(Speedwell Members click here for a PDF version of this post)
3Q26.
Meta reported another stellar quarter with Family of App revenue growth up 26% y/y, driven by impression growth of 10% and ad price growth of 14%. Family of App Other Revenue (non-advertising) grew 59% y/y, which is largely from WhatsApp Business Paid Messaging and subscription revenues.
Total operating income was up 11% y/y, after backing out $9.9bn in legal charges from their teen mental health settlement. The total settlement value is $18.6bn, with part of the remainder payment contingent on YouTube and TikTok agreeing to similar safety restrictions.
As mentioned in our 4Q25 update, “They seem to be one of the clearest beneficiaries of AI”, and 3Q26 has only extended that claim. Instagram and Facebook Reels reached a global all-time high of time spent, growing double digits y/y. Even Threads engagement grew 20% y/y. Key to these engagement improvements have been advances in AI that can recommend better content.
Business Update.
The big story for Meta isn’t their ads business, which to be clear has been very strong, but rather their progress with their first successful AI consumer product, Muse.
On the call they announced that 10mn users have set up Muse and 25% of them use the app daily. Mark Zuckerberg noted that this is twice the usage level that Threads enjoyed when it was launched.
More impressive was that there seems to be real signs of adoption for users using Muse to purchase items, which is a primary way they plan to monetize the AI agent service.
While it is hard to extrapolate future consumer behavior from this small subset, it is a sign that users are open to letting an agent make purchases for them. Although it should be noted that they currently do not charge a fee; the introduction of which could dissuade users from purchasing through Muse. On the other hand, it isn’t rare for a consumer to be surcharged for credit card usage at some small businesses, and they readily accept that cost rather than use a debit card or cash. We could move to a future where “AI Agent Transaction fees” are common and either the business eats them in order to get more purchases or the consumer faces an additional “convenience fee” for using the AI agent. It is also possible that with a paid subscription, the AI Agent provider waives these fees when they would be borne by the consumer. All of this is of course speculative, but the early data shows that AI Agent purchasing is something consumers are interested in, and so it is just a matter of figuring out how much they are willing to pay for that service.
Of course, the sticking point with AI Agent purchases is that large businesses that are threatened are unlikely to willingly go along with this plan. Amazon already restricts Muse access and we can expect similar restrictions from businesses that feel they have more to lose than gain from AI Agent purchases. Notably though, Walmart so far welcomes it.
While consumer AI Agent adoption is early in the S-curve, this could be a new vector that ecommerce businesses leverage to gain share over incumbents (in ecommerce, oddly enough, Walmart is the new entrant). Amazon has spent decades building the consumer top of mind position, so that whenever someone wants to purchase something they think “Amazon”, and click the app, type in their search, and click buy. This is already an extremely easy buying experience and combined with fast shipping and the trust that comes with an Amazon purchase, they have a strong value prop even if they restrict AI Agents. But this advantage only extends to high-intent purchases when the buyer already is pretty sure of what they want.
The risk comes for the category of commerce where you have an idea of what you want, but are not sure exactly what you want. Take for example a computer adapter or supplement. With these purchases there is more research needed and sifting through Amazon’s result page and clicking through a bunch of different listings is a suboptimal experience. This is the ideal environment for an AI Agent like Muse because it can run a broader search, pull up research and blogs, and then return results. At that point maybe some users do then plug that answer back into Amazon to purchase it, but others might just buy it right in the Muse app. Walmart is happy to let Muse connect to its inventory because they think they have more incremental ecommerce purchases to gain than lose in this circumstance. There will be many other e-retailers that are in a similar position, especially those that are hard to discover. (Amazon has their own AI Agent, Rufus, but as of today it leaves much to be desired in terms of experience.)
Even if Muse takes a small transaction fee, that still might be cheaper for a merchant than paying Amazon’s fees, especially after including the sponsored ads that many ecommerce businesses have to pay for visibility. So businesses will simply think of the ROI on demand from AI Agents with a small fee versus listing on a traditional market place, which itself has its own fees—and it is quite possible that the AI Agent fee will be smaller. Brands are not beholden to one strategy and are likely to do both. Why not see if Meta’s Muse can generate incremental demand?
What matters most is consumer preferences. Amazon is already fulfilling many consumer preferences by being very convenient and easy to shop on, but as mentioned that is not the case for more “discovery” purchases. If consumers build a preference to have a “trusted” AI Agent shop on their behalf for some purchases, over time that may become a habit and some purchases that used to go through Amazon could go through Muse. Now this isn’t to say Amazon is necessarily in trouble, their logistics network, trust, return points, and decades of consumer habit are hard to displace, but just that it isn’t that farfetched that Muse could build a real business through ecommerce purchases.
