Meta (META.US) Q3 Conference Call: Muse has achieved product-market fit, computational efficiency will increase tenfold next year.

According to Zhitong Finance APP, Meta (META.US) reported total revenue of $64.6 billion for Q3, up 26% year-on-year, with Q4 revenue guidance between $73 billion and $76 billion. The earnings call focused on the progress of the Muse personal agent: over 10 million US and Canadian users set up Muse within 1.5 months of launch, with a daily usage rate of about 25%. Shopping is the largest use case, with an average purchase amount exceeding $200. The management expects a more than tenfold improvement in compute efficiency next year, and is planning multiple monetization paths including subscriptions, transaction fees, and enterprise APIs.

Meta’s Q3 total revenue was $64.6 billion, up 26% year-on-year, including $63.1 billion from advertising (up 26% YoY) and $1.1 billion from other app family revenues (up 59%). GAAP operating profit for the quarter was $13 billion, down 37% YoY with an operating margin of 20%, mainly impacted by $9.9 billion in multi-state attorney general legal settlement costs. Excluding those costs, operating profit grew 11% YoY.

Net profit was $9.9 billion with earnings per share of $3.83. Capital expenditures totaled $34.9 billion, free cash flow was negative $2.1 billion, end-of-quarter cash and marketable securities amounted to $82.4 billion, and debt was $83.7 billion. Staff headcount exceeded 66,000, a 12% decrease from Q2. The company also announced the creation of a new non-voting share class, to be distributed to existing shareholders as a stock dividend, effectively a 2-for-1 stock split, expected to begin trading in late January 2027.

The core focus of the earnings call was on the progress and monetization paths of the Muse personal agent. In 1.5 months since launch, more than 10 million users in the US and Canada have set up Muse, with a daily usage rate of around 25%—over twice the initial rate of Threads. Shopping is the most common use case, with over a quarter of users participating and average purchase amounts exceeding $200. Management clarified that Muse has not yet launched globally, mainly restricted by compute intensity, but expects to improve compute cost efficiency by over tenfold next year, at which point profitability becomes feasible once revenue scales.

The business model is based on a free tier by default, with multiple monetization paths including transaction fees, premium subscriptions, enterprise APIs, and Muse Code. The Meta One unified subscription service is expanding globally, with over 15 million subscriptions and trials. On the enterprise side, the company is targeting 500 million self-identified small businesses, launching Muse for Small Business and previewing the Personal Agent Protocol, collaborating with Walmart, Stripe, Shopify, and others to set agency commerce standards.

Looking ahead, Meta guides Q4 revenue between $73 and $76 billion, full-year capital expenditures at $135–145 billion, and full-year total expenses at $175–179 billion (including $9.9 billion legal costs). Reality Labs' loss is expected to narrow to about $15 billion next year. Management did not provide specific 2027 revenue guidance, but stated AI investments are already delivering incremental engagement, improved recommendations, and better monetization for the core business, and expects robust revenue growth for the full year of 2027.

On advertising AI progress, the generative retrieval architecture now covers about a quarter of non-connected Feed views and half of Reels views. GEM model expansion has lifted offsite conversions by 1–1.7% and app ad conversions by 9.6%. Nearly 10 million advertisers use generative AI creative tools, while partnership ads are running at over $15 billion in annualized revenue.

Below is the Meta Platforms Q3 2026 earnings call transcript:

Executive Remarks

Mark Zuckerberg

Founder, Chairman, and CEO

Okay. Hi everyone, thanks for joining today. The last few months have been busy and exciting. Our Family of Apps now reaches 3.64 billion daily active users, with record engagement on Facebook and Instagram thanks to strong video growth. Threads’ monthly active users have surpassed X and continue to grow quickly. WhatsApp message volume hit all-time highs and is becoming an increasingly important platform for agents.

In recent months, we’ve achieved several major milestones towards our goal of providing “personal superintelligence for everyone.” Our Muse personal agent is one of the most successful product launches ever. We released Muse Wonder as our first cutting-edge model and rolled out several Muse-integrated new glasses and accessories, including audio glasses launching soon.

We’ve discussed our personal superintelligence vision for over a year. We want to help everyone have a highly capable personal agent that understands your goals, helps you achieve them, and improves your life in any way you choose—whether that’s improving relationships, saving time, saving money, starting a business, being creative, or anything else. Muse is the product realization of this vision. So now we have the foundational technology to achieve this. There’s much work ahead, but our initial position is very strong.

In just 1.5 months, over 10 million people in the US and Canada have set up a Muse personal agent. Growth remains strong, with us topping the App Store charts almost every day. Engagement and retention have started very strong: about 25% of those who set up Muse use it daily. For our earliest adopters, the longer they use Muse, retention appears even better. For comparison, this is more than double Threads’ initial rate and above what we’ve seen for other AI apps at launch.

