23 days after launch, consumer-grade Agent explodes! Morgan Stanley's in-depth analysis of "6 key issues about Muse"

Meta’s AI agent application Muse went viral in just 23 days since its launch, rapidly igniting the consumer Agent market and becoming one of the fastest-growing AI applications recently.

According to data from market research firm Sensor Tower, as of September 30, Muse’s cumulative downloads have surpassed 5 million. Muse launched on September 8 in the US and Canada, reaching this milestone in just 22 days; by comparison, ChatGPT, Grok, and Claude took 56, 103, and 492 days respectively to surpass 5 million downloads.

Muse’s rapid growth has sparked market attention toward its costs, computational power, and commercialization capabilities. Recently, Brian Nowak’s team, chief internet analyst at Morgan Stanley, released an in-depth research report analyzing six key issues regarding Muse: monetization, service costs, compute power reserves, enterprise market, partnership with Amazon, and valuation.

The core contradiction is: the faster Agent user numbers grow, the higher the compute cost; Meta needs to demonstrate that future advertising, transaction, enterprise subscriptions, and API business can cover this investment, converting the massive user base into revenue.

I. Monetization: Focus on User Base First in Short-Term, Eyeing the $30 Trillion Consumer Market Long-Term

Currently, Muse is not in a rush to make money. Morgan Stanley expects that short-term subscription income will contribute little, transaction commissions will stay low, and Meta’s priority is to expand its user base, boost usage frequency, and cultivate user habits for conducting business transactions through Agents.

Large-scale monetization may only arrive in 2027 to 2028. Over the long run, Meta may build a real-time bidding commercial marketplace for Agents: enterprises bid based on anticipated ROI and pay Meta transaction fees.

This model’s potential market size reaches $30 trillion, spanning retail, travel, mobility, food delivery, advertising, logistics, and wearable device scenarios.

Before direct charging, Muse can also enhance user engagement and profiling capabilities to further strengthen the monetization efficiency of Meta’s ad business.

II. Cost: Up to $130 Per User Per Month, High Costs May Pose a Barrier

The hotter Muse gets, the more Meta immediately faces cost issues.

Morgan Stanley estimates that Muse’s service costs range from $3 to $130 per user per month, depending mainly on token consumption. GPU inference costs rise with token usage, while CPU sandbox costs depend on user activation frequency and virtual machine run time.

Assuming the user consumes 25% of weekly token allowance, per-user monthly cost is about $37. Since this estimate assumes Meta rents all CPU sandbox capacity from AWS at on-demand prices, actual costs could be lower.

This also means Agent is not an easy business to copy with “burning money.” Meta boasts around $250 billion in core ad business revenue (up 27% year-over-year) as a financial pillar, generating roughly $70 in global average revenue per user, enough to withstand heavy early AI infrastructure investment.

III. Compute Power: 2026 CPU Purchases Can Theoretically Support 1 Billion DAU

If Muse continues rapid growth, does Meta have enough compute power?

Morgan Stanley believes Meta’s current CPU procurement can theoretically support 1 billion daily active Muse users.

According to estimates, Meta’s 2026 CPU procurement totals about 643 million vCPU, equivalent to about 321 million Muse instance capacity. Assuming 3 hours daily activity, a 2x peak multiplier, and a 1.2x buffer, supporting 1 billion DAU would require around 300 million simultaneously active instances, which current capacity can cover.

This means that as Muse’s user base grows, the real short-term challenge for Meta isn’t “having enough compute,” but how to turn massive compute investment into user growth and commercial revenue.

The report estimates Meta’s 2027 capital expenditures at $225 billion, with about $59 billion in depreciation; third-party computing contract costs are projected at $26 billion, up from $17 billion in 2026.

IV. Enterprise Market: Tapping into the $25 Trillion Knowledge Economy from Consumer Agents

Muse’s business potential extends beyond individual users.

Meta is pushing Agents further into the enterprise market, targeting the roughly $25 trillion global knowledge-based work economy. Associated products include Muse for SMBs, WhatsApp and Messenger intelligent tools, enterprise offerings, Muse Coding, and the future “Watermelon” model API.

Morgan Stanley argues that the traditional Web 1.0 architecture is not suitable for direct Agent-to-Agent interaction, and that companies in the future will need to provide machine-native interfaces to allow Agents to directly call products, services, and enterprise systems.

This offers Meta a new revenue model: charging enterprise subscriptions, while providing developers and enterprises with API compute and model services.

Sensitivity analysis shows that if 60 million users pay $15 per month, by 2028, EPS could increase by about 8.2%; if 100MW dedicated compute is used for the API business, 2028 EPS could rise by around 4%.

V. Amazon: Once Agents Enter E-Commerce, Transaction Entry Becomes Key Competition

Once Muse starts shopping for users, Meta can’t avoid Amazon.

Amazon accounts for about 40% of the US e-commerce market, holding massive product, merchant, and logistics systems. The core issues for potential cooperation include who records merchants, who controls the shopping experience, how to allocate Prime benefits and retail media value, and what data can flow back to Muse.

At the same time, Meta relies on Amazon’s AWS cloud resources, including CPU and GPU compute, as well as distribution channels for model APIs through Bedrock.

Therefore, potential cooperation between Meta and Amazon is not just a routine e-commerce partnership, but might simultaneously involve transaction entry, user data, compute resources, and model distribution.

VI. Valuation: 21x PE Already Reflects Some Muse Expectations, Next Step Depends on Commercialization

After the Muse launch and Meta’s settlement with state attorneys general, Meta’s NTM consensus EPS valuation has risen from about 14x to 21x.

Morgan Stanley believes that the current valuation already reflects some first-mover advantages established by Meta rolling out Muse with a free model. Whether the valuation can rise further depends on three things: First, whether Muse can continue scaling its user base and usage rate; second, whether Agents can generate large-scale commercial transactions; third, the pace of competitors like Google and SPCX.

Meta reached a 28x valuation in February 2025, so if Muse continues to grow its user base and further demonstrates commercialization potential, the market may still reassess its valuation level.

On fundamentals, Morgan Stanley focuses on three upside drivers: AI boosting core user engagement, growth in commercial messaging and Agent business, and scaled monetization of the API platform.

Ultimately, the most important value of Muse right now is not how much revenue it contributes, but whether Meta can leverage its vast user base, ad cash flow, and compute investment to transform a high-cost AI application into an Agent platform connecting consumers, enterprises, and commercial transactions.