analysis shows that 60 million users paying $15/month could lift EPS by about 8.2% by 2028.
Once Muse starts shopping on behalf of users, Amazon becomes unavoidable — Amazon holds roughly 40% of U.S. e-commerce. The key negotiation points: who is the merchant of record, who controls the shopping experience, how Prime benefits and retail-media value are split, and what data flows back to Muse.
In plain terms = Meta and Amazon must cooperate (Meta needs AWS compute and e-commerce fulfillment) while competing for the consumer gateway — a classic "co-opetition" dynamic.
Is the current valuation expensive?
After the Muse launch, Meta's trailing-twelve-month consensus EPS multiple rose from roughly 14× to 21×. Morgan Stanley views the current valuation as partially reflecting Meta's first-mover advantage from offering Muse for free.
Whether the multiple expands further hinges on three things: sustained growth in user scale and engagement; whether agents can drive commercial transactions at scale; and how fast competitors like Google advance.
This reflects the market's core judgment today: Muse's most important value is not near-term revenue contribution, but whether Meta can transform a high-cost AI app into an agent platform connecting consumers, businesses, and commercial transactions — 2027–2028 is the window that will test this thesis.
市场有风险，内容仅供研究参考，不构成投资建议。