Meta Platforms Inc (META.US) reported third-quarter total revenue of $64.6 billion, up 26% year over year, with fourth-quarter revenue guidance of $73 billion to $76 billion. The earnings call centered on the Muse personal agent: within 1.5 months of launch, more than 10 million users in the US and Canada have set it up, daily usage stands at about 25%, shopping is the largest use case, and average purchase value exceeds $200. Management expects compute efficiency to improve more than 10x next year and is planning multiple monetization paths including subscriptions, transaction commissions, and enterprise APIs.
Meta's third-quarter total revenue was $64.6 billion, up 26% year over year, including advertising revenue of $63.1 billion, up 26%, and Family of Apps other revenue of $1.1 billion, up 59%. GAAP operating income for the quarter was $13 billion, down 37%, with an operating margin of 20%, mainly dragged down by $9.9 billion in legal expenses related to the multistate attorneys general settlement; excluding that expense, operating income would have risen 11% year over year. Net income was $9.9 billion, or $3.83 per share. Capital expenditures were $34.9 billion, free cash flow was negative $2.1 billion, and cash and marketable securities at quarter-end were $82.4 billion against $83.7 billion in debt. Headcount exceeded 66,000, down 12% from the second quarter. The company also announced it will create a new class of non-voting stock to be distributed to existing shareholders through a stock dividend, effectively equivalent to a 2-for-1 split, with trading expected to begin in late January 2027.
The call's core focus was the progress and monetization path of the Muse personal agent. Muse has been live for 1.5 months, with more than 10 million people in the US and Canada having set it up and daily usage at about 25%, more than double Threads' early level; shopping is the most common use case, with more than a quarter of users participating and average purchase value above $200. Management made clear that Muse has not yet launched globally, with compute intensity the main constraint, but expects compute cost efficiency to improve more than 10x next year, at which point it can be profitable once revenue reaches scale. On the business model, the free tier is the foundation, with monetization through multiple paths including transaction commissions, premium subscriptions, enterprise APIs, and Muse Code. The Meta One unified subscription service is expanding globally, with more than 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 a personal agent protocol, partnering with Walmart, Stripe, Shopify and others to establish agentic commerce standards.
On the outlook, Meta guided fourth-quarter revenue of $73 billion to $76 billion, full-year capital expenditures of $135 billion to $145 billion, and full-year total expenses of $175 billion to $179 billion (including $9.9 billion in legal expenses). Reality Labs losses are expected to narrow to about $15 billion next year. Management did not provide specific 2027 revenue guidance but said AI investments have already delivered incremental gains to the core business through higher engagement, better recommendations, and improved monetization efficiency, and it expects strong full-year revenue growth in 2027.
On advertising AI progress, the generative retrieval architecture now covers about a quarter of non-connected Feed views and half of Reels views, the GEM model expansion has lifted off-site conversions by 1% to 1.7% and app ad conversions by 9.6%, nearly 10 million advertisers use generative AI creative tools, and partnership ads have an annualized revenue run rate above $15 billion.
The following is the transcript of Meta Platforms' third-quarter 2026 earnings call:
Executive Remarks
Mark Zuckerberg, Founder, Chairman and CEO
All right. Hey everyone, and thanks for joining us today. It's been a busy and exciting few months. Our family of apps now has 3.64 billion people using it daily, and engagement on both Facebook and Instagram is at an all-time high, driven by strong growth in video. Threads has more monthly actives than X and continues to grow quickly. WhatsApp has record messaging volume and is increasingly an important platform for agents.
Over the past few months, we've hit several major milestones toward our goal of delivering personal superintelligence to everyone. Our Muse personal agent is one of the most successful product launches we've ever had. We launched Muse Wonder as our first frontier model and released several new glasses and accessories with Muse integration, including coming soon to our new audio glasses. We've been discussing the vision of personal superintelligence for more than a year. We want to help every person have an extremely capable personal agent that understands your goals, can help you achieve them, and can improve your life in whatever way you want — whether that's improving relationships, helping you save time, save money, start a business, create new things, or anything else you can think of. Muse is the product form of that vision. So we now have the foundational technology to achieve this. There's still a lot of work ahead, but our starting position is very strong.
In just 1.5 months, more than 10 million people in the US and Canada have set up the Muse personal agent. Growth remains strong, and we've been at the top of the App Store almost every day. Engagement and retention have also started strong. About 25% of people who set up Muse use it every day. For our earliest adopters, it seems the longer they use Muse, the better retention gets. As a comparison, that's more than double where Threads was at launch and above what we believe other AI apps start at. Muse's most common use case is shopping, with more than a quarter of Muse users participating in shopping and an average purchase value above $200. Plaid told us they've never seen this kind of activity on their connectors for any other service. In the first week, we had to work with Stripe to raise credit limits because transaction volume exceeded what any of us expected. Beyond shopping, the most common connectors are personal productivity and connecting Muse to your other services like email and calendar. This shows the importance of the Muse Secure VM we built to deliver state-of-the-art privacy and security, and we look forward to launching Muse Confidential VM soon, where even Meta won't be able to see your content.
