Meta Update: Underpriced AI Opportunity or Costly Gamble?

Get smarter on investing, business, and personal finance in 5 minutes.

Business Update

This newsletter is adapted from a recent video Drew Cohen did on Meta.

Meta stock rose 42% over the past few months before falling 19% after the company reported 3Q26 earnings.

This week’s Five Minute Money is going to be a short update on Meta!

Business Update.

Family of apps revenue grew 26% y/y, reaching ~$64bn billion for the quarter.

This growth rate represents a meaningful acceleration from 2022 and 2023, when growth had slowed considerably.

Growth climbed into the low 30% range during the recovery that followed and now sits in the mid-20% range, a strong figure for a business of this size.

Family of apps other revenue, a much smaller category built primarily around paid messaging and click-to-message advertising inside Messenger and WhatsApp, grew 59% y/y.

What you need to know is that revenue growth is not the issue for Meta today; by most measures, it has been exceptional.

Operating income tells a different story, growing only 11% y/y against that 26% revenue growth, a significant gap that reflects real operating deleverage.

R&D spending has risen substantially, driven largely by the cost of hiring and retaining AI researchers, who command very high compensation.

Capital expenditures have also risen sharply, and as that spending continues, it increasingly appears in the income statement as higher depreciation charges.

Both increases reflect deliberate choices where Meta could reduce this spending and report a considerably higher margin if it chose to.

We do have to make one adjustment from the operating income line.

Meta recorded a litigation settlement charge of $9.9bn this quarter, which is about half of the total $18.6bn settlement that they agreed on with attorney generals over teen usage.

Meta may not end up paying the full amount, since part of that liability depends on whether TikTok and YouTube adopt similar rules for teen and pre-teen users that Meta has already implemented across Instagram and Facebook.

Several other legal matters involving Meta have been resolved this year as well, though some cases remain pending.

Removing a significant piece of litigation risk from the stock counts as a genuine positive, even with the one-time charge attached to it.

AI Initiatives.

The Family of Apps business itself has stayed strong, by some measures stronger than ever.

AI has been one of the clearest beneficiaries to Meta's business for a couple of years now, and that benefit shows directly in the reported numbers.

This quarter, Meta disclosed that time spent on Facebook and Instagram grew in the double digits, and ad impression growth accelerated alongside it, since more time spent on the apps creates more opportunities to display ads.

Improvements in content recommendations and ad targeting, both powered by AI, have fed directly into this revenue growth.

How long this particular source of growth can continue remains an open question (more on this later).

Also in the quarter, they launched an AI agent called Muse.

Muse differs from a typical chatbot in one key respect, it takes real action on a user's behalf, beyond simply producing a text response.

Meta disclosed that Muse now has around 10mn users, growing twice as fast as Threads did at a comparable point after its own launch, with roughly 25% of those users active daily.

These numbers remain small relative to the billion-user scale of Meta's largest existing apps, but the pattern matters more than the current size: people are actually using Muse and finding real value in it.

That pattern carries weight given Meta's history with new consumer products.

A standalone app called Meta AI launched several years earlier and generated little traction.

Most of Meta's other app launches outside its core platforms have followed a similar pattern.

Threads stands as something of an exception, though its early growth owed more to users looking for an alternative to Twitter under Elon Musk's ownership than to any particular strength of the product itself, which makes it an imperfect precedent for Meta's ability to execute on AI specifically.

Genuine organic adoption of Muse, on that basis, counts as a meaningfully positive and largely unexpected signal for Meta.

Meta's plans around Muse matter well beyond its own business, since the same questions apply to Amazon, Google, and commerce broadly.

Meta has said its current plan is to monetize Muse through a transaction fee charged on purchases, along with the possibility of future subscriptions for access to higher-tier models.

Meta has specifically said it will not run Muse as an ad-driven product for now, likely because of the general distrust many people already hold toward Meta's handling of personal data, distrust that could make people more hesitant to adopt an agentic AI product if it were also built around advertising.

