Muse Wonder Adds Fuel to the Agent Frenzy, Market Buzzes Over Meta (META.US) Second Growth Curve! Can "AI FOMO" Ignite Earnings Season?

Wall Street financial giant Citigroup's latest research report shows that Meta (META.US), the parent company of Facebook and Instagram, is very likely to strengthen its product cycle with memory, reasoning, and continuous execution capabilities through the Muse Wonder series of personal AI agents/AI large models, codenamed "Watermelon," launched locally on Monday. User growth, channel distribution advantages, and Meta-led massive expansion of AI computing resources are expected to accelerate the commercialization of the company's AI agent business. Citigroup's strategist team stated that Meta's newly launched Muse Wonder series of AI application products strengthens complex task processing, long-term programming, and professional scenarios such as finance and law through Active Memory (active recall, active invocation) cross-conversation memory and Active Focus continuous execution capabilities, further enhancing Meta's AI product cycle. Data cited in the research report shows that Muse accumulated approximately 8.6 million downloads in the United States within 34 days of launch, with daily active users exceeding 2 million, while Meta's massive application distribution advantages have yet to be further unleashed. Citigroup emphasizes that the Muse series of AI agents is expected to drive Meta's expansion from advertising monetization to long-term paid subscriptions for AI agents, model APIs, and professional work services. The AI agent-related business landscape is expected to become a "second growth curve" second only to the hundred-billion-dollar-level digital advertising business, and the future commercialization space for agents is far stronger than digital advertising. As Meta Muse and OpenAI's GPT-6 Astra comprehensively ignite the massive wave of AI agents, "AI FOMO trading" (referring to investors' fear of missing out on gains in AI-related assets, thus chasing purchases or replenishing previously reduced positions) can be said to have made a comeback. From the perspective of the underlying architecture of AI computing infrastructure, Muse and Astra are expected to further expand the scope of AI agent usage. The massive expansion of AI inference workloads they bring will simultaneously increase demand for model inference, tool execution, and state management, while deepening competitive pressure on some software companies. Accelerators such as GPUs and TPUs handle model computation; high-performance server CPUs run browsers, virtual machines, product retrieval, database queries, and transaction orchestration; HBM and server DRAM carry model data, context, and concurrent working environments; enterprise-grade SSDs store product indexes, task records, and persistent states, while high-speed networks and optical interconnects support distributed data exchange. In the view of Citigroup's strategist team, this AI FOMO trading frenzy may even ignite the bullish frenzy of this round of U.S. stock earnings season that began in Octobermainly based on the fact that the popularity of AI agent-related products is expected to translate on a large scale starting this year into long-term use by more and more users and Token invocation/long-term subscription payments, and together with the digital advertising growth ecosystem, support continuously expanding computing investment, allowing model/agent competitiveness to further manifest as strong earnings growth expectations. Citigroup's strategist team stated that Wall Street analysts' consensus expectation for S&P 500 third-quarter year-over-year earnings growth has been significantly revised upward from 26.7% at the beginning of the quarter to 29.6%, with energy and AI technology themes being the main sources of incremental upward revisions. All 11 sectors are expected to achieve year-over-year earnings growth. Citigroup's proprietary model predicts that 66.2% of Russell 1000 constituents will post positive earnings surprises (i.e., earnings exceeding market consensus) and related positive stock price return trajectories (i.e., Citigroup's model also simultaneously predicts a positive direction for related stock price returns), significantly higher than the already strong 60.9% last quarter, reaching the highest level since the fourth quarter of 2021. Wall Street financial giant Jefferies Financial Group Inc. recently stated that, driven by the dual engines of the AI investment frenzy and AI-related corporate earnings exceeding expectations, the S&P 500 is expected to soar to 8,000 points by the end of 2026 and further reach 9,000 points in 2027. The core logic of Jefferies Financial Group Inc. is clear and powerful: in a cycle where AI-driven earnings growth exceeds the historical average by more than twice, it is dangerous to fight against the earnings trend. Jefferies Financial Group Inc.'s 2026 baseline forecast of 8,000 points for the S&P 500 is based on earnings per share (EPS) reaching $373 (35% year-over-year growth, far above the market consensus of 29%) and a price-to-earnings ratio of 21.5 times. From AI large model competition to leading the personal AI agent trend, Muse is expected to open a second growth curve Meta's newly launched Muse Wonder series is the underlying AI large model driving the most cutting-edge AI agents, focusing on "deeply remembering context, continuously executing, and completing delivery." Active Memory retrieves relevant information across conversations, while Active Focus supports switching approaches when encountering obstacles, error correction, and follow-up, with key coverage of long-term coding and finance, legal, and health tasks. Its differentiated competitiveness compared with Anthropic's and OpenAI's flagship products is concentrated in context continuity and execution reliability for long tasks, with the goal of reducing repeated user explanations and manual intervention. Wonder has been integrated into Meta Model API and Muse Code, with Full available to Muse Code subscribers; upgrades to the Muse app will be prioritized for subscription users. On Monday, the 12th, U.S. Eastern Time, Meta officially released the Muse Wonder series of models, including Muse Wonder for fast frontier performance and the full version Muse Wonder Full. Zuckerberg said on social media that Muse Wonder is a larger and more advanced class of model than the previous Muse Spark, marking Meta's move toward the industry frontier in AI capabilities. Muse Code is Meta's programming agent tool for software development and is a different product from the general-purpose Muse agent app. According to the