We're also seeing Muse Wonder further improve quality, engagement, and retention, which shows Muse will get better as we continue to raise intelligence. Now that Muse has clear product-market fit, we're entering a new phase — a more operational phase focused on scaling. The goals for the next phase will leverage many of Meta's strengths. We need to do three things: scale the product, grow revenue, and optimize compute costs. We expect to scale Muse to serve billions of people. That's both our mission and our strategy. On mission, we believe we should put the power of personal superintelligence in everyone's hands. We believe this is the path to ensuring a positive future for all, with individual empowerment as the source of prosperity and a balance of power as the foundation of safety. On values, we believe broadly providing this capability is the best thing we can do.
It's also our strategy, because we believe a scaled network of billions of Muses will be able to accomplish more for individuals and businesses by working together. So we want to create a large-scale Muse network as quickly as possible. From a business perspective, average revenue per user across our existing Muse users — sorry, average revenue per user across our existing family of apps users is several hundred dollars per year, and we expect Muse to reach higher levels over time, since we're already seeing a lot of commercial activity happening. This will take some time, but we're starting now.
Our business model will have several components. For most people, the product should be free, and we'll build commercial tools and charge fees based on the transactions we help facilitate. We expect this will be paid by businesses. For users who want the most advanced models or to use extra tokens, they subscribe and pay directly. For business owners who want extra features to grow their businesses using Muse, or companies that want their employees to use Muse, they can also subscribe and pay directly. For those who want Muse to help improve the personalization and quality of content and ads on Meta platforms, they can connect those experiences too. To clarify, when I said average revenue earlier, I meant Family of Apps revenue for our existing Muse users.
On compute, we're building capacity to serve more than 1 billion people over the next few years. Near term, running Muse is very compute-intensive, but we expect to improve its cost efficiency by more than 10x next year, which will make it profitable once the revenue line reaches scale. If compute efficiency continues to improve rapidly over the next few years — which we think is highly likely — then Muse will be highly profitable for many years. Near term, the strength of our core business allows us to scale the Muse network before achieving profitability, and we believe we're the best-positioned company to advance this strategy.
Another point about compute economics is that since training is a huge fixed cost, now that we have leading models, we want to serve as much high-margin inference demand as possible. That's why we're building the Meta enterprise platform led by CJ Desai. Muse Wonder and the more advanced models we're training can serve additional high-margin services outside the Muse network, including our APIs, Muse Code, and other enterprise services. A fundamental feature of our business is that we have many good uses for compute. It's hard to predict exactly how the Muse network, our enterprise platform, and Muse Wonder and subsequent models will accelerate the growth rate of our core business. But by building multiple ways to deploy these advanced models, we increase the chance that one or more of them significantly exceeds our expectations. Beyond serving other businesses, we haven't yet found the limit of applying compute to accelerate our own core business. We've implemented an evaluation of Muse Wonder to improve its effectiveness in autonomously improving our recommendation and ads systems. As our upcoming models become more advanced, I expect we'll see these systems improve faster.
All right. Now on models. Muse Wonder is state of the art across many dimensions that matter for personal superintelligence. This validates the importance of training models ourselves to deliver the best experience. Beyond Muse, we've also seen that since we began launching new models in April, per capita engagement with Meta AI has more than doubled. While Muse Wonder is our most important launch, I'm very pleased with our execution speed. Over the past few months, we released Muse Spark 1.2 and 1.3, Muse Code, 2 open-weight models Muse Glimmer and Muse Spark Open, and Muse Voice.
Looking ahead, as the field advances quickly, we plan to both push the frontier of intelligence and distill these models into smaller, more efficient product models optimized for personal agents. My view is that we'll be able to build superintelligence through scale. So we're methodically climbing that ladder. We're already developing larger models after Muse Wonder that will be very impressive and capable. We're also developing upcoming new small models that will deliver Muse Wonder-level intelligence with higher efficiency.
Next. Our work on personal superintelligence is now also the focus of Reality Labs. Muse will come to Meta glasses starting next month, which will be an important step toward this goal. People already love many features of the glasses — music, calls, capturing moments, hearing assistance, and more. But over time, we expect Muse and having your personal superintelligence with you all day to become the most important part of the experience. Muse is also central to the glasses business model. Again, our mission and strategy is to serve everyone. We believe in putting the power of this technology in everyone's hands. So we price the device not to make large profits on hardware, but to build a large software and superintelligence business on top of it over time. Glasses sales continue to grow quickly, and now with Muse, we expect paid subscriptions to increase and this business to improve significantly from here.
I also shared at Connect that we're changing our approach to virtual reality, only releasing new devices when we achieve major breakthroughs, like the Meta VR glasses we launched this year — of course, the first VR experience in a glasses form factor. I still believe the path to delivering holographic experiences that blend the physical and digital worlds into everyday glasses will first come through VR, but we'll pace our investments there more deliberately to prioritize delivering personal superintelligence.