The most common use case for Muse is shopping; over a quarter of Muse users participate in shopping, with average purchase amounts of over $200. Plaid told us they’ve never seen this level of connector activity on any other service. In the first week, we had to increase credit limits with Stripe, as transaction volume exceeded all our expectations.

Aside from shopping, the main connectors are personal productivity and connecting Muse to your other services like email and calendar. This highlights the importance of the Muse Secure VM we built for state-of-the-art privacy and security, and we’re looking forward to launching Muse Confidential VM soon, at which point even Meta won’t be able to see your content.

Muse Wonder has further improved quality, engagement, and retention, showing that as we continue to make the model smarter, Muse gets better. Now that Muse has clear product-market fit, we’re entering a new, more operational phase focused on scaling. The next phase will play to many of Meta’s strengths. We need to do three things: grow the product, grow revenue, and optimize compute costs.

We expect to scale Muse to serve billions of people. This is both our mission and our strategy. In mission terms, we believe everyone should have the power of personal superintelligence. We see this as the path to a positive future for all, with personal empowerment as the source of prosperity and power balance as the basis of safety.

On values, we believe broad access to this capability is the best thing we can do. This is also our strategy, because we think a network of billions of Muse agents at scale will be able to accomplish more for both individuals and businesses through collaboration. Therefore, we want to create a large-scale Muse network as quickly as possible.

On the business side, our current average revenue per MAU in our app family is several hundred dollars per year, and we expect that Muse will reach even higher levels over time, since we’re already seeing a lot of business activity taking place. This will take time, but we’re starting work on it now. Our business model will have several components. For most, the product should be free; we’ll build business tools and take a fee based on transactions we help facilitate. We expect businesses to pay this.

For those who want the most advanced models or more tokens, they can subscribe and pay directly. For business owners who want extra features to use Muse for their operations, or companies wanting employees to use Muse, they can also subscribe and pay. For users who want Muse to improve Meta platform content and ad personalization and quality, they can connect these experiences too. Just to clarify, when I previously mentioned average revenue, I meant current app family average revenue per Muse user.

On compute, we’re building the capacity to serve over a billion people over the next few years. In the near term, running Muse is compute-intensive, but we expect to improve cost efficiency by more than tenfold next year, making profitability feasible once revenue scales. If efficiency continues to improve rapidly—which we think is quite likely—Muse will be highly profitable for years. For now, our core business strength allows us to scale the Muse network before it’s profitable, and we believe we are the best-positioned company to execute this strategy.

Another aspect of compute economics is that since training is a massive fixed cost, now that we have a leading model, we want to serve as many high-margin inference requests as possible. That’s why we’re building the Meta Enterprise Platform led by CJ Desai. Muse Wonder and more advanced models in training can power high-margin services outside of the Muse network, including our API, Muse Code, and other enterprise services.

A key feature of our business is that we have many good uses for compute. It’s hard to predict precisely how the Muse network, our enterprise platform, and Muse Wonder (plus subsequent models) will accelerate the growth rate of our core business. But by establishing multiple ways to deploy these advanced models, we increase the odds that one or more of them significantly exceed our expectations.

Aside from serving other businesses, we have not yet found the limit for applying compute to accelerate our own core business. We’ve already run a Muse Wonder test to improve its ability to autonomously enhance our recommendation and ad systems. As our upcoming models get even more advanced, I expect to see these systems improve even faster.

Now, about the models. Muse Wonder is state of the art on many dimensions important for personal superintelligence. It validates the importance of training our own models to deliver the best experience. Beyond Muse, since launching new models in April, Meta AI engagement per user has more than doubled.

While Muse Wonder is our most important release, I’m very pleased with our execution speed. In the past few months, we released Muse Spark 1.2 and 1.3, Muse Code, two open-weight models Muse Glimmer and Muse Spark Open, and Muse Voice. Looking ahead, as the space moves quickly, we’ll continue to push the frontier of intelligence while also distilling these models into smaller, more efficient production models optimized for personal agents. My view is that we can achieve superintelligence through scale, so we’re systematically climbing that ladder. We’re already developing larger models following Muse Wonder, which will be very impressive and capable. We're also developing new smaller models that will deliver Muse Wonder–level intelligence at much higher efficiency.

Next. Our work on personal superintelligence is now Reality Labs’ focus as well. Muse will launch on Meta glasses starting next month, which is an important step towards that goal. People already love many features of the glasses—music, calls, capturing moments, hearing assistance. Over time, though, we expect Muse and having your personal superintelligence with you all day to become the most important part of the experience.