Meta's stance is that Muse will be free, the data stays separate from its advertising system, and monetization runs purely through a take rate on completed transactions.

So far, Meta has disclosed that roughly 25% of Muse's users have used the product to make a purchase, with an average transaction value of $200.

These figures come from a genuinely early stage of adoption, driven by early adopters who may behave differently than the broader population eventually will, so they deserve some caution.

The underlying premise, that agentic AI will eventually make purchases on behalf of users at meaningful scale, still looks reasonably likely, and a percentage-based transaction fee is a plausible way for a company like Meta to capture value from that activity.

Agentic Shopping.

Reaction to AI agents making purchases has already split companies into different camps.

Amazon has restricted access for AI agents outright where Muse currently cannot complete a purchase on Amazon.

Walmart has taken a more open approach and allows Muse to connect to its catalog, and Shopify has gone further still, allowing AI agents to complete purchases directly across its merchant network.

Amazon's caution traces to what it stands to lose.

If a third-party AI agent increasingly sits between Amazon and the consumer, Amazon loses the direct interaction that currently lets it display its own search ads, a substantial source of Amazon's revenue.

Amazon also holds real advantages worth protecting: fast shipping that can deliver an item within a day, a high degree of consumer trust, and, perhaps most importantly, default consumer behavior.

In the United States, a consumer who needs something routine already tends to go straight to Amazon, search for it, and buy it, a process that is already fast and requires little effort.

For that kind of purchase, routing the request through an AI agent adds little.

Someone buying shaving cream again is unlikely to type a request to an AI agent when opening Amazon directly and buying it takes about the same amount of time.

Agentic AI looks far more useful for higher-intent purchases that involve real research, the kind of shopping that historically began with a search on Google.

A shopper looking for running shoes, for example, benefits from an agent that can search multiple sites, compare options, and present a shortlist, something closer to assisted research than to a single fast transaction.

Travel looks like an even clearer case where agentic AI adds value.

Finding and booking a flight has traditionally meant searching through Google or Google Flights, and that process becomes considerably harder for Google to defend against an AI agent that already knows a user's calendar, can identify which travel times actually work, and can move directly to booking the flight itself.

Instinct, a company with a product similar to Muse, has reported over $1bn in transaction value so far, more than half of it coming from travel bookings specifically, evidence that agentic commerce already has at least one genuinely large, real use case.

A consistent principle helps explain which of these purchases are likely to shift toward agentic AI and which are not: the deciding factor is whether the agent genuinely benefits the consumer.

Having Muse buy something from Amazon today offers little advantage and can even feel like a worse experience, since it takes longer and still requires entering payment information into an extra step, when tapping straight into an already familiar app is just as fast.

A purchase that already involves real research and comparison across multiple sources benefits far more from an agent that can do that work directly.

This same dynamic, a monetizable query moving away from a traditional search engine and onto an AI agent instead, mirrors the risk facing Google Search as agentic AI adoption grows.

Meta's transaction fee plan remains somewhat unsettled, though it looks plausible.

The basic idea is that Muse would charge something like a 3% fee each time it completes a purchase for a user, and whether that fee lands on the consumer or the merchant remains an open question.

A merchant has real reason to accept it: advertising already costs many businesses 10% to 30% of sales, sometimes more for categories built around recurring purchases, so a 3% fee on a sale Muse generates that would not otherwise have happened looks comparatively cheap.

This same mechanism could extend to similar products built by OpenAI, Anthropic, or Google, which already has its own early agentic product called Google Spark, even though that product has not gained much traction yet.

The more important question for Meta is not really whether a transaction fee can work technically; it is whether people genuinely like using Muse, since a product people keep choosing to use can be monetized in more than one way over time.

A useful historical comparison is the introduction of credit card fees, typically 2% to 4%, which merchants initially resisted.

Merchants eventually accepted those fees once they saw how much additional sales volume credit card acceptance generated, along with higher average ticket sizes from customers able to spend more freely on credit.