So putting these factors together, we expect Reality Labs' P&L to improve from here, with operating losses narrowing to about $15 billion next year and trending toward profitability from there.
Before I wrap up, I want to take a moment to talk about the amazing speed we're seeing. An important part of our approach has always been moving fast so we can learn quickly and serve users at scale better. But recently, things have accelerated. Over the past year, we launched Edits, which has surpassed 50 million active users. New apps like Forum and Seller were built by small teams and are scaling quickly to serve power users. Muse went from an idea on a whiteboard to a successful product in about 8 months. We're about to launch Muse Charm, which will go from concept to a hardware device in people's pockets in less than a year.
I'm proud of what we're building. People have different views about the future, but at Meta, we believe the future belongs to everyone, and putting power in people's hands is how we get there. I think this matters. There needs to be a company that stands for these ideas. But it's not just ideas, it's execution. So I'm grateful to everyone working so hard to deliver this to billions of people around the world. As always, thank you to everyone on this journey with us. Now over to Susan.
Susan Li, Chief Financial Officer
Thanks, Mark, and good afternoon everyone. Let's start with segment results. Unless otherwise noted, all comparisons are year over year. Third-quarter Family of Apps total revenue was $64.2 billion, up 26%. Third-quarter Family of Apps advertising revenue was $63.1 billion, up 26% on both a reported and constant-currency basis. In the third quarter, total ad impressions across our services grew 10%. Impression growth was healthy across all regions, driven by engagement and user growth as well as ad load optimization. The global average price per ad increased 14% year over year, driven by improved ad performance and an easier macro comparison to last year's third quarter. This was partially offset by strong impression growth, especially from lower-monetizing surfaces and regions.
Family of Apps other revenue was $1.1 billion, up 59%, driven mainly by WhatsApp paid messaging and subscription revenue. In the Reality Labs segment, third-quarter revenue was $443 million, down 6%, as Quest headset sales declined, partially offset by continued strong growth in AI glasses revenue.
Now on consolidated results. Third-quarter total revenue was $64.6 billion, up 26% on both a reported and constant-currency basis. Third-quarter total expenses were $51.6 billion, up 68% year over year, including the $9.9 billion expense related to the August 2026 multistate attorneys general agreement. The year-over-year increase was driven mainly by legal-related costs, infrastructure costs, and third-party AI token costs. Infrastructure cost growth was driven by higher third-party cloud spend, depreciation, and data center operating costs. We ended the third quarter with more than 66,000 employees, down 12% from the second quarter following the May 2026 layoffs.
Third-quarter GAAP operating income was $13 billion, down 37% year over year, with an operating margin of 20%. Excluding the third-quarter legal expense related to the aforementioned legal settlement, third-quarter operating income would have grown 11% year over year. Our tax rate for the quarter was 24%, above our guidance range, driven mainly by 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 network infrastructure. Free cash flow was negative $2.1 billion. Cash and marketable securities at quarter-end were $82.4 billion and debt was $83.7 billion.
Now turning to business performance. There are two main factors driving our revenue performance: our ability to deliver engaging experiences for our community and our ability to effectively monetize that engagement over time. On the first point, we continue to see significant gains from our content recommendation initiatives. On Instagram, third-quarter time spent per user reached an all-time high globally, growing double digits year over year, with Reels and Feed both strong. On Facebook, Reels time spent reached an all-time high in both global and US, growing double digits year over year. Threads also continued to scale, with time spent per user growing more than 20% year over year.
Engagement growth across our services continued to benefit from improvements in relevance, personalization, and freshness of recommended content. Our investments in retrieval and ranking for the recommendation engine continued to drive positive results on Instagram and Facebook. For example, on Instagram Reels, our relevance release in the third quarter increased Reels time spent by more than 4%, helped in part by our upgraded model architecture — ranking a series of Reels at once rather than evaluating each Reel independently — and improvements we made to add richer watch-time signals. On Facebook Feed and Reels, we completed the global rollout of our new generative retrieval architecture, which is LLM-inspired and uses semantic representations to tag content pieces and generate latent recommendations, helping our systems surface relevant content before it accumulates significant engagement. This new architecture now covers about a quarter of non-connected Feed views and half of Reels views, increasing Reels watch time by 1%.
Our recommendation work also helps us deliver fresher, more personalized content. Nearly 45% of recommended Instagram Reels are now less than 1 day old, up more than 11 percentage points from the second quarter. On Facebook Reels, we doubled the amount of history our ranking models can consider, increasing US Reels watch time by 1%. LLMs also help us understand content more deeply. We continue to process public Instagram posts daily through LLM-based content understanding systems. By the end of the quarter, these signals had improved engagement on Instagram search results by 1%. During the quarter, we also began migrating content understanding models to our first-party Muse model family. In offline evaluations, these models outperformed our previous production models and reduced per-unit content processing costs across Instagram, Threads, Facebook, and ads by more than 80%. Beyond improving quality, having more of this stack gives us greater flexibility to develop models around the signals that matter most for personalization.
