By DaiDai and Frank, MSX Maitong
When Mark Zuckerberg starts preaching the next big thing in Silicon Valley, Wall Street’s first instinct is often to check Meta’s capital spending table. That pattern has held from the metaverse to AI.
The PANews analysis says that by 2026, Meta was still buying GPUs at scale, building data centers, and hiring model and agent talent, while also lifting its full-year capital expenditure guidance to $130 billion-$145 billion. Once spending reaches that level, the market’s question becomes blunt: how does the company earn that money back?
The old answer was straightforward. Feed stronger models into recommendation systems, sharpen ad targeting, and keep the Facebook and Instagram cash machine running more efficiently. The article argues that this logic has limits. For a company generating hundreds of billions of dollars in annual revenue, spending well over $100 billion on AI only to make feed ads slightly more precise is unlikely to create a fresh valuation story.
That is why Muse stands out in the piece. It is presented as the first product that connects Meta’s expanding AI investment to something outside advertising while staying close to real consumption and transactions. From model capability to an agent product to commerce and monetization, the authors say a more complete path is starting to take shape.
Stock gains revived the debate over whether AI can produce a second growth curve
The article points to Meta’s recent share performance. From late August to Sept. 21, Meta stock rose from a little over $570 to $741.25, a gain of nearly 30%. On Sept. 21 alone, the stock climbed 11.34%.
At the same time, the authors caution against attributing that entire move to Muse. Meta’s rally had already started before the product launch, and the earlier part of the rebound was described as a repair trade after pessimism around capital spending, cost growth and margins.
What changed after Muse went live on Sept. 8, according to the article, was the tone of the discussion around Meta AI. Investors began asking whether all that spending could actually produce a second growth curve beyond ads.
The piece says Muse is not built around the standard chatbot logic. It is framed instead as a personal AI agent. Once a user grants permission, it can enter websites, call interfaces, fill out forms, compare prices across platforms, and continue working in the background even when the phone is locked. It only pulls the user back in when a payment or a key email needs confirmation.
That shift is summed up in simple terms. Earlier AI systems mostly answered the question, “What should I do?” Muse is trying to move one step further: “Okay, I’ll do it.” The article argues that this is not a minor product tweak. Once AI can take over actions, it moves naturally into search, shopping, booking and payments, the places closest to money.
The authors add that Meta does not necessarily need Muse to become the world’s top benchmark model. Most ordinary users do not care whether a model scores two points higher or lower. They care whether the task they hand over actually gets done.
The article draws a parallel with Google, saying that as the capability gap between foundation models narrows, companies with entry points, product ecosystems and distribution may be better positioned to benefit in the next phase of AI. In that framing, Muse is not Meta sitting for another exam on who has the strongest large model. It is Meta changing the venue and trying to place an agent quietly inside the daily routines of billions of users once base model performance is already good enough.
From attention to intent
The article says Meta’s biggest advantage here is native distribution. Many AI startups are still struggling with customer acquisition costs, while Meta already controls Facebook, Instagram, WhatsApp, Messenger and a push into smart glasses. Billions of users are already inside its products.
But scale alone is not the full story. The more interesting point, the authors argue, is that Muse could let Meta move from controlling “attention” to getting closer to “intent.”
The piece uses a simple example. A user watches a soy milk maker video on Instagram for a few extra seconds. The algorithm may infer interest in kitchen appliances and serve more related ads, allowing Meta to earn exposure revenue. But if that user actually wants to buy, the next step is likely to happen elsewhere, on Amazon or Temu, where the user searches, compares, reads reviews and places the order.
In other words, Meta has historically known what a user might like, while the actual purchase decision and transaction happened on someone else’s turf. Muse, in the article’s telling, could restructure that path. A user might simply tell Muse to find a soy milk maker under $100 that is easy to clean and suitable for a two-person household. The filtering, comparison, confirmation and payment could then run inside the agent flow.
The distance between attention and purchase intent may look small, but the article says that step is worth a great deal in commercial terms. That is also why major platforms are reacting so differently.
Amazon, the piece says, has treated Muse as a serious threat and moved quickly to restrict agent-assisted ordering, citing authorization compliance and privacy security. Shopify, by contrast, is described as welcoming the shift and wanting Shop Pay embedded directly into Muse’s transaction flow.