On the call Mark explicitly said that he believes that Muse ARPU could exceed FOA revenues, which is no doubt a tall task.
Not all of this will come from ecommerce purchases though. Just like other AI companies they plan to charge for more advanced model access as well as extra usage through subscriptions.
An even bigger opportunity than consumer AI Agents though could come from small businesses. One of the reasons why early Facebook had so much success with advertising was that tens of millions of businesses had Facebook pages. This meant that Facebook could sell advertising services to these businesses with simple Post Boosts without having to have a sales force to acquire them. The fact that businesses already were on Facebook made it a much easier sell to have them advertise on Facebook. Similarly, already 500mn small businesses are on Meta’s apps, which gives them a distribution channel to directly sell them small business services.
The example used in the call was a small maple syrup producer who used Meta AI Business Assistant to improve their campaign messaging and as a result saw a 57% increase in purchases. It is easy to imagine a prompt showing up to an advertiser (small business) that tells them how to improve their advertising. Then after they see the results they would be much more open to further assistance. Maybe the Business Agent then says how they can help schedule deliveries, optimize inventory, or take over customer service. This is a much more natural way to introduce AI to most businesses: first with a task with a clear ROI and then gradually more over time. The early proof point from the improved effectiveness of advertising is likely to build trust to let the AI Agent do more.
While eventually there may not be a big difference between Meta’s AI or Anthropic’s, the real question will be who can productize AI and distribute it to millions of consumers and businesses first, and Meta is showing they are a leading contender for that.
It is true that consumers have a general distrust of Meta and it consistently ranks as one of the least trustworthy tech companies, but if Muse’s limited success shows anything, it is how forgiving people can be if you create a product that creates value for them.
Now all of this though comes with immense compute costs. They have only rolled out the service in the US and Canada in part because they need to bring costs down before Muse is scalable to a billion users. They talked about how they expect cost efficiency to improve by more than 10x in the next year, to help them scale the business.
Nevertheless, even then, they are making a big capex bet. 2026 capital expenditures are expected to be $115-135bn (including financing leases) and it is expected 2027 will be even higher. Despite having a cash generating machine in advertising, it looks like they will still need to raise more money. They announced the introduction of C shares, which will be non-voting and dividend out at a 1:1 ratio to all existing shareholders (similar to Alphabet’s GOOG and GOOGL shares). This will have the effect of splitting the stock price in half, but also makes it easier for them to raise capital in the future without diluting Mark Zuckerberg’s control.
They are guiding 4Q revenues to 24-29% y/y, which would put total 2026 revenues at around $255bn. At a combined margin of 41% (2025 level), that is $105bn in EBIT. This figure still includes around $18bn in Reality Labs losses. If we back that out and apply a 20% tax rate, that comes out to 2026 EPS of $38 (excludes one-time legal charges too). On an after-hours price of $720, that is 19x TTM core earnings.
The Reality Labs losses are likely to continue for some years (guiding to $15bn of losses next year) and margins will continue to be compressed as elevated capex runs through the P&L in the form of depreciation, but nevertheless many investors will not find this to be a demanding valuation.
The real risk is if there is a larger AI slowdown and they have committed to spending billions on capex that turns out to not have the associated revenue opportunities they are counting on. Is it possible that the recommendation model and ads targeting gains start to plateau, even with more advanced AI and compute? And perhaps the early success of Muse is only a sign of product-market fit, but not Meta’s dominance of this area with Apple, Google, and the AI Labs having yet to come out with a similar product (that they are no doubt working on).
Nevertheless, so long as their core FOA business continues to hum, it significantly derisks the need for the AI Agent endeavors to work out. And for the first time in a long time, it seems like Meta has a real shot of building a business outside of advertising.
For further reading, check out our Meta Extensive Research Report here.
The Synopsis Podcast.
Follow our Podcast below! We will talk more about Meta on there soon!
Speedwell Research Reports.
Become a Speedwell Research Member to receive all of our in-depth research reports, shorter exploratory reports, updates, and Plus members also receive Excels.
(Many members have gotten their memberships expensed. If you need us to talk with your compliance department to become an approved vendor, please reach out at info@speedwellresearch.com).
*At the time of this writing, one or more contributors to this report may have a position in stocks mentioned. Furthermore, accounts one or more contributors advise on may also have positions in stocks mentioned. This may change without notice.
Leave a Reply