Turning to the second driver of our revenue performance — improving monetization efficiency. We see significant opportunities to continue improving the performance of businesses using our services. In our ads system, we continue to drive performance gains by deploying deeper prediction models that understand users more deeply and deliver ad experiences better aligned with their unique preferences, while also making our infrastructure more efficient. Last quarter, we introduced Meta Generative Recommender, which uses LLMs to reason about ad content and user preferences to predict the best ad for each person. In the third quarter, we launched this generative retrieval approach on Facebook Feed, increasing conversion rates in targeted segments by nearly 2%. Additionally, we expanded our GEM model for ad ranking to more surfaces, increasing conversion rates for off-site conversion campaigns by 1% to 1.7% and app ad conversion rates by 9.6%.
We continue to scale our retrieval system efficiently. In the third quarter, improvements to Andromeda — such as expanded model complexity and throughput — increased 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. The Meta AI business assistant is becoming a more proactive and capable partner for advertisers. We estimate that advertisers who adopt its suggestions achieve nearly 5% higher conversion rates at the same cost per conversion and increase ad investment by 9%, compared with similar advertisers who don't adopt them. We're rolling out more capabilities, including creative insights, stronger business context, and more agentic actions. These features will let marketers plan, create, and optimize campaigns through a conversational experience, and schedule tasks and reminders, while maintaining control through approvals and previews.
Let me share an example of the Meta AI business assistant empowering an advertiser. Simply Organic Maple is a family-owned maple syrup producer in Vermont that used the Meta AI business assistant to improve its campaign messaging. After the business assistant suggested testing longer, more descriptive ad copy, the business increased purchases by 57% and lowered cost per purchase by 36%. Its strongest revised ad achieved a 3x return on ad spend, a 52% improvement over the original. Adoption of our generative AI ad creative tools is also expanding. In the third quarter, nearly 10 million advertisers used at least one of these tools. Advertiser adoption of our video generation capabilities grew 25% quarter over quarter, helped in part by pre-made videos reducing generation time from a single image upload to just a few seconds.
We're also expanding opportunities at the intersection of creators and ads. Partnership ads, which allow advertisers to run ads together with creators, brands, and other businesses, continued to scale quickly in the third quarter. Our partnership ads product grew more than 97% year over year, with an annualized revenue run rate above $15 billion. In the third quarter, we also completed the global rollout of Creator Marketing Hub to all advertisers, centralizing creator discovery, content organization, and partnership ad activation in one place, with more than 10 million creators now discoverable. We're integrating the Meta AI business assistant into the hub so advertisers can search for creators conversationally, identify content for partnership ads, and get guidance in their creator marketing workflows.
Beyond ads, we're also building more ways to monetize engagement across the ecosystem. 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 more than 50 features across Instagram, Facebook, and WhatsApp, including Meta Business Agent and Meta AI, with more than 15 million subscriptions and trials to date. The core experiences of our apps and Meta AI will remain free, while subscriptions offer more specialized capabilities and expanded AI usage.
In addition to monetizing engagement in our 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, with the top three categories being shopping, personal productivity, and creativity. We're encouraged by the breadth of use cases, especially shopping, which more than 25% of Muse weekly active users use. While our near-term priority is expanding adoption and scaling Muse's reach, we see multiple monetization paths, starting with subscriptions.
We also see a major opportunity to extend Muse to small businesses. Based on a recent user survey, we estimate that at least 500 million self-identified small businesses use Meta's family of 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.
Beyond growth, we're focused on building sustainable unit economics. Muse's compute intensity means scaling its user base will require significant investment, and we expect total cost of service to rise. At the same time, we expect to improve compute cost efficiency by more than 10x next year, helped by better use of off-peak capacity, more efficient models, and inference optimization. As revenue scales, these efficiency gains give us confidence in our ability to build a highly profitable business over time.
Now turning to our financial outlook. We expect fourth-quarter 2026 total revenue to be in the range of $73 billion to $76 billion. Our guidance assumes an approximately 1% foreign exchange headwind to total revenue growth year over year based on current exchange rates.
Turning to expense and capital expenditure outlook. We are raising our expense outlook to incorporate the $9.9 billion legal expense recognized in the third quarter. We now expect full-year 2026 total expenses to be in the range of $175 billion to $179 billion. Excluding the third-quarter legal expense, we continue to expect 2026 operating income to be higher than in 2025. At the segment level, we now expect Reality Labs losses to decline to a range of $18 billion to $19 billion from 2025.
We expect 2026 capital expenditures, including principal payments on finance leases, to be in the range of $135 billion to $145 billion, narrowed from our prior outlook of $130 billion to $145 billion. Absent changes in the tax environment, we expect our fourth-quarter 2026 tax rate to be in the range of 15% to 17%. In addition, we continue to monitor active legal and regulatory matters that could significantly impact our business and financial results. For example, we continue to see scrutiny on teen-related issues in multiple markets, with several teen-related trials scheduled in the US in the coming months.