The authors tie that split to business structure. Amazon is a closed center for search and transactions. Its most valuable asset is not only cloud services, but the fact that consumers must open Amazon at the moment they make a purchase decision. If shopping is increasingly handled by outside agents through a single prompt, Amazon risks being reduced to a back-end shelf, while its internal search bidding and ad system lose leverage.
Shopify sits in a different position. Orders can come from Instagram, Google or Muse. As long as payment settlement and merchant rails run through Shopify, it can still take a cut. The article describes this as a practical power reshuffle in the agent era: whoever sits closest to the user’s final decision controls the key choke point in e-commerce.
Trust remains the hard part
The analysis does not present the story as a straight line upward. It notes that while many people spend time on Instagram, far fewer are ready, at least for now, to hand over email passwords, calendar permissions, credit card numbers and CVV details to an AI system.
The difference between a chatbot mistake and an agent mistake matters. If GPT or Gemini gives a wrong answer, users may shrug it off as another AI error. If an agent buys the wrong quantity, books the wrong hotel, or sends an email that should never have gone out, trust can disappear in a single incident.
That is why the article says Meta’s first challenge in capturing the value of “intent” is basic: are users actually willing to let the system handle things on their behalf?
Subscriptions are small compared with transaction revenue
The article then turns to the numbers. The easiest revenue model to understand is subscription pricing. Muse is said to offer a free version as well as paid tiers at $20 and $100 per month. If 10 million users were eventually to keep paying $20 a month, that would translate into $2.4 billion in annual revenue.
For most software companies, that would already be a major business. For Meta, the authors say, it would not be enough. The article notes that Meta generated $60.8 billion in revenue in the second quarter this year, including $59.3 billion from advertising alone. Membership fees by themselves would not fill a spending gap of this size.
MSX Maitong therefore argues that subscriptions are not the most important part of the Muse story. The bigger opportunity lies in transaction take rates. If an agent becomes the digital manager for hundreds of millions of people, the gross merchandise volume it influences could, in theory, far exceed its own subscription revenue. That would open up room for matching commissions, merchant traffic fees for preferred placement, and revenue sharing on payment channels.
The article frames the difference this way: merchants used to buy a probability of exposure on Meta, a chance that a user might purchase. An agent could hold something much closer to a confirmed order from a user already prepared to spend. In that setting, the authors ask whether merchants would pay commissions for conversions, whether payment companies would share revenue for agent checkout, and whether merchants would buy services to become a preferred option.
Monetization creates a conflict Meta will have to manage
The same path also creates risk. People use an agent because they believe it is on their side and trying to maximize their interests. If Muse starts pushing a hotel because that hotel offers Meta the highest rebate rather than because it is the best fit, the trust foundation breaks down immediately.
That leaves Meta with a central tension, according to the article: how to make money from user intent without making users feel that their intent has been sold.
The authors say this is why Meta has to keep stressing that Muse’s private conversations and Secure VM will not be directly connected to the ad system. In their view, Meta needs to establish a separate charging and trust framework outside advertising.
The importance of that point comes back to Meta’s AI bill. The article says revenue rose 28%, but total costs and expenses jumped 55%, pushing operating margin down from 43% to 31%. Quarterly operating cash flow reached $31.8 billion, while capital expenditures plus lease principal consumed more than $31 billion, squeezing free cash flow to $784 million.
The piece notes that one-off charges are part of the picture, but says one fact is already clear: Meta is reinvesting cash into AI infrastructure at a pace not seen before.
If all those GPUs and infrastructure purchases end up doing little more than adding a few extra clicks to feed ads, the spending would amount to patching the existing business rather than supporting a new valuation framework. If Muse can open a transaction entry point led by agents, the article argues, Meta’s capital-markets story could shift from an advertising intermediary buying huge amounts of compute to a dispatcher for the next generation of internet life. The authors call those two very different valuation models.
A more concrete commercial path, but still unproven
The article closes by saying Muse begins to connect Meta’s increasingly large AI bills to a more concrete business path. Meta’s old strength was the world’s largest attention machine. Muse is trying to extend that chain, from knowing what users like, to understanding what they actually want, to getting the task done for them.
If that chain works, Meta may gain more than a ChatGPT rival. It may gain a new layer of internet entry point.
Still, the article notes that Meta shares are already trading above $700 and the market has already offered plenty of applause. Wins and losses on model leaderboards may fade quickly. The more important question for the next phase is how Meta turns this AI spending into returns.