Meta’s Muse Hits 5M Downloads in 23 Days, Morgan Stanley Maps the Agent Economy
TREE NEWS reports: Meta’s consumer AI agent app Muse has crossed 5 million cumulative downloads just 23 days after launching in the U.S. and Canada on September 8, making it one of the fastest-growing AI applications on record. Sensor Tower data cited in a new Morgan Stanley research note shows Muse reached the milestone far quicker than ChatGPT (56 days), Grok (103 days) and Claude (492 days). The surge has turned investor attention to a harder question: can Meta convert a massive, compute-hungry user base into durable revenue?
What Happened
Morgan Stanley’s internet team, led by analyst Brian Nowak, published a deep-dive framing six focal questions around Muse: monetization, serving costs, compute reserves, the enterprise market, an Amazon partnership, and valuation. The central tension is straightforward — the faster agent usage grows, the higher the inference bill. Meta must prove that advertising, transactions, enterprise subscriptions and API revenue can eventually cover that spend.
Market Implications
Costs as a moat. Morgan Stanley estimates Muse serving costs at $3 to $130 per user per month depending on token consumption, with a baseline case near $37 per user monthly. GPU inference scales with tokens; CPU sandbox costs scale with activation frequency and virtual machine runtime. That economics profile means consumer agents are not easily replicated by undercapitalized challengers. Meta’s roughly $250 billion core advertising business, growing 27% year over year, funds the buildout.
Compute is not the near-term bottleneck. The bank estimates Meta’s 2026 CPU purchases — about 643 million vCPUs, or roughly 321 million Muse instance equivalents — could theoretically support 1 billion daily active users under assumptions of three active hours per day and peak and buffer factors. The real challenge is converting capital expenditure into engagement and revenue. Meta’s 2027 capital spending is projected near $225 billion, with roughly $59 billion of depreciation and about $26 billion of third-party compute contracts.
Monetization is a 2027–2028 story. Subscription revenue will be limited near term and transaction take rates low. The larger prize is a real-time bidding marketplace for agent commerce, spanning retail, travel, mobility, food delivery, advertising, logistics and wearables — a pool Morgan Stanley sizes at up to $30 trillion. Enterprise is a second front, targeting a roughly $25 trillion knowledge-work economy through SMB tools, WhatsApp and Messenger agents, Muse Coding and a future model API. Sensitivity analysis suggests 60 million users paying $15 monthly could lift 2028 EPS by about 8.2%, while dedicating 100MW to API could add roughly 4%.
Amazon is the unavoidable counterparty. If Muse shops for users, it collides with Amazon’s roughly 40% U.S. e-commerce share. Negotiations would touch merchant of record, control of the shopping experience, Prime benefits, retail media value and data flows — while Meta simultaneously depends on AWS for CPU, GPU and model distribution via Bedrock.
Investor Takeaways
- Meta’s forward consensus EPS multiple has re-rated from about 14x to 21x following Muse’s launch and a state attorneys general settlement. That already prices in some first-mover advantage.
- Further multiple expansion depends on sustained user and usage growth, proof of scaled agent transactions, and the pace of rivals such as Google.
- Meta touched 28x in February 2025, so re-rating is possible if commercialization evidence accumulates.
- Watch AI-driven engagement gains, business messaging, agent commerce and API monetization as the key fundamental upside drivers.
The near-term value of Muse is less about revenue and more about whether Meta can turn a high-cost AI application into an agent platform connecting consumers, enterprises and commerce.