Muse is also at the core of the glasses business model. Again, our mission and strategy are to serve everyone. We believe in putting this technology in everyone’s hands. That’s why we price the hardware so we don’t make large profits on devices, but rather build a large software and superintelligence business on top of them over time. Glasses sales continue to grow quickly, and now with Muse, we expect paid subscriptions to increase, and this business will begin to improve significantly.

At Connect, I also shared that we’re changing our approach to VR, only releasing new devices when we achieve a major breakthrough—just like with the Meta VR glasses we released this year, the first ever VR in glasses form factor. I still believe the path to providing holographic experiences and integrating the physical and digital worlds into everyday glasses will initially happen through VR, but we’re going to pace investments more thoughtfully to prioritize delivering personal superintelligence. With all these factors combined, we expect the Reality Labs P&L to improve from here, with operating losses narrowing to around $15 billion next year and trending toward profitability.

Before I wrap up, I want to touch on the remarkable speed we’re seeing. A key part of our approach has always been to move fast to learn quickly and better serve users at scale. But lately, things have accelerated. Over the past year, we launched Edits, which has exceeded 50 million active users. New apps like Forum and Seller, built by small teams, are scaling quickly to serve power users. Muse went from an idea on a whiteboard to a successful product in only about eight months. We’re about to ship Muse Charm, taking under a year from concept to shipping hardware in people’s pockets.

I’m proud of what we’re building. People have all kinds of views about the future, but at Meta, we believe the future belongs to everyone, and putting power in people’s hands is our way of delivering on that. I think it’s important. There needs to be a company that stands for those ideas. But it’s not just about ideas—the execution matters, too. So I’m grateful to everyone who’s working so hard to bring all of this to billions of people globally. As always, thank you to everyone partnering with us on this journey.

Now let’s turn it over to Susan.

Susan Li

Chief Financial Officer

Thank you, Mark, and good afternoon everyone. Let’s begin with segment performance. Unless stated otherwise, all comparisons are year-over-year. Q3 Family of Apps total revenue was $64.2 billion, up 26%. Q3 Family of Apps ad revenue was $63.1 billion, up 26% on both a reported and constant currency basis. In Q3, total ad impressions grew 10% across our services. Impression growth was healthy in all regions, driven by engagement, user growth, and optimized ad load. Global average ad price grew 14% YoY, helped by improved ads efficacy and a better macro environment versus Q3 last year. This was partly offset by strong impression growth, especially from lower-monetizing surfaces and regions.

Other Family of Apps revenue was $1.1 billion, up 59%, primarily driven by paid messaging and subscription revenue from WhatsApp. In Reality Labs, Q3 revenue was $443 million, down 6% due to lower Quest headset sales, partially offset by continued strong growth in AI glasses revenue.

Now, consolidated results. Q3 total revenue was $64.6 billion, up 26% on both a reported and constant currency basis. Q3 total expenses were $51.6 billion, up 68% YoY, including a $9.9 billion charge for the August 2026 multi-state attorney general settlement. The expense growth was driven mainly by legal-related costs, infrastructure, and third-party AI token costs. Infrastructure costs were up due to higher third-party cloud spend, depreciation, and data center operating expenses. As of Q3, our headcount exceeded 66,000, down 12% from Q2, following layoffs in May 2026.

Q3 GAAP operating income was $13 billion, down 37% YoY, with a 20% operating margin. Excluding the above legal settlement charge, operating income would have grown 11% YoY. The quarter's tax rate was 24%, above our guided range due to the non-deductible portion of the legal expense. Net income was $9.9 billion, or $3.83 per share.

Capital expenditures (including principal payments on finance leases) were $34.9 billion, driven by investments in servers, data centers, and networking infrastructure. Free cash flow was negative $2.1 billion. Quarter-end cash and marketable securities were $82.4 billion, with debt at $83.7 billion.

Now, business performance. Two main factors are driving our revenues: our ability to deliver engaging experiences for our community, and our effectiveness at monetizing that engagement. On the first, we continue to see meaningful gains from our content recommendation initiatives. On Instagram, Q3 global per-user time spent hit an all-time high, up double digits YoY, with both Reels and Feed strong. On Facebook, Reels watch time set all-time highs both globally and in the US, up double digits YoY. Threads also continues to expand, with per-user time spent up more than 20%. Engagement gains across our services are benefiting from our improved relevance, personalization, and timeliness of recommended content.

Investments in our recommendation engines’ retrieval and ranking are continuing to show positive results on Instagram and Facebook. For example, on Instagram Reels, the relevance updates we made in Q3 drove over a 4% increase in Reels watch time, partly driven by our upgraded model architecture (which now ranks a batch of Reels rather than each Reel independently), as well as improvements in incorporating richer watch time signals.

On Facebook Feed and Reels, we completed a global rollout of a new generative retrieval architecture inspired by LLMs, which uses semantic representations to match content fragments and generate potential recommendations, helping our systems surface relevant content before it accumulates significant engagement. This new architecture now covers about one-quarter of non-connected Feed views and half of Reels views, driving a 1% increase in Reels watch time.

Our recommendation work is also helping us deliver fresher, more personalized content. Currently, nearly 45% of recommended Instagram Reels have been published within a day, up more than 11 points from Q2. On Facebook Reels, we doubled the amount of history the ranking model can consider, increasing US Reels watch time by 1%.

LLMs are also helping us understand content more deeply. We continue to process public Instagram posts daily using LLM-based content understanding systems. By the end of this quarter, these signals improved engagement with Instagram search results by 1%.

This quarter, we also began migrating our content understanding models to our first-party Muse model series. In offline tests, these models outperform our previous prod models and reduce per-unit content processing costs for Instagram, Threads, Facebook, and ads by over 80%. Beyond improved quality, having more of this stack in-house gives us greater flexibility to specialize models for the signals most important to personalization.

Turning to the second driver of revenues—monetization efficiency. We see significant opportunity to further improve outcomes for businesses using our services. In our ads system, we’re continuing to drive performance through deeper predictive models that better understand users and deliver more relevant ad experiences while making our infrastructure more efficient. Last quarter, we rolled out Meta Generative Recommender, which uses LLMs to reason over ad content and user preferences to predict the best ad for each person.

In Q3, we launched this generative retrieval approach on Facebook Feed, driving nearly a 2% conversion lift for the target segment. We also expanded the GEM model for ad ranking to more surfaces, lifting offsite conversion rates by 1–1.7% and app ad conversion rates by 9.6%.

We continue to rapidly scale retrieval systems. In Q3, Andromeda improvements—including model complexity and throughput—lifted Facebook conversion rates by 1.4%. We’re also using AI to make it easier for businesses to manage campaigns, develop creative, and capture performance opportunities.

Meta AI business assistant is becoming a more proactive, capable partner for advertisers. We estimate that advertisers adopting its recommendations have seen nearly a 5% increase in conversions at the same cost per conversion and a 9% increase in ad spend—compared with little change for those who did not adopt. We’re rolling out more capabilities, including creative insights, richer business context, and more agentic operations. These features will enable marketers to plan, create, and optimize campaigns through a conversational interface, organizing tasks and reminders, while maintaining control through approvals and previews.

Let me share an example of Meta AI business assistant empowering advertisers. Simply Organic Maple, a family-run maple syrup producer in Vermont, used the assistant to improve their campaign messaging. After following the recommendation to test longer, more descriptive ad copy, the business saw a 57% increase in purchases and a 36% lower cost per purchase. Their best-performing revised ad achieved a 3x return on ad spend, up 52% from the original.

Adoption of our generative AI ad creative tools is also expanding. Nearly 10 million advertisers used at least one of these tools in Q3. Adoption of our video generation feature grew 25% quarter-over-quarter, as pre-produced videos reduced generation time from uploading a single image to just seconds. We’re also increasing opportunities at the creator-ads intersection. Partnership ads, which allow advertisers to run ads with creators, brands, and other businesses, continued rapid Q3 expansion. Our partnership ad product grew more than 97% YoY, now running at an annualized revenue rate over $15 billion.

In Q3, we also completed the global rollout of Creator Marketing Hub to all advertisers, centralizing creator discovery, content organization, and partnership ad activations, with over 10 million discoverable creators. We’re integrating Meta AI business assistant into the hub, enabling advertisers to conversationally search for creators, identify content for partnership ads, and get guided help with creator marketing workflows.

Outside of ads, we’re also building more ways to monetize engagement. In September, we announced the global expansion of Meta One, our unified subscription service for consumers, creators, and businesses. Meta One is rolling out gradually, covering 50+ features across Instagram, Facebook, and WhatsApp, including Meta Business Agent and Meta AI, and already has 15+ million subscriptions and trials. The core experiences in our apps and Meta AI will remain free, while subscriptions provide more professional features and extended AI usage.

In addition to monetizing engagement from existing services, we’re building new experiences and revenue streams around consumer agents. Mark highlighted Muse’s impressive early momentum. Users are finding many reasons to use Muse—the top three categories are shopping, personal productivity, and creativity. We’re encouraged by the breadth of use cases, especially shopping—over 25% of Muse weekly actives have used it.

While our near-term priority is scaling adoption and extending Muse’s reach, we see multiple monetization paths, starting with subscriptions. We also see a major opportunity to extend Muse to small businesses. According to a recent user survey, we estimate at least 500 million self-identified small businesses use Meta’s apps to reach customers and generate revenue. Muse can help these businesses acquire customers and operate more efficiently, while helping people complete transactions. Over time, consumer and business agents can connect and transact directly, creating opportunities around paid usage, commerce, and enterprise services.

In addition to growth, we’re focused on building sustainable unit economics. Muse’s compute intensity means scaling its user base will require major investment, and total service costs are expected to rise. At the same time, we expect to improve compute cost efficiency by more than tenfold next year, aided by better utilization of off-peak capacity, more efficient models, and inference optimization. With scaling revenue, these efficiency gains give us confidence in building a highly profitable business over time.

Now, our financial outlook. We expect Q4 2026 total revenue to be in the range of $73–76 billion. Our guidance assumes FX headwinds of approximately 1% YoY based on current exchange rates.

Turning to expense and capex outlook. We’re raising our expense outlook to include a $9.9 billion legal charge recognized in Q3. We now expect full-year 2026 total expenses to be $175–179 billion. Excluding the Q3 legal charge, we still expect 2026 operating income to be higher than 2025. For segments, we now expect Reality Labs losses to decline from 2025 to the $18–19 billion range.

We now expect 2026 capex (including principal payments on finance leases) to be $135–145 billion, a narrower range from the prior $130–145 billion. Absent tax law changes, we expect Q4 2026 tax rate to be in the 15–17% range.

Additionally, we continue to monitor active legal and regulatory matters that could materially impact our business and results. For example, we continue to see scrutiny of teen-related issues in multiple markets, with several teen-related trials scheduled in the US in the months ahead.

Looking beyond 2026, we’re still finalizing next year’s budget and capacity plan, but we expect to continue actively investing to support our core business and emerging AI initiatives. Our current capacity and ramping capacity give us flexibility to allocate resources between model training, core business growth, and new product development and support.

Our AI investments have already delivered substantial incremental ad revenue through stronger engagement, better recommendations, and increased monetization efficiency. Building on this, our MSL investments are creating the technology and infrastructure needed to unlock major new revenue opportunities and enhance our overall business capabilities. Although returns on these investments will take longer than our core business, the breadth of monetization opportunities and internal efficiency give us confidence in their long-term potential. Looking ahead, while we aren’t giving full-year 2027 revenue guidance, the benefits our AI investments bring to our core business, combined with opportunities from model advancements and new product launches, enable us to sustain strong overall revenue growth throughout 2027.

We also remain focused on improving Reality Labs’ profitability. As Mark mentioned, next year losses are expected to narrow to around $15 billion, as we shift investment to other priorities, mainly AI.

Finally, today we announced plans to create a new class of non-voting capital stock, to be distributed as a stock dividend to all existing shareholders. For each existing share, shareholders will receive one new non-voting share, resulting in an effective 2-for-1 split. We expect trading to begin in late January 2027.

In summary, our business momentum continued in Q3, with strong execution on core ad and engagement initiatives, and we’re excited by our progress toward personal superintelligence.

Now, let’s open it up for questions.

Q&A

Brian Nowak

Morgan Stanley Research

I have two questions, one about Muse and one about the enterprise business. First, congratulations on Muse’s success. When you analyze user behaviors and the product, could you discuss any top constraints on forward user growth trajectory—whether it’s capability, model efficiency, compute capacity, etc.—so we can understand the main factors limiting future user growth that you hope to address?

The second is about the enterprise business. Mark, could you briefly outline your vision for Meta’s enterprise product offerings and the key value drivers you hope to deliver for businesses adopting the service?

Mark Zuckerberg

Founder, Chairman, and CEO

Yeah, I think I can speak to both and then Susan can add. So for Muse now, the main thing is we have over 10 million people who have set it up in the US and Canada, but I think the main constraint so far is that because this is such a compute-intensive service, it hasn’t rolled out globally. Part of that is wanting to optimize for each locale and some work around connectors, but part is just making sure we can do this efficiently.

There are pros and cons. The benefit is, making it to 10 million users in the first six weeks, all US and Canada users in a few concentrated markets, is much more impressive. If it were rolled out more broadly, you’d expect the number to be much higher than we see today. But on the flip side, we have work to do to improve efficiency to a reasonable level.

So that’s an important part of what we need to do. Compute efficiency work also matters because we want to have enough built-out compute to serve what we see as 1 billion+ users if we hit certain efficiency goals, but there’s still a lot to do. This will be a limiting factor next year.

We want to build a business around it. I spoke about that. I think improving compute efficiency and building the business in parallel is key to making this viable commercially. I’m optimistic, but it’s something we have to do.

