Wall Street Puts a Price on Muse

Citi says Muse can clear more than $27 billion a year by 2030. About $23 billion of that is transactions, and about $4.5 billion is subscriptions. Deutsche Bank’s upside path runs toward $36 billion, which is about 8% of Meta’s gross revenue.

Those figures sit on top of an ad business that already prints money. Meta did about $200 billion of revenue in fiscal year 2025 and sits near $228 billion in the last trailing 12 months. This is not a startup story. It is a bank stacked on an ad machine, and the model is the punch.

Later in the same discussion, Citi’s take is framed as Meta Muse generating somewhere between $27 billion and $30 billion in the next two to three years. That is a large add-on to the more than $200 billion Facebook is already making, and it is treated as upside rather than a replacement for the core business.

Free Tokens, Then a Cut at Checkout

At Connect, Zuckerberg said Muse is free for a huge number of tokens, and Meta plans to profit from a small fee on the transactions Muse helped close. The free tier is not charity. It is bait for shopping. Invite codes and 100 million tokens sit on top of that as growth candy.

The comparison drawn here is the App Store. Apple did not make its money by selling its own apps. It took a slice off everyone else’s. Muse is being positioned the same way: not as the seller of everything, but as the layer that takes a thin slice off the top when the agent helps close a purchase.

That is also the strategic split with the frontier labs. Meta is not trying to out-meter Sam and Dario. It is trying to own the agent people open first, then skim a cut when that agent buys stuff. A product designer at Substack put the contrast bluntly: ChatGPT and Clod sell tokens, and Meta sells attention. ChatGPT and Clod’s pricing reflects a token business, and their branding is geared toward techies. Muse is being sold as something else.

Early Downloads and Daily Use

The early scoreboard is already loud. Citi cites 6.6 million downloads and about 1.8 million daily active users, and calls Muse a front door to the internet. The growth being shown around the product is described as huge.

On mobile, Muse has outpaced ChatGPT in its early launch. ChatGPT was one of the most explosive growth products in history, and Muse is already being described as ahead of that pace. Downloads of the Muse app in the first 12 days on the market were higher than ChatGPT’s first 12 days.

An exact comparison has been hard. When ChatGPT arrived on mobile, it was available globally, but only on iOS. Muse launched on both app stores. Overall, Muse saw 2.8 million total installs in the first 12 days. Growth has not stagnated. The app moved up from an original number-two position, and the climb has continued.

Availability on both iOS and Android does give Muse something. Even if the result is only close to what ChatGPT did at the start, the reading here is that Zuckerberg is onto something. Few apps have exploded at that level.

Why Didn’t Google Build This?

One reaction to the launch is that Meta shipped the AI assistant Google should have built a long time ago. The number-one response to that thought is distrust: people saying they will never trust Meta with their data. The number-two response is the question itself — why didn’t Google build this? — even as people are already trusting Muse with data.

One stated reason Google did not ship this is that Gemini Spark is not fully released yet. The point is not a full Google story. It is that Google is being described as having missed the boat.

The Advantage Zuckerberg Is Being Credited With

A White House moment is treated as the point where Zuckerberg looked like he knew he had won. Dario is described as not fitting the room of leaders, and Zuckerberg as enjoying it. The argument attached to that moment is about regulation and capital.

Anthropic is described as desperately asking for regulations because tokens are being commoditized by open-source alternatives. The only option left in that framing is to keep cutting token costs to compete. The problem is commitments for hundreds of billions in capital expenditure. At a certain pace, the return on capital does not make sense. The only ways cited to sustain a trillion-dollar valuation are to become entrenched through regulation, or to fight up-market and become vertically integrated.

The second path is where Zuckerberg is given the advantage over both frontier-model companies. Facebook and Meta have over a billion people signed in. That hook has not been fully capitalized yet, but the ability to use it is treated as a major edge.

The other reason is cash flow. Meta could stop spending on the family of apps tomorrow, and those platforms would still generate billions of free cash flow. That is the bankroll for outspending Anthropic and OpenAI, and for subsidizing losses on an AI platform such as Muse longer than competitors can. The Metaverse spend is cited as proof that Meta has already shown it will keep spending. The claim is that this buys time to bleed out Anthropic and OpenAI, especially at the consumer-facing application layer. Trump refusing to give in on Dario and Sam’s request is called a massive win for Meta, and the reason Zuckerberg could not hold his smile.

