Meta's AI agent app Muse rapidly ignited the consumer agent market just 23 days after launch, becoming one of the fastest-growing AI applications recently.

Data from market research firm Sensor Tower shows that as of September 30, Muse's cumulative downloads exceeded 5 million. Muse launched in the United States and Canada on September 8, reaching that scale in just 22 days; by comparison, ChatGPT, Grok, and Claude took 56 days, 103 days, and 492 days respectively to surpass 5 million downloads.

Muse's rapid growth has drawn market attention to its costs, computing power, and monetization capabilities. Morgan Stanley's internet sector chief analyst Brian Nowak's team recently published in-depth research analyzing six key questions surrounding Muse: monetization, service costs, computing power reserves, the enterprise market, Amazon partnership, and valuation.

Among these, the core contradiction is: the faster agent users grow, the higher the computing power costs; and Meta needs to prove that future advertising, transactions, enterprise subscriptions, and API businesses can cover this investment and convert the massive user base into revenue.

Monetization: Prioritizing User Scale in the Short Term, Targeting a $30 Trillion Consumer Market in the Long Run

Muse is currently not in a hurry to make money. Morgan Stanley expects that short-term subscription revenue contribution will be limited, and transaction commissions will remain at a low level. Meta's more important task is to expand user scale, increase usage frequency, and cultivate user habits of completing commercial transactions through agents.

Truly large-scale monetization may have to wait until 2027 to 2028. In the long run, Meta may build an agent commerce marketplace around real-time bidding: enterprises bid based on expected return on investment and pay transaction fees to Meta.

The potential market size corresponding to this model reaches $30 trillion, covering consumer scenarios including retail, travel, transportation, food delivery, advertising, logistics, and wearable devices.

Before direct charging, Muse can also first strengthen the monetization efficiency of Meta's advertising business by enhancing user engagement and user profiling capabilities.

Costs: Up to $130 Per User Per Month, High Costs May Actually Become a Barrier

The more popular Muse becomes, the first problem Meta faces is cost.

Morgan Stanley estimates that Muse's service cost is approximately $3 to $130 per user per month, mainly depending on token consumption. Among these, GPU inference costs grow with token usage, while CPU sandbox costs depend on user activation frequency and virtual machine runtime.

Based on users consuming 25% of their weekly token quota, the cost is approximately $37 per user per month. Since this estimate assumes Meta rents all CPU sandbox capacity at AWS on-demand prices, actual costs may be lower.

This also means that agent is not an easy business to replicate by burning money. Meta has a core advertising business worth approximately $250 billion, growing 27% year-over-year, as financial support, with global average revenue per user of approximately $70, enabling it to withstand high upfront AI infrastructure investment.

Computing Power: 2026 CPU Procurement Theoretically Sufficient to Support 1 Billion DAU

If Muse continues to grow rapidly, does Meta have enough computing power?

Morgan Stanley believes: Meta's current CPU procurement scale is theoretically sufficient to support 1 billion Muse daily active users.

According to its estimates, Meta's total CPU capacity procured in 2026 will be approximately 643 million vCPUs, corresponding to approximately 321 million Muse instance capacity. Under assumptions of 3 hours of daily activity, a 2x peak factor, and a 1.2x capacity buffer, the simultaneous active instances needed to support 1 billion DAU would be approximately 300 million, and existing computing power can basically cover it.

This means that as Muse's user scale expands, what Meta really needs to solve in the short term may not be "whether there is computing power," but how to convert massive computing power investment into user growth and commercial revenue.

The report estimates that Meta's 2027 capital expenditure will be approximately $225 billion, corresponding to approximately $59 billion in depreciation; third-party computing contract spending will be approximately $26 billion, higher than $17 billion in 2026.

Enterprise Market: Entering the $25 Trillion Knowledge Economy from Consumer Agents

Muse's commercial imagination is not limited to individual users.

Meta is further pushing agents into the enterprise market, targeting approximately $25 trillion in the global knowledge work economy. Related products include Muse for small and medium-sized enterprises, WhatsApp and Messenger agent tools, enterprise products, Muse Coding, and the future model API "Watermelon."

Morgan Stanley believes that traditional enterprises' Web 1.0 architecture is not suitable for direct interaction between agents. In the future, enterprises will need to provide machine-native interfaces so that agents can directly call products, services, and enterprise systems.

This provides Meta with a new charging model: charging enterprises subscription fees while providing API computing power and model services to developers and enterprises.

Sensitivity estimates show that if 60 million users pay $15 per month, it could boost EPS by approximately 8.2% by 2028; if 100MW of dedicated computing power is used for the API business, it could also increase 2028 EPS by approximately 4%.

Amazon: As Agents Enter E-commerce, the Transaction Gateway Becomes the Core Contest

Once Muse starts shopping for users, Meta cannot avoid Amazon.

Amazon accounts for approximately 40% of the U.S. e-commerce market and controls a massive product, merchant, and logistics system. The core issues in potential cooperation between the two sides include who is the merchant of record, who controls the shopping experience, how Prime benefits and retail media value are distributed, and what data can flow back to Muse.

At the same time, Meta also relies on Amazon's AWS cloud resources, including CPU and GPU computing power, as well as model API distribution channels provided through Bedrock.

Therefore, potential cooperation between Meta and Amazon is not just an ordinary e-commerce partnership, but may simultaneously involve transaction gateways, user data, cloud computing power, and model distribution.

Valuation: 21x PE Already Reflects Some Muse Expectations, Next Step Is Commercialization

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

Morgan Stanley believes that the current valuation already reflects some of the first-mover advantages formed by Meta launching Muse in a free model. Whether the valuation can further improve depends on three things: first, whether Muse's user scale and usage rate can continue to grow; second, whether agents can truly form large-scale commercial transactions; third, the pace of advancement by competitors such as Google and SPCX.

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

From a fundamental perspective, the upside drivers Morgan Stanley focuses on mainly include: AI-driven improvement in core user engagement, growth in business messaging and agent businesses, and gradual large-scale monetization of the API platform.

Ultimately, Muse's most important current value is not how much revenue it contributes, but whether Meta can use its massive user base, advertising cash flow, and computing power investment to turn a high-cost AI application into an agent platform connecting consumers, enterprises, and commercial transactions.