Morgan Stanley: Muse Costs Up to $130/User/Month, User Growth Plunges 62%
nashnova research
Morgan Stanley published its most detailed cost model for Meta's AI agent Muse, pegging the base-case serving cost at $37 per user per month — already far above Meta's global average annual revenue per user. Meanwhile, daily net user additions have plunged 62% from their September peak, pushing the monetization window further out than the market expects.
How much does it cost to serve a single user?
Morgan Stanley splits Muse's cost into two buckets: GPU inference cost, which scales linearly with token consumption, and CPU sandbox cost, which scales with how often users activate and how long virtual machines run.
The base case assumes 25% quota utilization — about 108 million tokens per month — yielding a monthly cost of $37.12, or $445 per year. This means → serving one user for a year costs roughly 6× Meta's global average annual ad revenue per user (about $70).
Extreme scenarios are starker: Power-tier users at full utilization hit $488/month in inference costs; Maximum-tier users hit $2,925/month. In plain terms = the more a user engages, the more Meta loses — this is pure cash burn for scale.
If costs are so high, why does Morgan Stanley call them a "moat"?
The core argument: the cost itself is the barrier — smaller competitors simply cannot sustain the burn.
Meta's cushion is its core advertising business, growing at 27% annually and generating $250 billion in revenue — a cash engine that can subsidize Muse indefinitely. This reflects a shift in AI-agent competition toward a war of attrition: whoever has the strongest legacy business absorbs losses the longest.
Even at just 5% free-tier utilization, annual serving cost per user is about $124 — still above Meta's average annual revenue per user. This means → every Muse user is a net loss in the short term; profitability depends entirely on future monetization design.
Is user growth actually slowing? By how much?
TickerTrends data shows Muse's daily active users (DAU) hit a new high of 2.73 million on October 6, but daily net additions cratered — from roughly 198,000/day at the September peak to about 74,000/day, a 62.4% drop.
Downloads are weakening too: 169,600 on October 6, down 21.1% from the September 24 peak of 214,900.
Different data providers diverge sharply on total user count: TickerTrends says 2.73 million DAU, Similarweb says 2.18 million, Sensor Tower says roughly 1.8 million — a ~50% gap between the highest and lowest, yet none exceeds 3 million. In plain terms = no one can pin down the exact user base, but every source points the same way — growth is visibly decelerating.
Can the "one-billion-user" story still hold?
Morgan Stanley's capacity analysis is modeled on 1 billion DAU. By TickerTrends figures, Meta's purchased CPU capacity is roughly 366× current actual users. This means → hardware has been massively front-loaded, betting on explosive user growth that has not yet arrived.
At the current pace of 74,000 net adds per day, reaching 1 billion DAU would take about 37 years. This reflects the vast gap between the "perfect growth curve" embedded in the current valuation and the actual trajectory.
Morgan Stanley explicitly warns that slowing growth will push the monetization milestone — currently pegged at late 2027 to 2028 — further out. This is the single most important validation risk right now.
What is the monetization plan, and when does it kick in?
Analyst Brian Nowak estimates substantive monetization no earlier than late 2027 to 2028. He wrote: "We expect Muse subscription revenue to remain very limited for a considerable period, as Meta is focused on growing the user base and engagement, not monetization."
Bank of America sees material revenue from ads, subscriptions, or commissions as unlikely before 2028; Goldman Sachs flags "whether monetization can follow quickly" as a key question. This means → three major banks converge on the same call — do not expect Muse to make money any time soon.
The long-term blueprint envisions a real-time auction marketplace — similar to ad bidding — where businesses pay for transactions completed by the agent. The target spending pool is roughly $30 trillion in consumer spending, plus $25 trillion in global knowledge work, totaling $55 trillion. In plain terms = the addressable market is enormous on paper, but the entire model requires a massive, active user base before businesses will bid real dollars.
Could external platform blocks choke Muse?
Amazon blocked Muse from its platform on September 20, citing Muse's failure to identify itself as an AI agent and inadequate disclosure of user-credential storage.
TickerTrends tracked user reports of similar restrictions from Costco and Netflix; telecom carriers have begun routing Muse-initiated AI customer-service calls away from live agents.
This reflects a defensive reaction from platforms against Muse's "act on behalf of the user" model — if more platforms refuse to cooperate, the range of tasks the agent can perform shrinks, and so does its value proposition to users.
市场有风险，内容仅供研究参考，不构成投资建议。