release arrangement, Muse Wonder Full is only available to Muse Code subscribers, and non-subscribers cannot currently use Full; the standard Wonder version has already been provided through Meta Model API and Muse Code. In addition, the official general-purpose Muse agent app covering Meta's ecosystem platforms Facebook and Instagram has not yet announced that general-purpose Muse can immediately use Wonder Full. The Artificial Analysis leaderboard cited by Citigroup ranks Wonder Full third; the advantages in the research report's comparison are concentrated in memory, law, finance, and long-horizon software development. This means Meta is entering competition in professional workflows. The product chain reviewed in the research report extends from continuous iteration of Spark to Muse Code, the open-source model Glimmer, voice and vision, smart glasses, and the SecureVM execution environment: underlying capabilities, tool interfaces, and execution environments are gradually becoming connected, which is conducive to translating model progress more quickly into deliverable services. Application data provides preliminary validation: Muse accumulated approximately 8.6 million downloads in the United States within 34 days of launch, with daily active users exceeding 2 million. Early download speed led ChatGPT, Gemini, and Meta AI at their respective launch stages, and downloads over the past four weeks exceeded those of each individual Meta app. Citigroup believes the group's distribution advantages have only just begun to be unleashed and looks forward to experiences such as Muse Voice real-time conversation and Business Calls making calls on behalf of users. The investment implication lies in the potential incremental contributions from subscriptions, APIs, and professional services; when agents take over cross-software workflows, some software companies' interface entry points and per-seat charging will also come under competitive pressure. From 1GW training to 14GW expansion: earnings season will test how Meta's massive computing resources become Token cash flow Citigroup survey data shows that Wonder is trained at the 1GW-level Prometheus hyperscale data center. The research report's expectation of approximately 14GW of total computing capacity by 2027 implies that computing supply may double year over year, supporting more intensive model and agent releases and possibly leaving room for short-term computing partnerships. From an engineering mechanism perspective, the spread of agents simultaneously increases demand for model inference, tool execution, and state management: accelerators and HBM handle inference, CPUs execute tools, persistent storage and retrieval support long-term memory, and networks connect all links. Citigroup emphasizes that Meta's advantage depends on whether these resources can work together to improve task completion rates and lower the cost per successful delivery. The core change brought by cutting-edge AI agents such as Muse and Astra is upgrading one-off Q&A into continuously running task execution processes. Long-term memory, continuous reasoning, and tool invocation allow a single user request to unfold into planning, retrieval, code execution, result verification, and repeated correction, while multi-agent collaboration may also increase parallel tasks and model calls; as more real work enters this process, computing demand is expected to expand simultaneously toward computation, memory, storage, and communication. Within the computing system handling massive inference workloads across the entire process, GPUs take on complex training and diverse inference through programmability; for mature workloads that have completed software adaptation, have large invocation scale, and high utilization, ASICs such as TPUs can reduce inference costs through optimization of computation, memory access, and interconnect. CPUs are responsible for tool operation, browser environments, and task scheduling; HBM and server memory carry model weights, KV cache, and runtime states; SSDs and databases preserve long-term memory, knowledge materials, and task records; optical interconnects and network infrastructure support cross-node and cross-rack model collaboration and data exchange. Therefore, what the agent frenzy catalyzes is demand for heterogeneous computing that can deliver complete, continuously running services: chip performance must be matched with memory bandwidth, storage response, network latency, and power and cooling in order to more successfully completed tasks. When the scope of use and number of tasks in agentic Agent workflows in the AI inference era expand far beyond improvements in single-task computational efficiency, AI demand is expected to be further amplified from single accelerator purchases into full data center investment. This is the key point where the Muse agent product cycle can provide industrial logic support for AI FOMO trading. Before the agent business becomes a firm second growth curve, in Citigroup's view Meta's earnings foundation remains advertising. Citigroup estimates revenue for 20262028 at $254.2 billion, $310.1 billion, and $371.6 billion, respectively, with GAAP earnings per share of $28.11, $36.92, and $46.17. Total advertising revenue in 2027 is approximately $296.8 billion, accounting for about 95% of total revenue; daily active users in the family of apps are expected to grow only 2.8%, while revenue per person on a daily active basis grows 18%, highlighting the importance of Reels and advertising monetization efficiency. Other revenue in the family of apps is expected to increase from $4.4 billion in 2026 to $19.5 billion in 2028, but not all of it can be classified as Muse revenue. The financial model compiled by Citigroup also expects GAAP operating margin to rise from 30.7% to 38.1%, with Reality Labs losses gradually narrowing, reflecting operating leverage where revenue growth exceeds expense growth. Citigroup's $800 target price corresponds to about 22 times 2027 GAAP earnings per share and 11 times EV/EBITDA, equivalent to about 20 times at the current price; at the same time, depreciation and amortization for 20262028 is expected to rise from $29.8 billion to $59.8 billion, and Muse adoption may also increase inference costs. Meta will report earnings after the market close on October 21, and Citigroup expects third-quarter revenue of $63.6 billion, up 24.2%. Citigroup emphasizes that further upside for Meta's stock price requires advertising growth, agent payments, and cost control to be delivered together, while also digesting uncertainties related to competition, privacy regulation, and returns on non-advertising investment.