Looking beyond 2026, we're still developing next year's budget and capacity plans, but we expect to continue investing aggressively to support the core business and emerging AI initiatives. Our current capacity and the capacity coming online give us flexibility to allocate resources across model training, core business growth, and the development and support of new products.
Our AI investments are already delivering substantial incremental ad revenue through stronger engagement, better recommendations, and more efficient monetization. Building on that, our MSL investments are creating the technology and infrastructure needed to unlock major new revenue opportunities and enhance capabilities across the business. While these investments will take longer to pay off than core business investments, the breadth of monetization opportunities and internal efficiencies give us confidence in their long-term potential.
Looking ahead, while we are not providing a full-year 2027 revenue outlook, the benefits our AI investments are delivering to the core business, combined with emerging opportunities from model advances and product launches, position us to sustain strong overall revenue growth throughout 2027. Separately, we remain focused on improving Reality Labs profitability. As Mark mentioned, we expect losses to narrow to about $15 billion next year as we reallocate investment to other priority areas, primarily AI.
Finally, today we announced plans to create a new class of non-voting capital stock to be distributed to all existing shareholders through a stock dividend. For each share of existing stock held, shareholders will receive 1 share of the new non-voting stock, effectively equivalent to a 2-for-1 stock split. We expect trading to begin in late January 2027.
In summary, our business momentum continued in the third quarter, with strong execution across core advertising and engagement initiatives, and we're also excited about our progress toward delivering personal superintelligence. Now, let's open the line for questions.
Q&A Session
Brian Nowak, Morgan Stanley Research
I have two questions, one on Muse and one on the enterprise business. First, congratulations on Muse's success. But as you analyze user behavior and the product, could you talk about any constraints on the forward user growth slope you're most focused on — whether capability, model efficiency, capacity, etc. — so we can understand the forward user growth limiting factors you hope to address? Second, on the enterprise business. Mark, could you briefly lay out your vision for Meta's enterprise product offerings and the key value drivers you hope to provide to businesses that adopt the service?
Mark Zuckerberg, Founder, Chairman and CEO
Yeah, I think I can take both of those, and then Susan can add. All right. So for Muse right now, one of the things I pointed out is that we have more than 10 million people who have set it up in the US and Canada, but I think one of the biggest constraints so far is that because this is a compute-intensive service, we haven't rolled it out globally. Part of that is wanting to optimize for each place, and there's some connector work. But part of it is we want to make sure we do this efficiently. So there are pros and cons. The upside, of course, is that it makes the 10 million people using and signing up in the first 6 weeks even more impressive, because these are all Americans and Canadians and concentrated in a few markets. If you actually rolled it out more broadly, you'd expect that number to be significantly larger than what we're seeing today. But on the other hand, we actually have to do the work to get efficiency to a reasonable level. So that's an important part of what we need to do.
The compute efficiency work is also important here, because we want to make sure we've built enough compute to serve what we think is more than 1 billion users if we can hit certain compute efficiency targets, but we clearly have a lot of work to do. That will be one of the limiting factors next year. We want to build a business around it. So I talked about that. I think doing that in parallel with lowering compute — or improving compute efficiency — is important to making this viable from a business perspective, and I'm optimistic about it, but I think it's one of the things we need to do.
Then there's continuing to add product value, which I think we mainly do by continuing to develop new models. That's why I mentioned we first launched Muse on Muse Spark, which is currently the default model for the free tier. We launched Muse Wonder, which people can access today on the paid tier. We know it performs much better, so we're developing more efficient versions of Muse Wonder so we can roll it out more broadly. But you can imagine, as we iterate 2 or 3 models forward, when you have significantly higher intelligence and the ability to solve more problems, delivered at a more affordable price point, I think that will unlock more value for people. So we're very excited about that.
Turning to the enterprise part, because I think the two are somewhat connected, the main enterprise products near term will be model APIs and Muse Code. We've had enterprise agents scaling for a while, and all three of these things are growing quickly. So we're excited about that. Over time, we want to build a full Muse agent commerce network. There are business agents, which we've talked about on previous earnings calls, that help businesses with sales and customer support. But we see a lot of people actually using Muse to help run small businesses in particular. We think there's an opportunity to help create an agent that people use to really automate, to handle many of the difficult and time-consuming but non-core parts of starting a small business. So I think that's a big deal. And then making a version of Muse that people can bring into different business environments and work there, we think that will be a big deal.
Beyond that, building what I'd call the Muse network, meaning making Muse not just a single-player game — you have your Muse — but making it so your Muse can collaborate with your friends and family. If you want to buy something, rather than just navigating to that business's website, you're talking to their agent. I think making all of this work well together will unlock enormous value at scale. So I'm very excited about all of it. I think the enterprise use case will be an important part of it, but it will also create very valuable utility gains for consumers. So I'm very excited about both.
