This is the Facebook / Instagram / WhatsApp company. The investment case: their AI assistant becomes something billions of people use every day, and Meta figures out how to get paid for it: ads, subscriptions, or a cut of everything the assistant helps you buy. The catch: at ~$709 you're already paying for a lot of that future. This is a "great company, picky about price" situation.
The base case clears the 12% growth bar but not the 30%. Hence selective. Only the bull case touches 30%, and only just.
Before any AI story: the core business is still compounding. Ad revenue grew 28% on a roughly $244 billion run rate, and the stock trades around 21 times next year's earnings, the cheapest of the mega-cap AI names. That cash flow, about $68 billion a year, funds the entire AI buildout without raising a dollar of outside money. Nobody else in AI can say that.
In September 2026 Meta launched Muse, a personal AI agent living inside WhatsApp, Instagram, and Facebook: a distribution surface of 3.6 billion daily users that no standalone assistant can match. The interesting part is not the chatbot. It is the wallet. Muse can approve one-time payment cards and buy things on your behalf, which puts Meta in the middle of transactions, not just ads. At just 1% of daily users paying a $30 blended monthly rate, that is roughly $13 billion a year in subscription revenue at software margins. The bigger prize is a cut of agentic commerce, a market forecasters size anywhere from hundreds of billions to trillions. Meta is now pushing the agent into the physical world: open-source hardware kits that turn cheap circuit boards into Muse gadgets, a TV stick, a desk buddy. The idea is to make Muse the operating system everything plugs into, the way Android became the default for phones. OpenAI noticed. Three weeks after Muse launched, it announced Dots, its own always-on agent with a strikingly similar look. The race is real, but so is Meta's head start: 3.6 billion daily users and a free product against a $100-a-month gated rollout. The deeper bet is the Apple playbook. OpenAI and Anthropic sell intelligence you have to configure, down to reasoning toggles. Muse hides the model entirely and sells the outcome. That is the product shape that crosses from enthusiasts to everyone, provided the hidden intelligence stays good enough. None of this is revenue yet. The first proof arrives October 28, when Meta reports how many people actually pay.
An agent that can spend your money lives or dies on trust, and Meta's history here is checkered. So they overbuilt the security: every user gets an isolated virtual machine, the agent browses in a sandbox, it never sees your real passwords, and a system called Sentinel watches everything leaving the machine. Arguably the most serious agent security architecture in the industry. The binding constraint is whether people believe it, given who is selling it.
Here is the underappreciated angle. The AI labs run a tollbooth model: raise billions, sell intelligence at a premium. Meta breaks that model twice. First, ad profits fund the data centers outright, no fundraising treadmill. Second, its open-weights Llama models commoditize the model layer itself. If AI gets cheap, the labs' business model cracks and Meta's gets stronger. What the labs fear is not that Muse is good. It is that Meta can afford for AI to be cheap.
Alexandr Wang gave a friendly 68-minute interview on October 7. His numbers, unverified and talking his own book: Muse hit 5 million downloads in 22 days, which he frames as the biggest launch since ChatGPT. If directionally right, that is strong top-of-funnel heading into the October 28 earnings checkpoint. More interesting: "Muse Confidential," a more-encrypted tier designed by WhatsApp's encryption creator, aiming for WhatsApp-like guarantees against government data requests. That reads as the likely paid product, which makes the paid-conversion and ARPU disclosure at earnings the thing to watch. On safety: every user gets an isolated virtual machine, a separate "sentinel" agent watches the main agent, and consent prompts appear on new website contacts. On money: still free today, ads "not the current focus," and business models floated around purchases and savings the agent generates. His honest admission: agent memory is "still largely unsolved." None of this changes the conditional verdict, but the paid tier is the next real data point.
The math: the base case reaches about $1,350 by 2031, roughly 15% a year. That clears the 12% growth bar comfortably and fails the 30% growth bar. Only the leader case, around $2,540, just barely clears 30%. Hence selective: a great company that is picky about price, interesting on a pullback toward $625, fairly valued where it trades. The October 28 earnings call is the first checkpoint. Watch for paid conversion numbers on the new subscription tier.
Meta is roughly doubling data-center spending toward $130 to $145 billion while free cash flow is deeply negative and getting worse, and it has paused share buybacks to preserve cash. Even if the AI bet works, depreciation from this building spree will weigh on reported profits for years. The October earnings call is the first real checkpoint on whether any of this spending turns into revenue.
About 98 percent of revenue still comes from advertising, and ad spending runs in cycles. If advertisers pull back at exactly the moment Meta is spending the most on AI infrastructure, there is no second revenue line to cushion the fall. The subscription and AI commerce bets are 2027 stories at best. Everything depends on the ad engine holding up while the company rebuilds itself around AI.
The AI assistant lives or dies on trust, and this company's privacy history is checkered: a 200 million euro fine in Europe, the Apple privacy change that cost an estimated $10 billion in ad revenue, the abandoned Libra currency project. An assistant that sends the wrong email or buys the wrong thing would damage the brand in a way a bad ad product never could. Meanwhile OpenAI, Google, Apple, and Amazon all field their own assistants, monetization is unproven, and no user numbers have been disclosed.
Operating margins fell from 43 percent to 31 percent in four quarters. The AI talent war means enormous pay packages, and the Reality Labs division loses roughly $4.6 billion every quarter. None of that corrects itself. The company can afford it because the ad machine is so profitable, but each point of margin lost is a point of cushion gone.
Around $709 the stock sits in the fair-value zone, which is exactly why the verdict stays conditional. The math only turns positive on a pullback toward $625 or below, where the expected return clears the hurdle with room to spare. Regulatory scrutiny is a permanent background hum with this company, but it rarely decides anything.