A similar pattern could play out with agentic AI: merchants that resist it early may eventually experiment with it and adopt it once competitors who embraced it earlier start capturing more sales.

Amazon's likely response is to build and control its own AI agent, keeping the consumer relationship inside its own walled garden and avoiding a dependency on Muse.

Plenty of other merchants, by contrast, may be glad to accept a modest fee for a sale they otherwise would not have captured, and from a return-on-investment standpoint, a transaction fee of around 3% on incremental sales could compare favorably to many existing marketing channels.

Early signs point toward this becoming a genuine positive for Meta, more so than seemed likely even a year ago.

Enterprise AI.

A second, potentially larger opportunity for Meta involves agentic AI built specifically for businesses.

Meta may be particularly well positioned here because of a relationship it already has with an enormous number of small businesses, a dynamic that mirrors how Facebook originally built its advertising business in the first place.

When Facebook first built its advertiser base, most of those advertisers already had a Facebook business page, which gave Facebook a simple, low-friction opening: offering to boost a post for as little as $5 and reach a meaningfully larger audience.

That low-cost entry point led many small businesses to later develop more sophisticated advertising creative specific to Facebook and Instagram, and because the channel performed well, spending continued to grow from there.

Facebook did not need a large outbound sales effort to reach these businesses, since the businesses were already present on the platform, and simple, self-serve advertiser tools made the next step easy.

That combination helped build the ad density that makes Meta's auction system work well today.

A similar dynamic may now apply to AI tools built for small businesses.

Meta has not disclosed an exact number of advertisers, though the figure sits in the high tens of millions, and more than 500 million businesses maintain some presence on Meta's platforms overall.

Meta has disclosed one case study involving a Vermont maple syrup producer that used agentic AI to improve its advertising creative and ad-buying process, resulting in a 57% increase in sales.

A small business already using Meta's advertising tools, and already aware of the platform's AI capabilities, is a natural candidate to test a tool like this.

A positive early result builds confidence and makes that same business considerably more open to adopting further AI tools later, covering areas such as inventory management, customer service, or even payroll.

Most of these broader small-business AI tools have not rolled out widely yet, though Meta has said it plans to expand them soon.

If that expansion succeeds, it could eventually become a business comparable in scale to Meta's current advertising revenue, priced either through subscriptions or usage.

None of this needs to happen for Meta's current valuation to make sense; it functions more as a call option.

A broader principle sits behind all of this: the company that ultimately wins in AI will not necessarily be the one with the single best underlying model.

It will be whichever company turns AI into the most useful product and distributes that product the most effectively.

Meta does not currently operate a leading frontier AI model, yet it has already shown early evidence that it can build a large, genuinely useful AI product and get it in front of a very large number of people quickly.

Capex Spend.

Meta is now guiding to $115-135bn in capex for the year, with some reporting suggesting 2027 spending could run even higher.

On the earnings call, Meta also announced a new, non-voting share class, called Class C, that existing shareholders will receive in addition to their current shares.

The practical effect resembles a stock split, but the larger purpose is strategic: this new share class gives Meta a way to raise equity capital in the future without diluting Mark Zuckerberg's voting control.

Meta is already acting on that ability.

The company announced an initial stock offering of $22.5bn, convertible into an additional $12.5bn, for $35bn in total, alongside discussion of up to $60bn more in future equity raises, implying as much as $95bn in additional capital over time.

Most of that capital will likely flow toward further capital expenditures.

As a result, free cash flow has turned negative over the past couple of quarters.

Meta is now covering that financing gap partly with its remaining cash on hand and partly through additional equity and debt, a real shift in its capital structure, though not yet an extreme one.

In effect, this capital raising pulls forward roughly 1 to 2 years' worth of earnings.

Dilution could become a more serious concern later, but the outcome ultimately depends on the return Meta generates on this invested capital.

Meta's AI spending is already improving results inside its core Family of Apps business specifically, independent of how Muse, business AI tools, or any Neo Cloud-style opportunity eventually perform.