Susan Li, Chief Financial Officer
I'd add to Mark's comments on enterprise. AI is creating a lot of disruption in enterprises today. From what we're seeing, we really don't think the traditional enterprise companies have locked this up. In fact, some of the leading companies in this space today didn't exist a few years ago. Those companies invested heavily in model development, and we've done the same, and we have increasingly competitive models that can benefit all users, including small businesses and enterprises. There are already 500 million businesses using our products today. So we really believe AI will reshape how businesses operate. We expect businesses of all sizes to use AI agents across software development, productivity, sales, marketing, customer service, and more. We have differentiated assets to serve this market, combining models, infrastructure, global distribution, and long-term relationships with businesses. We hope this will become a new diversified recurring revenue stream for us, supporting multiple business models including subscriptions and usage-based APIs and messaging, as well as other revenue lines. We expect this to be a multiyear opportunity, and we intend to scale investment based on demonstrated customer demand and product-market fit.
Eric Sheridan, Goldman Sachs Group Research
First question, following up on Brian's question. As you think about the friction points in agentic commerce, as a company, how do you think about addressing the friction where aggregators or suppliers don't want agents interacting with their platforms or services, and how do you think about reducing that friction or creating the right balance between existing services and how agent behavior may change over the medium to long term? Second question, with the new stock class also announced alongside tonight's earnings, is there any updated thinking on how you view sources of capital and optimizing the capital structure to ensure appropriate investment to capitalize on the opportunities AI presents over the next few years?
Mark Zuckerberg, Founder, Chairman and CEO
I can take the first one. So right now, what we're seeing is that the vast majority of businesses are very excited to support this, because they recognize that this is not only a new opportunity to increase sales and experience quality for existing customers, but could also bring in new customers. So across the commerce ecosystem, different players will adopt this at different speeds, and how our strategy evolves here will depend on how progress goes. But overall, I'd say we've been very positively surprised by how many businesses not only want to figure out how to make Muse work more effectively with their services, but are proactively reaching out to us to integrate around this. I think that's a very good sign. Obviously, not everyone wants to be an early adopter. That's fine. We'll work with people on terms that make sense for them. But I do think, over time, making all of this work well will be very valuable for everyone, and I think we're currently on a good track there.
Susan Li, Chief Financial Officer
Yeah. I'd add, and I'll take your second question as well, I think agents will be an important part of how reality evolves, and I think everyone will have to adapt to what the new era of agentic commerce means. We obviously have a lot of businesses very excited to work with us. There are also businesses understandably focused on protecting customer relationships. We expect the market will ultimately move away from agents trying to navigate every digital experience like a human. I think over the long term, this will head toward more permissioned, structured interactions, including authenticated APIs, approved agent partnerships, or direct communication between consumer personal agents and businesses' own agents. We're helping establish the standards needed to support that model. Earlier this month, we previewed the personal agent protocol, an open standard across Muse and Meta business agents that we're developing with industry partners including Sierra, Walmart, Stripe, and Shopify. The protocol is designed to define how consumer personal agents can interact securely and reliably with business agents, while preserving appropriate controls for businesses and customers. So again, I think adoption may be uneven and take time, and near term we expect many experiences — many experiences will remain AI-assisted rather than fully autonomous. But as trusted integrations develop and agents prove they can generate incremental value and demand and operate safely, we expect more and more businesses will be excited to participate.
Your second question was about the announcement of the new stock class. I should note that the purpose of creating this new stock class is to create a publicly traded non-voting stock that can be used for capital raising, employee compensation, acquisitions, and other general corporate purposes, while preserving the voting rights of existing Class A and Class B shareholders. We believe this stock class will maintain our founder-led structure. It will broaden our financing tools and give us flexibility to invest in the unprecedented AI-driven momentum we see across parts of our business. We currently have no specific financing plans to announce, but we think establishing the right capital structure framework is prudent to ensure we can access diversified sources of funding.
Douglas Anmuth, JPMorgan Research
I have two questions. First, I want to follow up on Susan's last comments about Muse's evolution and more enterprise adoption of agents. I'd like to understand your view on how things change and shift when moving toward more permission-based and agent-to-agent transactions. What exactly does that unlock over time? Second, on expenses and capital expenditures, thinking about 2027. I know you're not really there yet. But if I go back to some of the prior comments, you talked about maximizing compute in 2026 and 2027 and then having more flexibility in 2028. A year ago, you talked about 2026 dollar growth being significantly greater than 2025, and then you also used annual operating income growth. I'm just curious if any of these can be connected to help think about 2027.
