TL;DR
·JPMorgan upgraded Meta from Neutral to Overweight, raising its price target from $640 to $820, implying roughly 25% upside from the stock price at the time of the report.
·Muse Spark 1.3 has entered the frontier model tier, marking Meta's initial transformation from a laggard to a major competitor.
·Consumer agent Muse rose to No. 3 on the U.S. App Store free chart on its second day, with early user engagement intensity reaching 10 times that of the internal testing group.
·Meta is expanding AI monetization from advertising to transaction commissions, paid subscriptions, enterprise agents and model APIs, creating multiple new revenue paths.
·JPMorgan expects Meta's capital expenditures to reach $243 billion and $284 billion in 2027 and 2028, respectively, putting significant pressure on free cash flow.
·The research report argues that current forecasts have already priced in most AI investment but have not yet priced in revenue from new products such as Muse and model APIs, and therefore may underestimate Meta's earnings potential.

Why does Meta deserve a re-rating?

Meta shares have rebounded about 20% from recent lows, but are still down 1% year-to-date, while the S&P 500 has gained about 12% over the same period. As the market remains concerned about excessive AI spending and deteriorating free cash flow, JPMorgan upgraded Meta from Neutral to Overweight and sharply raised its December 2027 price target from $640 to $820.

Based on the September 9 stock price of $653.69, the new price target implies about 25% upside. JPMorgan values Meta at roughly 23 times its 2028 GAAP earnings per share of $35.44, above the S&P 500's valuation level of about 16 times.

The core judgment supporting this valuation premium is that Meta is transforming from a social platform that uses AI to optimize recommendations and advertising into an AI platform that also offers frontier models, consumer agents, enterprise agents, developer APIs and paid subscriptions. In the past, the market mainly saw capital expenditures; now Meta is beginning to show how these investments will translate into revenue.

Model capability is the foundation of this commercial system. When Meta Superintelligence Labs was rebuilt in the summer of 2025, the company proposed launching a frontier-competitive model within one year. After releasing Muse Spark 1.1 in July 2026, Meta quickly iterated to Muse Spark 1.3.

Based on third-party test results, Muse Spark 1.3 has entered the frontier model tier across multiple capabilities including agentic tasks, coding, instruction following, and long-context processing, significantly narrowing the gap with leading models such as Claude and GPT. JPMorgan accordingly believes that Meta has essentially delivered on its previously stated goal of catching up on models.

Artificial Analysis Model Intelligence Index. Muse Spark 1.3 has entered the frontier model tier, with competitive capabilities in agentic tasks, coding, and instruction following.

The next-generation model, Watermelon, is expected to further advance model capabilities. It employs a higher level of pre-training, and the generation of models after Watermelon has already begun scaled training on Meta's Prometheus gigawatt-scale compute cluster in Ohio.

Meta's true competitive advantage lies not just in the models themselves, but in the combination of models and distribution channels. The company's products reach approximately 4 billion users and connect hundreds of millions of businesses and advertisers. Once the underlying models reach frontier-level performance, Meta can rapidly embed them into Facebook, Instagram, WhatsApp, and Messenger, creating a scale advantage that other AI labs find difficult to replicate.

From Muse to Meta One, AI Begins to Find a Monetization Entry Point

Muse is a consumer-grade AI agent recently launched by Meta, now available in the U.S. on iOS, Android, and web. It can browse websites and operate user interfaces through its own virtual computer, completing tasks on behalf of users such as online shopping, restaurant reservations, filling out forms, and sending emails and messages.

Currently, Muse is already able to work with platforms including Instagram, WhatsApp, Spotify, DoorDash, Etsy, Reddit, Yelp, Outlook, and Gmail. The day after launch, it rose to No. 3 on the U.S. App Store free app chart, with early user engagement intensity reaching 10 times that of the internal testing group, exceeding Meta's own expectations.

Muse's ranking on the U.S. App Store free app chart. Muse rose to No. 3 on the U.S. App Store free app chart the day after launch, demonstrating strong early user acquisition capability.

Muse currently offers free users a weekly quota of 100 million tokens, and has set up two paid tiers at $20 and $100 per month. However, Meta is not preparing to rely solely on subscription revenue. Since Muse can directly shop, book, and complete transactions on behalf of users, the company is more likely to take commissions from transactions in the future.

This means Muse is not just targeting the chatbot subscription market, but a consumer transaction market that could reach tens of trillions of dollars. As automated interactions between agents increase, merchants may also gain more orders, customers, and operational data through Muse.

Muse application scenarios and task execution workflow. Muse uses a virtual computer to call different websites and apps, completing searches, decisions, and transactions on behalf of users.

Commercialization on the enterprise side has already gone further. Meta Business Agent can help businesses answer questions, recommend products, book services, and qualify sales leads. As of the second quarter of 2026, more than 1 million businesses were using this product on WhatsApp and Messenger each week.

The Business Agent Platform for large enterprises can connect to hundreds of external systems such as Shopify, Zendesk, and Shopee, allowing agents to directly perform operations on behalf of businesses. The platform officially launched token billing on August 1: $2 per 1 million tokens, with the cost of each message interaction at about 4 to 5 cents.

In the future, Meta may also adopt an "outcome-based pricing" model similar to ad bidding, allowing businesses to bid around actual results such as sales, bookings, or lead conversions. Since Meta has already established business relationships with hundreds of millions of advertisers and small and medium-sized enterprises, Business Agent does not need to build enterprise sales channels from scratch.