Then it’s about continuing to add product value, mainly through ongoing model development. That’s why we initially launched with Muse Spark, which is today’s default model for the free tier. We released Muse Wonder, which people can use in the paid tier. We know it performs much better, so we’re developing a more efficient version of Muse Wonder to roll out more broadly. But you can imagine iterating 2–3 more models forward—when you have much higher intelligence and can solve more problems at a lower price point, I think that unlocks more value for people. We’re excited about that.

Shifting to the enterprise part, because I think the two are somewhat related, the near-term main enterprise products will be the model API and Muse Code. We already have enterprise agents scaling for a while, and all three are growing fast. So we’re excited about that. Over time, we want to build out a full Muse agent business network. There are business agents, as we’ve discussed previously, that help with sales and support for companies.

But we’ve seen a lot of people using Muse specifically to help run small businesses. We think there’s an opportunity to create an agent people use to automate the hard and time-consuming, non-core business stuff when starting a small business. That’s a big deal.

And having a version of Muse people can use across business environments—also big. Beyond that, building what I call a Muse network—not just a single-player game “you have your Muse,” but your Muse can collaborate with your friends and family’s Muses. If you want to buy something, instead of just navigating to a business website, you interact with their agent. Making that collaborative is a massive value unlock at scale.

I’m excited for all of that. Business use cases will be important, but there will also be a ton of consumer utility. So I’m excited for both.

Susan Li

Chief Financial Officer

I want to add to Mark’s enterprise comments. AI is creating massive disruption in the enterprise today. From what we see, we really don’t think traditional enterprise companies have locked this up. In fact, some of today’s leaders in the field didn’t exist a few years ago. Those companies invested heavily in model development, and we have too. Our increasingly competitive models benefit all users—including small businesses and the enterprise. Today, 500 million businesses are already using our products.

So we really believe AI will reshape how businesses operate. We expect companies of all sizes to use AI agents in software development, productivity, sales, marketing, customer service, and more. We have differentiated assets for this market, combining models, infrastructure, global distribution, and long-term enterprise relationships. We see this as a new, diversified recurring revenue stream supporting multiple business models, including subscription and usage-based API/messaging, plus other revenue lines. We expect it will take years to play out, and we’ll invest in response to demonstrated customer demand and product-market fit.

Eric Sheridan

Goldman Sachs Research

First, a follow-up on Brian’s question: when you think about friction for agentic commerce, how do you as a company see addressing cases where aggregators or suppliers are reluctant to let agents interact with their platforms or services, and how do you see reducing that friction or balancing the evolution between current services and agentic behavior over the medium/long term?

Second, with the new share class announced tonight, any updated thinking on sources of capital and optimizing capital structure to ensure suitable investment to capture AI opportunities over the coming years?

Mark Zuckerberg

Founder, Chairman, and CEO

I can answer the first one. Right now, we see that the vast majority of businesses are really excited to support this because they realize it’s a new opportunity to increase sales and improve experience for existing customers, as well as potentially reach new customers. Different actors in the broader commerce ecosystem will adopt at different paces, and our strategy will shift depending on the progress here. Overall, we’re very pleasantly surprised by the speed at which many businesses not only want to make Muse more effective on their services but are proactively reaching out to us for integrations. It's a great sign.

Of course, not everyone wants to be an early adopter, and that’s fine. We’ll work with people on terms that make sense for them. But I do think that over time, making this all work well will be extremely valuable for everyone, and we’re on the right track right now.

Susan Li

Chief Financial Officer

I’ll add to that, and also address your second question. I think agents will be an important part of this evolution, and everyone will have to adapt to what agentic commerce means in the new era. We have a lot of companies excited to partner with us, but also some that understandably focus on protecting customer relationships. We expect the market will eventually move away from agents “navigating” experiences like humans and more towards permissioned, structured interactions—certified APIs, approved agent partnerships, or direct consumer-personal agent to enterprise agent communications.

We’re helping build the standards for that model. Earlier this month, we previewed the Personal Agent Protocol—an open standard across Muse and Meta business agents—developed with partners including Sierra, Walmart, Stripe, and Shopify. This protocol defines how consumer personal agents can securely and reliably interact with enterprise agents, retaining proper controls for both businesses and customers.

So again, adoption may be uneven and will take time. In the near term, many experiences will stay AI-assisted rather than fully autonomous. As trusted integrations grow, and as agents prove they can generate value safely, we expect more and more businesses will want to participate.