The Data Center Tax Credit

Meta classified its AI data centers as experimental pilot facilities on its tax return. That lets the NVIDIA chips inside them qualify for a federal research tax credit. Meta’s research credit savings went from $700 million to $3.9 billion. That $3.9 billion is over 10% of the entire federal research credit claimed by one company.

The facilities include a five-gigawatt campus. The critique is that Meta is calling a $50 billion data center campus an experiment. The research credit was designated in the 1980s for companies that tried things that might not work. Meta used it to write off NVIDIA GPUs by the tens of thousands in facilities that power Instagram, WhatsApp, and commercial AI properties.

$3.9 billion in tax savings is described as money the Treasury did not collect from a company worth $1.5 trillion. Meta’s own filings are also cited as a sign the company knows this might not hold up: it increased its IRS dispute reserve by 45%. The posture read into that is to try it and pay later if it gets nailed.

Why the Product Is Being Used

Any leading agentic harness is described as having roughly the same agentic capabilities. What Muse did was abstract away unnecessary complexity. That is why people liked it.

The cute character is part of the story. Kids are cited as starting to use Meta Muse simply because they liked the little warm, fuzzy figure. Codex, Clod code, and Open Clod are said to sound scary. Muse does not. It sounds slight, but it is presented as one of the first reasons given for trying the app.

More important than the mascot is the mental model. Muse is framed as a personal helper with the computer, not a tool that announces itself as code. Meta Muse Spark 1.3 is called an incredible model that people are using heavily because it is being given away for free right now. When the product is free and the fee sits on the checkout, that is the setup being watched.

Competition is still expected to be fierce. The open question is who can outspend the field. Meta stock is described as having fun on the back of this.

The Trust Line

Privacy concerns around Meta Muse are already loud, including a story about someone who shared an address. The view from StartupHakk is that Muse can be useful, but not with production data. It has been tried. It is easy to set up and easy to use, and marketing and other light tasks are the kind of work it is liked for. Personal data, and especially banking accounts that other people are said to be connecting, is where the line is drawn.

The alternative being built there is Openmonoagent.ai, also referred to as openmonoagent.ai. Production work, including Startuphakk Security, Swiftcase Legal, and Startuphakk Corp Trainer, is being run on it, with local inference on local machines so data does not go into the cloud. The platform is described as a terminal-native AI coding agent running entirely on local LLMs, with zero API costs, zero telemetry, and full ownership. A Muse-like feature for building agents is said to be coming, and the repository is open source.

Conclusion

Muse is being treated as a real product, even by someone who has criticized Zuckerberg’s Metaverse choices and called those choices a long run of mistakes. The Wall Street case is a transaction-and-subscription stack on top of an ad machine that already did about $200 billion, with Citi’s path above $27 billion a year by 2030 and Deutsche Bank’s upside near $36 billion, about 8% of Meta’s gross revenue. The product case is free tokens, invite codes, a cute front door people actually open, and a helper that abstracts away the complexity of tools that sound like code.

The competitive case is distribution. Facebook and Meta already have over a billion people signed in, and the family of apps could stop receiving spend tomorrow and still throw off billions of free cash flow. That is the bankroll for subsidizing Muse longer than Anthropic or OpenAI can subsidize a consumer app, especially if tokens keep getting cheaper and regulation does not lock the frontier labs in. The tax treatment of the data centers, with research credit savings rising from $700 million to $3.9 billion, is part of the same picture: Meta has more ways to fund the race than a token business does.

None of that settles the trust question. Downloads, daily users, and a free model do not erase the worry about handing Meta an address, a bank account, or production data. The split in the argument is simple. Muse can be the agent people open first and the thin slice Meta takes at checkout. It can also be the product some people will only use for light tasks, while the work that matters stays on machines they own. Whether the fee at checkout becomes the business the banks are pricing is still open. The installs, the daily users, and the decision to give the model away are the evidence being offered that Meta is already in it.