Susan Li, Chief Financial Officer
Thanks, Doug. On your first question about how the agent ecosystem evolves, I wish I could predict it more perfectly. I think it will take some time. Again, I think it will be somewhat uneven. But I certainly think that when I use Muse today, it's clear that many parts of the experience would be better if Muse could interact more seamlessly with a business agent on the other side. For example, if Muse has to call a business on my behalf, I think it would be easier for the business if that call were handled by a business agent representing the business. I think one thing that's really unique about Muse, and what makes the product experience so good, is the degree of initiative it shows on my behalf, right? I can give it a task, and Muse will go execute that task, and even if it hits obstacles or can't find the right thing, it needs to forge new paths to help me find what I'm looking for. So I think that dimension of initiative shown by consumer agents or business agents will cause more transactions to happen or more activity to happen broadly, compared with today when more actions are taken — when latent demand expresses itself — compared with me being able to have Muse continuously search for things for me, regularly give me updates on what I'm looking for, and send me 10 new books a month. Agents should do similar things for businesses, and I think that creates opportunities, again, for a new, more proactive framework over time for what agent-to-agent interactions look like.
On your second question about 2027, we're in the early stages of developing next year's budget and capacity plans. So I'd say we're at the stage of evaluating many investment proposals with attractive return profiles across different time horizons, and we want to figure out how to fund as many of them as possible within the framework of our overall budget and financial goals. But this is very dynamic and still quite early. So we don't have more specific or quantitative information to share until the fourth-quarter call, which is typically when we share guidance for the following year.
Justin Post, BofA Securities Research
All right, I have two questions, one for Mark and one for Susan. Mark, you have a powerful new model in hand, and it seems like there's a lot you can do internally in addition to externally. But specifically in the ads business, you've re-architected the system around AI. As you deploy that new model, how much more improvement is left? How do you think about that? And then for Susan, you mentioned strong revenue growth next year. I'd like to know whether that's advertising, or whether it includes some incremental AI revenue materializing next year?
Susan Li, Chief Financial Officer
I'm happy to take both of those. All right. On your first question, I think our ads business has benefited for years from our continued investment in what we call traditional AI machine learning models as well as the more recent generation of models. I'd say there are three themes worth outlining. The first is the traditional roadmap of performance improvements in our pipeline — what I'd describe more as business-as-usual machine learning work. That continues to be a strong and healthy pipeline. As we look at 2027, again, we're not providing an outlook, but I'd say our investments and product improvements in engagement and ad initiatives give us confidence in meaningful room for growth throughout 2027. That also relates to your second question. But I should say that, basically, I think the ranking and recommendation pipeline is healthy.
The second theme is that we're excited about some of the new research bets as we move toward more LLM-oriented AI infrastructure, to help us reason more deeply about people's preferences and what they're interested in across organic content and recommendations, and we expect we'll be able to unlock benefits for the ranking and recommendation engine from that research. The third thing is that we expect generative AI tools — already very valuable to advertisers — to continue to grow in terms of the capabilities we provide to advertisers through the Advantage+ suite, which makes it easier for advertisers to quickly and seamlessly generate creative, or take an image and generate video or generate voice, all of which I think serves to make ads increasingly personalized for the recipient. So I'd say we feel there's meaningful room for growth across all three areas.
Your second question about revenue growth next year. As I mentioned, I think we feel good about the budget investments we expect to make in the core ads and engagement initiatives next year. We're also building several new AI businesses that we expect can begin contributing incremental revenue in 2027, but more importantly, can scale meaningfully over time. These will include monetizing our AI products through subscriptions, transactions, and ads, and growing enterprise products, as we talked about in a previous question. So I'd say that while many of these opportunities are still early, and our near-term focus is really on building high-quality products and achieving product-market fit and scaling product-market fit, I think our track record in reaching billions of people and building businesses at scale makes us feel good that AI will become a meaningful source of revenue diversification over time.
Mark Shmulik, Bernstein Institutional Services LLC Research
Mark, coming out of Connect, I kind of got the feeling that we're at the dawn of a new era for the company, with the business recentered around Muse. Is that a fair assessment of how things are progressing internally? And if so, could you talk more about how some of the business mandates and outcomes might change? And then, Susan, it was interesting to see the release cadence, with the enterprise announcement coming before some of the small business announcements. I appreciate some of your comments on the SMB agent opportunity. But could you share more detail on servicing and scaling the SMB product?
Mark Zuckerberg, Founder, Chairman and CEO
Yeah. I can speak briefly to the first one. I mean, more than a year ago I wrote about personal superintelligence as the north star for our work. This basically fits with what we've been doing at the company for more than 20 years, focused on empowering individuals, giving people a voice, and giving people tools to connect in the ways they want. When we extend that to our work around AI now, we think the most important thing we can build for the next generation of technology is — basically putting the power of this technology in everyone's hands, and that's the personal superintelligence vision. So we've been working on this for a while, and that's when we restarted around MSL and built the lab. A lot of it is the idea that we need to build models specifically good at being personal agents, and we'll be able to build the best experience by owning, building, and designing end-to-end technology. That includes models and agents. This has been brewing for a while. So has the vision around glasses. We want to build the next generation of hardware platforms so we can deliver the best end-to-end experience for people. I think that's also starting to take shape, and I think it will be even more so in the coming years. These are long-term bets. I feel like that's what we're here to do, make long-term technology bets and design the best end-to-end experiences for billions of people around the world. I think as long as we keep doing that, then over time the business will take care of itself, as long as we keep executing well, and I'm very proud of how the team has performed on this.
