Meta precipitates the attention of billions of users, and Muse wants to go further and take away users' intentions and transactions.
Author: DaiDai, Frank, MSX Maitong
In Silicon Valley, whenever Zuckerberg prepares to passionately preach about the next generation of technology, Wall Street’s first reaction is often to look at Meta’s capital expenditure table.
From the Metaverse to AI, the script has barely changed.
Especially in 2026, while Meta continues to purchase GPUs, build data centers, and mine model and agent talents on a large scale, it also pushes its annual capital expenditure guidance to US$130 billion to US$145 billion. After the money is burned to this level, the secondary market will naturally wonder, how can it earn back this money?
The previous standard answer was very trouble-free - feed the model to the recommendation algorithm, so that the ads can be delivered more accurately, and the huge money printing machines of Facebook and Instagram can continue to improve efficiency.
The problem is, this logic also has its ceiling.
For a giant with annual revenue of hundreds of billions of dollars, if hundreds of billions of investments in AI are ultimately just to make information flow ads more accurate and click-through rates higher, then why should the market give you new valuation imagination?
That’s why Muse is interesting.
For the first time, Meta’s increasingly large AI investment in the past few years has been connected to a product that is outside of advertising and at the same time close enough to real consumption and transactions. From model capabilities, to Agent products, to transactions and commercialization, at least logically, a complete path has begun to emerge.
This is why I really don’t want to praise Xiao Zha, but I have to say that Muse’s move was extremely sophisticated.
1. Meta’s AI is not just a cost center
Let’s look at the recent market first.
From the end of August to September 21, Meta’s stock price rose from more than $570 to $741.25, a cumulative increase of nearly 30%, of which it rose 11.34% on September 21 alone.
Of course, it’s not accurate to attribute all of that 30% to Muse. The Meta market started before the release of Muse, and the previous market was more about repairing the market's pessimistic expectations for capital expenditures, cost growth, and profit margins.
But after Muse went online on September 8, the discussion around Meta AI did begin to add another layer of content: Is it possible that this money will really grow a second growth curve?
The difference lies here.
Muse is not built according to the logic of a chatbot (Chatbot). It is a very pure personal AI Agent. After authorization, it can jump to web pages, call interfaces, fill out forms, and compare prices across platforms. Even if you lock the phone screen, it will run errands for you in the background. When it comes to making a payment or sending a key email, it will pull you out to confirm.
In other words, AI in the past was more like answering "What should I do?" What Muse wants to do is to take a step forward - "Okay, I will do it."
This change seems to be just from Chatbot to Agent, but the business significance is completely different. After all, once AI can take over actions, it will naturally enter the entrances closest to money such as search, shopping, reservations and payments.
And this is precisely where Meta really deserves to be revalued.
It does not necessarily require Muse to be the largest model in the world in Benchmark. For the vast majority of ordinary users, it does not matter whether the model score is two points higher or three points lower. It only depends on whether it can actually complete a task after we give it to it.
This is similar to our judgment on Google two days ago. When the capability gap between the underlying models begins to narrow, companies that truly have access, product ecology and distribution capabilities will have a better chance of reaping dividends in the second half of AI (extended reading "Gemini 4.0 is so powerful that it "swipes the screen", Google's AI counterattack is coming?").
From this point of view, Muse is not Meta's retake of the exam "whose large model is the strongest". It simply changes the exam venue: when the underlying model capabilities are basically sufficient, I will take the lead in inserting Agent moisturizing and quietly into the daily life of my billions of user pools.
But when we arrived at this examination room, the deck of cards in Meta's hand began to look very different.
2. Intention and interception: Why is Amazon defensive and Shopify embracing?
To do this, Meta has a trump card that others cannot envy - native distribution.
AI startups worry about the cost of acquiring customers every day, but Meta owns Facebook, Instagram, WhatsApp, Messenger, plus smart glasses that are still being pushed forward, and billions of people are already enjoying its products.
But if there are just "more people", it is not enough to explain the value of Muse.
The really interesting thing is that it has the opportunity to push what Meta originally mastered from the past "Attention" to "Intent".
For example, if we saw a video of a soymilk maker on Instagram and paused it for a few seconds, the algorithm might have judged that you were recently interested in small kitchen appliances, so it would push a few more related ads to you, and Meta would have earned exposure fees.
But if you really want to buy it, you will most likely close Instagram, open Amazon/Temu, search, compare prices, read reviews, and place an order.
In other words, Meta has mastered "what you may like", but the real consumption decisions and transactions still occur on other people's territory.
And Muse can completely reconstruct the path - we only need to directly say to Muse: "Look for a soymilk machine that costs less than 100 dollars, is easy to clean, and is suitable for young couples." The subsequent screening, comparison, confirmation, and payment are all completed in the Agent's channel.
There may seem to be only one step between attention and purchase intention, but in the business world, this step is worth a lot of money.
After understanding this, you can understand why major manufacturers have mixed attitudes towards Muse.
Amazon is facing a major enemy and has quickly restricted Muse from placing orders on behalf of customers. The reason for moving out is nothing more than authorization compliance and privacy security, but Shopify has given the green light and hopes that Shop Pay can be directly embedded into Muse's transaction flow.