Meta Business Agent Platform pricing. Enterprise agents have already launched token billing, giving Meta a new entry point for enterprise service revenue.

The developer market is another monetization path. Meta Model API allows developers to call Muse Spark to build agents and multimodal workflows, while Muse Code is used to handle complex software engineering tasks.

Muse Spark 1.3 Standard is priced at $1.25 per million input tokens and $4.25 per million output tokens; the contributor version, which allows Meta to use data to improve its products, costs just $0.10 for input and $0.20 for output. Muse Code also offers three subscription tiers at $5, $15, and $50 per month.

The low prices indicate that Meta is prioritizing usage scale and developer adoption at this stage. Early data has already shown positive signals: the Muse Spark 1.3 contributor version ranks first in market share on OpenCode, having processed approximately 31 trillion tokens cumulatively since launch, covering 212,000 unique users.

In addition to agents and APIs, Meta has also begun selling subscription services to individual users, enterprises, and creators through Meta One. Individual plans are priced at $7.99 and $19.99 per month, offering higher Meta AI usage quotas as well as premium features across Instagram, Facebook, and WhatsApp.

JPMorgan estimates that based on an average annual revenue of $182 per individual user and a paid penetration rate of 2% to 3%, Meta One individual subscriptions could generate $14.2 billion to $21.3 billion in revenue and contribute $3.30 to $4.96 in GAAP earnings per share. If average annual revenue rises to $211, revenue could further reach $16.4 billion to $24.6 billion.

Meta One individual subscription revenue and EPS sensitivity analysis. Under a neutral penetration scenario of 2% to 3%, individual subscriptions could contribute approximately $14.2 billion to $24.6 billion in revenue.

Enterprise and creator versions are priced higher, ranging from $14.99 to $499.99 per month. Meta currently has over 200 million business users and tens of millions of professional creators. Based on a neutral scenario of $600 to $1,200 in average annual revenue and a penetration rate of 8% to 12%, this segment could contribute $12 billion to $36 billion in revenue by 2028, along with $2.80 to $8.40 in GAAP earnings per share.

Meta One enterprise and creator subscription revenue and EPS sensitivity analysis. Leveraging over 200 million business users, enterprise and creator subscriptions could become Meta's more elastic AI revenue source.

The above estimates do not yet include Meta AI's potential advertising revenue. In the future, Meta AI itself could become a new advertising channel; user interaction data with AI assistants could, in turn, improve ad targeting effectiveness on Facebook, Instagram, and WhatsApp.

Massive investment is not over yet, but revenue has not been fully factored into forecasts

Meta's AI commercialization path is gradually becoming clearer, but that does not mean capital expenditure risks have disappeared. On the contrary, investment over the next two years could be even more aggressive.

JPMorgan expects Meta to maximize its computing capacity expansion in 2026 and 2027. According to reports, the company plans to reach approximately 7GW of computing capacity in 2026 and double it to 14GW in 2027.

The research report estimates that Meta's capital expenditure will increase from $69.7 billion in 2025 to $142.5 billion in 2026, then grow another 70% to $242.7 billion in 2027, and reach $283.5 billion in 2028, all significantly above market consensus.

Massive capital expenditure will directly weigh on cash flow. JPMorgan expects Meta's free cash flow to drop from $47.1 billion in 2025 to negative $1.4 billion in 2026, and further to negative $58 billion and negative $49.6 billion in 2027 and 2028, respectively. The company may also shift from net cash to net debt in 2026, with net debt reaching approximately $165.2 billion in 2028.

Meta income statement and capital expenditure forecasts. Meta's capital expenditure will rise sharply over the next two years, expected to put continuous pressure on free cash flow.

However, this is also the key reason JPMorgan turned bullish on Meta: current financial forecasts already factor in most AI infrastructure costs, but have not yet included new revenue from Muse, model APIs, and other new AI products. In other words, costs are already in the model, but potential revenue has not yet been fully accounted for.

If Meta ultimately ends up with excess computing capacity, the company could also rent out some computing power to external customers. Recently, some high-end computing contracts have reached prices of $30 to $50 per watt, higher than the typical $10 to $20 per watt level for new cloud computing companies. But JPMorgan expects Meta will still prioritize using computing power for ad optimization, frontier model training, and its own AI products, because these internal scenarios could bring higher returns.

Meanwhile, the core advertising business remains Meta's foundation for shouldering AI investments. AI can improve content recommendations, increase user engagement time and ad targeting efficiency, and help advertisers generate creative materials at scale. JPMorgan expects Meta's revenue to grow from $201 billion in 2025 to $254.3 billion in 2026, reaching $305.3 billion and $354.4 billion in 2027 and 2028, respectively.

The risks facing Meta remain clear: AI investments may continue to exceed expectations, and product monetization speed may not keep pace; Muse and other agents may fail to retain users long-term; Google, TikTok, and OpenAI are also competing for user time and advertising budgets.

Therefore, whether the $820 price target can be realized depends not only on whether Meta can continue investing in AI, but on whether Muse, Business Agent, model API, and Meta One can generate real revenue in the next two years. JPMorgan's judgment is that Meta has crossed the threshold of insufficient model capabilities, and AI spending is beginning to transform from a pure cost into a commercial system that can be subscribed to, called upon, charged for, and used in transactions.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia