Meta Platforms: Muse Hits Product-Market Fit With 10 Million Users in 1.5 Months, but Serving Costs Must Fall 10x as Free Cash Flow Turns Negative
Q3 2026 earnings call, October 21, 2026: AI agent gains traction, a $9.9 billion legal charge distorts the quarter, and a nonvoting share class opens a new funding channel
Meta used its third quarter call to argue that its consumer AI agent, Muse, has crossed from experiment to scalable product, while also conceding that the economics are not yet in place. CEO Mark Zuckerberg said more than 10 million people in the U.S. and Canada have set up Muse personal agents in roughly 1.5 months, that the app has been #1 in the App Store almost every day, and that about 25% of people who have set it up use it on any given day. He called that "more than twice the level that Threads started at or where we believe that other AI apps started at as well." Meta is now moving into what Zuckerberg described as "a more operational phase" built around three tasks: scaling the product, growing revenue and optimizing compute costs.
Muse: Early Usage Is Commerce-Heavy, and Monetization Will Come Sooner Than Usual
The most notable data point was how people use Muse. Shopping is the most common use case, with more than 25% of users engaging in it and an average purchase size above $200. Zuckerberg said Plaid had not seen connector activity like this on any other service, and that Meta had to work with Stripe to raise credit limits in the first week because volumes exceeded what "any of us expected." Travel planning is another top use case. Personal productivity and creativity round out the leading categories.
That usage pattern is shaping the business model. Zuckerberg said the product should be free for most people, with Meta taking a fee on transactions it facilitates, paid by businesses. Heavier users would pay subscriptions for advanced models or extra tokens, and business owners or employers could subscribe for additional features. Users can also connect Muse to Meta's content and ad experiences to improve personalization. In response to a question from Evercore ISI's Mark Mahaney, Zuckerberg said monetization will come "somewhat sooner" than Meta's usual playbook of building engagement first and layering in revenue later, citing "the size of the numbers that we're talking about in terms of both the costs and the opportunity." He framed the goal as capturing "a percent of the value that we are helping to create for people."
Management gave no revenue figure for Muse. The reference point offered was that Family of Apps revenue per American user is "several hundreds of dollars annually," with Muse expected to eventually exceed that because of the commerce activity. Zuckerberg clarified that this figure refers to Family of Apps revenue from existing Muse users, not revenue generated by Muse itself. Susan Li said new AI businesses could begin contributing incremental revenue in 2027, but the emphasis was on scaling over time rather than near-term contribution.
Compute Economics: A 10x Efficiency Gain Is the Condition for Profitability
The cost side is the main caveat. Li said plainly that Muse's compute intensity "means that expanding its user base will require significant investment, and we expect total serving costs to rise." Meta is building capacity to serve more than 1 billion people over the next few years, and it targets a more than 10x improvement in compute cost efficiency within a year, driven by better use of off-peak capacity, more efficient models and inference optimization. Zuckerberg said this would let Muse be profitable once revenue lines are at scale, and highly profitable over multiple years if efficiency gains continue.
Management declined to share per-user serving costs when asked by Barclays' Ross Sandler. The compute constraint is also already limiting growth. Muse has not launched outside the U.S. and Canada, partly because of connector work by market and partly because Meta wants to improve efficiency first. Zuckerberg noted that a broader rollout would produce a "significantly larger" user count, but that "we actually have to go do the work to bring the efficiency in line to make that make sense." The free tier currently runs on Muse Spark, while the more capable Muse Wonder is limited to the paid tier until a more efficient version is ready. Zuckerberg also cautioned that user expectations will rise with each model generation, so today's efficiency targets are a moving baseline.
Models and Enterprise: Monetizing Training Spend Beyond Consumer
Meta shipped Muse Wonder as its first frontier model, along with Muse Spark 1.2 and 1.3, Muse Code, two open-weights models (Muse Glimmer and Muse Spark Open) and Muse Voice. Engagement per person in Meta AI has more than doubled since April, when the new models began rolling out. Zuckerberg said the plan is to push the frontier while distilling into smaller product models, with larger models already in development and smaller models delivering Muse Wonder-level intelligence "coming out soon."
The argument for the Meta Enterprise Platform, led by CJ Desai, is about fixed-cost absorption: "since training is a large fixed cost, now that we have leading models, we want to serve as much high-margin inference demand as possible." Near-term enterprise offerings are the model API, Muse Code and business agents, all described as growing quickly, though no figures were given. Li said Meta does not believe traditional enterprise vendors "have this locked down" and said it intends to "scale our investment behind demonstrated customer demand and product market fit." This is an early-stage effort, and management framed it as a multiyear opportunity.
Zuckerberg also said Meta has built an eval to test how well Muse Wonder can autonomously improve its recommendation and ad systems. He expects "an increasing rate of improvement" as more advanced models arrive. This is a stated aspiration, not yet a quantified result.