Meta's compute capacity is also somewhat fungible: if Muse does not continue scaling as hoped, Meta can redirect some of that capacity toward further improving ad targeting and content recommendations inside its existing apps, though that flexibility has real limits.

As a further option, Meta has started a new business called Meta Compute, designed to rent out excess capacity much like a neo cloud does.

This detail carries its own risk.

Building out a large amount of capital expenditure with a plan to simply rent out whatever demand does not materialize raises the possibility that several major technology companies attempt the same strategy at once, since if this approach does not work well for Meta, it likely will not work well for other hyperscalers either.

A resulting scenario with too much compute supply on the market could pressure prices broadly across the industry.

That risk is softened somewhat by the strength of Meta's core Family of Apps business, which appears capable of supporting much of the company's current valuation on its own, without assuming significant additional value from these more speculative initiatives.

Reality Labs.

Reality Labs lost $18bn over the past year.

Relative to the roughly $130bn Meta is spending on capex, a loss of that size now draws comparatively little attention from the market.

Meta has talked about reducing these losses over time and eventually bringing the division closer to profit neutral.

Reality Labs also functions as a real AI beneficiary in its own right.

The division's hardware, particularly its smart glasses, works well together with AI and represents a genuinely strong product.

Looking forward, a device resembling today's AR and VR-style glasses looks like a reasonable guess for what an AI-first hardware category eventually looks like.

Meta does not know what Apple may be developing in this space, but Meta currently holds a real lead here, and even a strong competing device from Apple would likely still leave Meta with a meaningful position in this category.

Valuation.

Meta guided to revenue growth of 24-29% for the next quarter.

Taking the midpoint of that range and rounding produces approximately $255bn in revenue for the full year.

Applying a margin assumption to that figure is somewhat complicated, since Meta's margin was 41% as recently as 2025, and margin pressure has become visible over the past two quarters as rising research and development spending and growing depreciation from capex have flowed through the income statement, even after removing the litigation settlement cost.

A 38.5% margin, applied to the $255bn revenue estimate and taxed at 20%, produces approximately $30.50 in earnings per share.

At a stock price of $630, that implies a trailing multiple of approximately 21x earnings.

This does include about $5.50 per share of Reality Labs losses, and a loss-generating business arguably should not carry a positive multiple at all, since doing so implies either permanent negative value or, more realistically here, several more years of losses before the division possibly becomes profitable.

Removing the $18bn billion Reality Labs loss produces an adjusted EPS of approximately $36, a 17.5x multiple.

Removing a portion of Meta's elevated growth spending as well, using the higher 2025 EBIT margin instead of today's pressured figure, reduces the implied multiple further, to 16.5x.

However an investor approaches the calculation, Meta currently trades below the broader market's average multiple while still growing considerably faster than that average, with revenue up 26% and EBIT up 11%.

Over time, operating income growth should either reaccelerate toward revenue growth, or the elevated spending driving the current gap will get reduced because it is not generating the results Meta expects.

Asked what the single biggest risk to Meta is today, the answer is neither the current level of AI spending nor Reality Labs.

The larger risk is that, at some point in the future, the improvements AI currently delivers to ad targeting and content recommendations simply stop.

There may be little or no visible warning before that happens: ad targeting could stop improving at its current high rate, or the recommendation system could stop driving growth in time spent on the apps, without much advance signal either way.

For now, continued improvement looks like the more likely path, since AI development across the industry remains relatively early with considerable room left to improve.

If that improvement were to plateau at some point without Meta's broader AI business already generating substantial revenue of its own by then, the effect on Meta's core growth would be serious.

That scenario still looks like it sits several years out, at minimum, based on everything visible today.

For more on Meta, check out this video below.

YouTube FAQ

Nothing in this newsletter is investment advice nor should be construed as such. Contributors to the newsletter may own securities discussed. Furthermore, accounts contributors advise on may also have positions in companies discussed. Please see our full disclaimers here.