Susan Li, Chief Financial Officer
Thanks, Mark. Your second question was about Muse for small businesses. So we launched it in late September, and you can think of it as an extension of Muse rather than a separate product — a new set of skills and connectors within the existing Muse app designed to help SMB owners operate and find new customers. It connects in just a few clicks to a business's Instagram account, analytics, or Facebook page or Meta ad account, as well as dozens of third-party connectors the business already uses, such as Shopify, QuickBooks, Canva, and more. It already understands what the business sells, its brand voice, and what customers are asking. Muse can take actions, including analyzing ad and content performance, drafting campaigns, customer replies, and flagging issues, and we're actively listening to SMB users to understand their main use cases. We see them using it for ad optimization, direct message organization, expanding their sales footprint to keep their website up to date based on inventory and new products, and handling employee load balancing. So we're watching fairly closely how businesses use Muse to make sure we're thoughtful about the new features we release for Muse and make it as useful and efficient as possible for SMBs. I'd also say we have unique touchpoints, again, I mentioned 500 million self-identified small businesses in my opening remarks. We think a lot of the investments we're making in consumer products will carry over to help drive these businesses. We have multiple strong channels to surface these opportunities to small businesses around the world. So this is an area where we feel we already have a very strong existing platform, and we're very excited about it. We also believe the transition to AI broadly unlocks businesses as a new opportunity for us.
Ross Sandler, Barclays
All right. Mark, you touched on this earlier, but could you elaborate a bit on some high-level metrics around Muse's cost structure per user today, and then how you see the cost curve declining over time? What are you doing to bring that cost curve down?
Susan Li, Chief Financial Officer
I'm happy to take that. Mark, you should add anything you feel I missed. We — we don't have specific numbers to share today on Muse's cost structure. But of course, as Mark said, it's quite compute-expensive, and making sure we focus on compute efficiency will be one of the key vectors for being able to scale Muse and grow it into the business we want. I think there are several dimensions to achieving that efficiency. I think some of the more near-term pieces include using off-peak or spare compute capacity to run certain sets of tasks, using smaller models. There's a lot of work to do across different inference efficiency initiatives, all really aimed at higher GPU utilization and other product efficiency work we can do.
I think over the long term, I think this is something we expect to be able to — as LLMs scale, do more with fewer tokens, and rely on AI models themselves to help us improve the efficiency of the software we use to run models. So I think this is an area where we both feel there's a lot of work on the efficiency roadmap that will help us bend the curve, and we also expect that the arc of AI-driven technological progress applied here will help us achieve even greater savings over time.
Mark Stephen Mahaney, Evercore ISI Institutional Equity Research
Two questions. For years Meta has had a playbook where you introduce a new surface and build user growth and engagement, and then layer on monetization. Do you think this is a similar playbook here? Or does it have to be faster, because the costs associated with Muse are arguably much greater than when you launched those surfaces in the past? And then just to clarify the commercial data points you gave, about 25% of users and $200-plus ASP. Are you including travel purchases? Or is that just retail data points?
Mark Zuckerberg, Founder, Chairman and CEO
My guess is that I think we'll move to monetization earlier than the normal playbook, for a few reasons. One is the scale of the numbers we're talking about — whether costs or opportunity — I think makes making progress on this important to being able to scale it. The other is that a large part of what people are doing is commerce. To answer your question, that also includes travel, I think we may treat travel as a different case, but travel is actually one of the other top use cases for Muse, basically people planning different kinds of things that they might otherwise get through our family of apps. So I think this will be an important thing for us to do to deliver the consumer value people want, and in doing so, I think we'll be able to build a model where we capture a portion of the value we help create for people.
So I think it will — figuring out all the math on this is very important so we can continue the compute buildout we want to build to reach everyone scale. Beyond everything Susan said in answering the previous question, I'd add that we're just seeing this trend, and year over year, I expect we'll be able to deliver the same intelligence with a much smaller model a year from now at a fraction of the cost of the current Muse Wonder or frankly any current frontier model. So if that represents the kind of baseline level that people seem — that's sufficient for Muse product-market fit, I think we'll make progress on efficiency through continued model development. What's less clear, of course, is that I don't think most people a year from now will want to use the current model level of intelligence, because that will progress too. So I think we need to work on all of these things, including the revenue pieces and scaling different aspects, as well as advancing model development and all the efficiencies, to be able to both fund and support the compute buildout and reach billions of people, again, and we want to get there as quickly as possible. Part of the reason we're not just charging everyone upfront for this is that we want to build this Muse network, which we think will be more valuable to people at scale when many people are using Muse. That way, they can collaborate and work together. I think that's really a very important state, and we're very excited about it. Again, I think there aren't many companies in the world positioned as well as we are to do this. I think we're very well positioned to do it, and we plan to move fairly aggressively.
Comments