It’s purely decided by profit:
Amazon itself is a closed center for search and transactions. Its most valuable thing is not the cloud service, but the moment when consumers make a decision when buying something. Amazon must be opened. If everyone relies on external agents to make purchases in the future, Amazon will be forced to become a mere back-end shelf, and the in-site search bidding and advertising systems it has worked so hard to develop will have to be removed.
Shopify is a water seller. Regardless of whether the order comes from Instagram, is searched from Google, or falls from the Muse Agent, as long as it uses its payment settlement and uses its merchant channel, it can steadily earn commissions;
This is the most realistic power restructuring in the Agent era. Whoever is closest to the user's final decision controls the "Strait of Hormuz" on the e-commerce map.
Of course, we cannot be blindly optimistic here. There are many people who use IG, but at this stage there are still only a few people who are willing to entrust their email passwords, calendar permissions, and credit card numbers/CVV to an AI.
After all, if GPT or Gemini answers a wrong question once, it will be considered as the AI has failed again. However, if the Agent helps you buy the wrong quantity, book the wrong hotel, or sends an email that should not be sent, the user's trust in it may be gone in one go.
Therefore, if Meta wants to eat the fattest piece of meat "intention", it must first solve a very simple problem: Do you dare to really leave things to it?
3. Subscription is just a small amount of money, the real big business is in the transaction
At the end of the conversation, we still have to settle the score. How can Muse help Xiao Zha regain his health?
The easiest thing to understand now is the subscription fee. Muse offers a free version, as well as paid tiers of US$20 and US$100 per month. Assuming that 10 million users can really pay US$20 per month for a long time in the future, the revenue will be US$2.4 billion a year.
For most software companies, this is already a very big business, but Meta does not quench the thirst at all.
You know, Meta’s single-quarter revenue in the second quarter of this year reached 60.8 billion U.S. dollars, and advertising alone accounted for 59.3 billion U.S. dollars. It can’t fill the huge hardware black hole just by selling members.
Therefore, in the view of Maitong Research Institute, subscription is not the most noteworthy part of Muse. What is truly imaginative is the transaction commission, which is the main meal.
If an Agent becomes a digital steward for hundreds of millions of people, the total transaction volume (GMV) it leverages will theoretically far exceed its own membership fee. At this time, the space for commercialization will be completely opened: commissions for transaction matching, diversion fees for merchants who want to be ranked at the forefront of recommendations, and profit sharing from payment channels...
In the past, when merchants advertised on Meta, they spent money to buy an exposure probability that "you might buy." But in the future, Agents will directly hold deterministic orders that "users are ready to pay for." Will merchants be willing to pay a commission for the transaction? Are payment companies willing to take a cut of Agent Checkout? Are merchants willing to purchase certain business services from it in order to become a preferred option?
These are the places more worth imagining.
However, this also raised a red flag for Meta.
The reason why everyone uses your Agent is because they believe it is a "silly assistant" who is on their side and helps us maximize our interests. But if one day Muse crazily recommends a certain hotel, and everyone discovers that it is not because that hotel is suitable, but because it gives Meta the highest rebate, then the trust foundation of this product will collapse on the spot.
Therefore, for Meta, the deeper the commercialization goes into the transaction side, the more it has to deal with a contradiction, that is, how to make money from users' intentions without making users feel that their intentions are being sold.
This is why Meta must repeatedly emphasize that Muse's private conversations and secure sandbox (Secure VM) will not be directly connected to the advertising system - it must establish a new charging credit system independently of advertising.
The reason why this matter is so important ultimately comes back to Meta’s increasingly exaggerated AI bill. Although revenue increased by 28%, total costs and expenses soared by 55%, and operating profit margins were directly hit from 43% to 31%. Operating cash flow in a single quarter was 31.8 billion, while capital expenditures plus lease principal ate up more than 31 billion, and free cash flow was squeezed into an extremely narrow range of US$784 million.
Although there are one-time fees mixed in, the fact is very clear that Meta is investing cash back into AI infrastructure at an unprecedented rate.
If these graphics cards are burned to the end and only add a few more clicks to the information flow advertisement, then it is at best patching the original business and cannot support the next round of valuation reconstruction.
But if Muse really opens up a transaction entrance led by Agent, Meta's story in the eyes of the capital market may be to use AI to create a new entrance, completely changing from an "advertising intermediary buying shovels" to a "scheduling station for the next generation of Internet life."
These are completely two sets of valuation models.
write at the end
I have to admit that for Meta, Muse has begun to connect the AI bills that have seemed increasingly scary in the past few years to a concrete enough business path.
Meta used to have the largest attention machine in the world; what Muse wants to do is continue to follow the attention, from knowing what you like, to understanding what you really want, to getting things done for you.
If this chain really goes through, Meta may not just get a ChatGPT competitor, but a new Internet entrance.
Of course, the stock price has risen to more than $700, and the market has given a lot of applause in advance.
The wins and losses on the model list may soon pass, but how to calculate this account is what is really worth trading in META in the next stage.