Your second question on the new share class: The purpose is to create a publicly traded, non-voting class of shares—usable for capital raising, employee compensation, acquisitions, and other corporate uses—while maintaining voting control for our existing class A and B shareholders. This maintains our founder-led structure, broadens financing tools, and gives us flexibility for investing across the business, especially given the AI-driven momentum we're seeing. We don’t have concrete capital plans to announce now, but believe it’s prudent to establish the right capital structure for diversified funding long-term.

Douglas Anmuth

J.P. Morgan Research

I have two questions. First, a follow-up on your comments, Susan, about Muse's evolution and broader agent adoption by businesses: how do you see things changing and shifting as we move to more permissioned, agent-to-agent commerce? What does that ultimately unlock over time?

Second, on expenses and capex thinking for 2027. I know you’re not quite there yet, but referencing past comments, you talked about maximizing compute in 2026–2027, then more flexibility by 2028. A year ago you said 2026 dollar growth would outpace 2025, and you also referenced annual operating income growth. Is there anything that can be tied together to help us think about 2027?

Susan Li

Chief Financial Officer

Thanks, Doug. On your first question about how the agentic ecosystem will evolve: I wish I could perfectly predict it—it’ll take some time and will be uneven, as I said. But it's clear to me when I use Muse today that a lot of experiences would be better if Muse could more seamlessly interact with an enterprise's agent on the other side. For example, if Muse calls a business for me, it would be easier for the business if the call were handled by their agent.

What’s unique about Muse, and what makes the product special, is the level of proactivity it demonstrates on my behalf. I can give it a task and it will pursue it, even if it hits obstacles or can’t find exactly what I asked for, it will find creative ways to help. So I think that sort of proactivity—on both the consumer and business agent sides—will result in more transactions and activities happening, compared to today where action only happens once there is explicit demand. Muse could keep searching for me, send me updates, and do things like send 10 book recommendations each month. Agents should do similar things for businesses. It’s an opportunity to create a more proactive agent-to-agent interaction framework over time.

On your second question about 2027: we’re in the early stages of next year’s budgeting and capacity planning. We’re evaluating many proposals with attractive, but varying, timeframes for returns—trying to figure out how to fund as many of these as possible within our budget and financial goals. It’s a dynamic process and still early. So before our Q4 call—traditionally when we share next year’s guidance—we won't have additional specifics to share.

Justin Post

BofA Securities Research

Okay, two questions, one for Mark and one for Susan. Mark, you have a strong new model, and aside from external opportunities, a lot you can do internally, especially for advertising. As you deploy the new model, how much room is there for further improvement? How should we think about it?

Then for Susan, you mentioned strong revenue growth next year—is that advertising, or will it include incremental AI revenue in 2027?

Susan Li

Chief Financial Officer

I’m happy to take both. So, on your first: our ads business has benefited for years from ongoing investments in traditional ML models and, more recently, next-gen models. I'd call out three themes: First, there is a strong pipeline of performance-driven, traditional ML improvements. This is still a healthy and strong pipeline—business as usual. As we look to 2027, again not providing guidance, but our engagement and ads investments give us confidence in meaningful growth opportunities throughout 2027. That extends to your second question. Sorting and recommendation pipelines are healthy.

Second, we’re betting on new, LLM-oriented AI infrastructure that can reason more deeply about people’s preferences and interests for organic content and recommendations—expected to unlock further improvements for ranking and recommendations.

Third, we expect generative AI tools—which are already highly valuable for advertisers—to continue growing in importance as part of our Advantage+ suite, making it easier to generate creative, turn a single image into a video or sound, and make ads more personalized for users. So all three have material upside for growth, we feel.

Your second question on next year’s revenue growth: as I mentioned, we feel good about the planned budget investments in core ads and engagement for next year. We are also building several new AI businesses, which we expect could begin to contribute incremental revenue in 2027—but more importantly, scale meaningfully over time. These include monetizing our AI products via subscriptions, transactions, ads, and enterprise offerings as previously discussed. While many of these are still early, our near-term focus is building high-quality products and achieving (and scaling) product-market fit. But our track record scaling businesses to reach billions gives us confidence that AI will become a meaningful source of revenue diversification over time.

Mark Shmulik

Bernstein Research

Mark, coming out of Connect, it feels like a new era for the company, with Muse as the new business center. Is that a fair assessment of where things are internally? If so, can you elaborate on how empowerment and business outcomes might change?

Susan, the release cadence is interesting—enterprise announcements preceded some small biz ones. Thanks for your comments on the SMB agent opportunity. Can you share more detail on serving and scaling the SMB product?

Mark Zuckerberg

Founder, Chairman, and CEO

Yes. I can briefly address the first. Over a year ago, I wrote about personal superintelligence as our “north star.” That fits basically with the work we’ve done for 20+ years—focusing on empowering people, giving them voice, the ability to connect how they want. Extending that into the AI era, we believe the most important thing we can do for the next generation of technology is to put that power in everyone’s hands—that’s the vision of personal superintelligence.

