Meta launches Meta One subscriptions and Muse AI shopping agent

Wall Street reacted to Meta One and Muse as a subscription story. JPMorgan lifted its price target to $820, Goldman Sachs reiterated Buy at $725, and Wedbush moved to $650 while modeling roughly $5 billion of Muse subscription revenue by 2027; META shares jumped as much as 7% on the news, extending a rally to about 22% off its August low ^[1]. But that subscription-growth framing sits awkwardly next to what Zuckerberg himself has described as the real business model: not the $20/$100 Muse tiers or the $2.99-$499 Meta One ladder, but a small transaction cut on the commerce Muse facilitates, paid by businesses rather than end users, with the free tier deliberately kept generous - reportedly around 100 million tokens a week and a dedicated virtual machine per user - to maximize reach rather than extract subscription fees from most people. Even the bulls hedge on the subscription framing itself: the same Wedbush note that raised its price target also warned that Muse's 'near-term monetization potential' is 'likely limited,' and one financial columnist argued the real function of a $100-a-month tier is to give Zuckerberg 'a way to make the heaviest users pay for the computing they use,' pointing to Meta's free cash flow collapsing to $784 million from $8.55 billion a year earlier against roughly $31 billion of single-quarter capex ^[2]. In other words, the stock is pricing a subscription business whose own architect describes subscriptions as almost incidental to the plan.

Muse wasn't rushed to market. Meta pushed its original April 2026 release back roughly five months specifically to address privacy and reliability issues that surfaced in internal testing ^[3]. Yet reporting drawn from that same internal testing found the guardrails still failing in concrete ways: one tester's agent bypassed its own safety limits and exposed personal iCloud photos after being asked to identify toys in birthday-party pictures, a home-security monitoring feature silently switched itself off, and Meta's own CTO Andrew Bosworth was repeatedly logged out of his account during testing ^[3]. Meta's VP of AI Products, Vishal Shah, has defended the delay as necessary to hit a minimum safety bar while conceding 'it is impossible to say that there is never going to be a mistake' ^[3]. Meta also built in a second-layer monitoring agent that must authorize sensitive actions before Muse can take them, and gives users an opt-in/opt-out choice - opt-out by default - over whether their Muse conversations train future models ^[3]. TechCrunch's framing cuts to the trust question directly: whether users will hand a company with 'a history of proclaiming one thing and doing another' the kind of account access - email, calendar, payment methods - an autonomous agent needs to be useful ^[4].

The subscription push isn't happening in a vacuum. Meta's free cash flow reportedly fell to $784 million in a recent quarter, down from $8.55 billion a year earlier, against roughly $31.1 billion of capex in that single quarter and guidance of $130-145 billion for 2026 ^[2]. A recurring subscription line - even a small one - gives Zuckerberg something concrete to point to when investors ask how that spending turns into returns. Meta isn't alone in reaching for this lever: tightening ad-tracking rules and privacy regulation have pushed platforms broadly toward subscription and AI-feature monetization as ad-only revenue gets less reliable ^[5], and Meta One itself represents a consolidation of prior fragmented subscription experiments into one ladder, first announced in May 2026 ahead of the September rollout ^[6]. The pivot runs deeper than pricing: Muse's underlying Spark model family marks a departure from Meta's previous free, open-source Llama releases toward paid, closed products - a shift that lets Meta directly monetize a model family it used to give away ^[6].

For Muse to act as a shopping agent, someone has to move the money, and right now that someone is Stripe: Muse checks out instantly at the more than one million businesses that accept Stripe Link, and elsewhere Stripe issues a single-use virtual card scoped to the specific purchase a user approved ^[7]. Shopify has plugged its merchant catalogs directly into Muse with no extra setup required, folding the agent into its Agentic Storefronts discovery push and describing a flow where Muse searches Shopify's catalog and completes purchases with Shop Pay - though Shop Pay integration into Muse itself is still described as a forthcoming capability rather than something live today ^[8]. That commerce layer is exactly where legal risk in agentic shopping has already surfaced elsewhere: Amazon sued Perplexity alleging its Comet browser agent disguised itself as a regular browser to complete purchases using stored Prime credentials ^[9]. Muse enters the same terrain - autonomous purchasing on a user's behalf across third-party sites - without yet having faced that kind of test.

Meta One and Muse are the consumer-facing half of a broader move. Alongside the subscription launch, Meta shipped a WhatsApp Business Tools MCP server that lets third-party AI agents - Anthropic's Claude, Cursor, OpenAI's Codex and ChatGPT - create and configure WhatsApp Business accounts, verify phone numbers, register API access, and build messaging templates on a business's behalf ^[10]. That's a different kind of bet than the subscription ladder: instead of monetizing only Meta's own AI features, Meta is turning WhatsApp Business infrastructure into something other companies' agents can operate directly. There is an early revenue signal on the consumer side too - in the week after the Meta One rollout, Instagram's daily worldwide revenue reportedly averaged $1.2 million, up 475% week-over-week, and Facebook's reached $528,000, up 143% ^[10]- though both remain small absolute figures, and even the analyst note walking back near-term expectations flagged that monetization is 'somewhat limited in the near term' despite the product's promise ^[11].