We’ve been working on this for some time. That’s when we rebooted around MSL, building the lab. The idea is we need models highly specialized for personal agents and by owning, building, and designing the end-to-end technology, we can deliver the best experiences. That includes models and agents. That’s been in the works a while.

The vision for glasses is similar—we want to build the next hardware platform to deliver the best end-to-end experience. That’s starting to come together and will be more evident in the coming years. These are long-term bets. That’s what we do here: long-term tech bets, designing the best end-to-end experiences for billions. As long as we keep doing that, the business will take care of itself, provided our execution remains strong. I’m proud of our teams’ work here.

Susan Li

Chief Financial Officer

Thanks, Mark. Your second question on Muse for small business: we launched it in late September as an extension of Muse, not a standalone product—it’s a new set of skills and connectors within Muse to help SMB owners operate and find new customers. With just a few clicks, it connects to a business’s Instagram account, analytics, Facebook page, Meta ads account, and dozens of third-party connectors like Shopify, QuickBooks, Canva, etc. It already knows what the business sells, its brand voice, and what customers are asking.

Muse can take actions from analyzing ad and content performance, drafting campaigns, replying to customers, flagging issues, and we’re actively listening to SMB users to identify key use cases. We see them using it for ad optimization, organizing DMs, expanding their sales footprint to keep websites up-to-date with inventory and new products, handling employee load balancing. We're closely watching how businesses use Muse to ensure new features are thoughtfully released and maximize utility for SMBs.

We also have unique touchpoints; I mentioned the 500 million self-identified small businesses at the start. Many consumer product investments will also help drive these businesses. There are multiple excellent channels to reach small businesses globally, so we already have powerful platform leverage. We’re excited about that and believe that the shift to AI unlocks business as a new opportunity.

Ross Sandler

Barclays

Great. Mark, you mentioned this before, but could you provide some high-level metrics for Muse per-user costs today and how you expect the cost curve to decline over time? What are you doing to bend that cost curve?

Susan Li

Chief Financial Officer

Happy to take that. Mark, please add anything I miss. We don’t have specific numbers for Muse cost structure today. As Mark said, it’s quite compute expensive, so we’re focusing on compute efficiency—that will be key to scaling Muse and developing it into the business we want.

There are a few aspects to achieving that efficiency. More recently, that includes using off-peak or spare compute for some workloads, using smaller models. There is work on inference efficiency efforts, all targeted at higher GPU utilization and other product efficiency levers.

Long-term, we expect—as LLMs scale—to do more with fewer tokens and to use the AI models themselves to help us run the software more efficiently. So we see a lot on the efficiency roadmap, and expect that AI-driven technical progress in this arc will yield much larger savings over time.

Mark Stephen Mahaney

Evercore ISI Research

Two questions. Over the years, Meta has followed a playbook: launch a new surface, build user growth and engagement, then layer on monetization. Do you see a similar playbook here, or does it need to move faster because Muse’s costs are much higher than past launches?

And, to clarify your commercial datapoint—roughly 25% users, and over $200 ASP—does that include travel purchases or just retail?

Mark Zuckerberg

Founder, Chairman, and CEO

My guess is we’ll move earlier than the usual playbook on monetization—for a few reasons. First, the scale of the numbers, both for costs and opportunity, means making progress here is important in order to scale it up. Second, a large portion of what people are doing is commerce.

On your question: yes, it includes travel. We may treat travel as a bit different, but it’s actually one of the top non-retail Muse use cases, since people are planning all kinds of things which may involve our apps. So that’ll be important to deliver consumer value, and in doing so, we can build a model where we share in the value created for users. Figuring out the math there is important in order to justify the compute build-out we want to reach everyone.

There’s also what Susan mentioned previously—I’d add that based on recent trends, my expectation is that in a year, we’ll be able to deliver the same intelligence in a smaller model at a fraction of the cost of Muse Wonder or any current frontier model. If that represents a sufficient baseline for Muse product-market fit, we’ll keep making efficiency progress through ongoing model R&D.

It’s unclear if most people will want to stick with today’s model level in a year, since progress will continue. So we’ll need to push all these things: the revenue side, scaling, ongoing model improvements, and efficiency. That will let us both fund/scale compute build-out and reach billions. We want to get there as fast as we can.

Part of why we don't charge everyone up-front is we want to build the Muse network, which we believe is more valuable for people when many users collaborate together. That’s really important and exciting. Again, few companies are as well positioned as us to do this. We’re in a great position; we plan to move aggressively.