Agentic Commerce: Meta Expects Friction and Is Pushing a Standard
Asked by Goldman Sachs' Eric Sheridan about platforms that resist agent access, Zuckerberg said businesses have been "overwhelmingly" enthusiastic, and that Meta was "very positively surprised" by inbound integration requests. Li was more measured. She acknowledged that some businesses are "understandably focused on protecting customer relationships" and said the market will "evolve away from agents attempting to navigate every digital experience like a human" toward authenticated APIs, approved partnerships and direct agent-to-agent communication. Meta previewed the Personal Agent Protocol, an open standard being developed with partners including Sierra, Walmart, Stripe and Shopify. Li cautioned that adoption will be uneven and that many experiences will stay "AI-assisted rather than fully autonomous" in the near term.
Core Business: Ad Growth Accelerates on Both Volume and Price
The core business was strong. Family of Apps revenue was $64.2 billion, up 26%, with ad revenue of $63.1 billion, also up 26%. Ad impressions rose 10% and average price per ad rose 14%, so growth came from both volume and pricing, helped by ad performance gains and a better macro backdrop than a year ago. Family of Apps other revenue reached $1.1 billion, up 59%, driven by WhatsApp paid messaging and subscriptions. Daily active people reached 3.64 billion, and Zuckerberg said Threads has overtaken X in monthly actives.
Management supported the revenue figures with model-level detail. Instagram Reels relevance launches lifted time spent by more than 4%. A new generative retrieval architecture on Facebook increased Reels watch time by 1%, and nearly 45% of recommended Instagram Reels are now less than a day old. On ads, Meta Generative Recommender drove a nearly 2% conversion lift on Facebook Feed, expanded GEM deployment lifted off-site conversions by 1.7% and app ad conversions by 9.6%, and Andromeda improvements added 1.4%. Partnership ads grew more than 97% to a run rate above $15 billion. Nearly 10 million advertisers used at least one generative creative tool. Meta One, the consumer subscription bundle, has passed 15 million subscriptions and trials.
One cost data point stands out. Moving content understanding models onto the first-party Muse family reduces per-unit processing costs by more than 80% across Instagram, Threads, Facebook and ads, according to offline evaluations. That is a concrete example of the efficiency Meta hopes to replicate in serving Muse.
Financials: A Legal Charge Cuts Operating Income by 37%, and Free Cash Flow Is Negative
Reported profitability was weak. Total expenses rose 68% to $51.6 billion, including a $9.9 billion charge tied to the August 2026 multistate attorneys general agreement. GAAP operating income fell 37% to $13 billion, a 20% margin. Excluding the charge, operating income would have risen 11%. That underlying growth is well below the 26% revenue growth, which reflects heavy infrastructure spend and what Li called third-party AI token costs, alongside higher third-party cloud spend and depreciation. Net income was $9.9 billion, or $3.83 per share, and the 24% tax rate was above guidance because of the nondeductible portion of legal costs.
Capital expenditures were $34.9 billion, and free cash flow was negative $2.1 billion. Meta ended the quarter with $82.4 billion in cash and marketable securities against $83.7 billion in debt, so it is now in a modest net debt position. Headcount fell to just over 66,000, down 12% from Q2 after the May reduction. Li also flagged that youth-related scrutiny continues and that several U.S. youth-related trials are scheduled in the coming months, a reminder that legal exposure remains an open risk.
Guidance: Q4 Revenue of $73 Billion to $76 Billion, Tighter Capex, Lower Reality Labs Losses
Meta guided Q4 revenue to $73 billion to $76 billion, with about a 1% currency headwind. Full-year 2026 expenses are now expected at $175 billion to $179 billion, raised to include the legal charge. Excluding that charge, management still expects 2026 operating income to exceed 2025. Capex was narrowed to $135 billion to $145 billion from $130 billion to $145 billion, which lifts the floor. The Q4 tax rate is guided to 15% to 17%.
On 2027, Li declined to give specifics, saying budgets are "very dynamic" and that guidance will come with the Q4 report. She did say Meta expects to "continue investing aggressively" and that core ad and engagement initiatives plus emerging AI products position the company "to sustain strong overall revenue growth throughout 2027." Asked by JPMorgan's Doug Anmuth about earlier framing on compute, expense growth and operating income, she offered no update. Investors will have to wait for the Q4 call to learn whether the commitment to grow operating income still holds as Muse serving costs rise. Li also described MSL investments as taking "longer to generate returns than those in our core business," a notable qualification.
Reality Labs: Pacing Back VR to Fund AI, With Losses Targeted at About $15 Billion in 2027
Reality Labs revenue fell 6% to $443 million on lower Quest sales, partly offset by strong AI glasses growth. Meta now expects 2026 Reality Labs losses of $18 billion to $19 billion and around $15 billion in 2027, trending toward profitability afterward. The improvement comes from a changed approach: Meta will ship new VR devices only when there are major breakthroughs, and it is reallocating investment toward AI. Zuckerberg said the glasses are priced "not to make a large profit on the hardware, but to build a large software and superintelligence business on top of them over time." Muse arrives on Meta glasses next month, and management expects paid subscriptions to rise. Muse Charm, a new hardware device, is also about to ship.
Capital Structure: A Nonvoting Share Class Creates Funding Flexibility
Meta announced a new class of nonvoting capital stock, distributed as a one-for-one stock dividend and effectively a 2-for-1 split, with trading expected to begin in late January 2027. Li said it is intended as a publicly traded currency for capital raising, employee compensation and acquisitions while preserving the voting power of existing Class A and B holders and the founder-led structure. She said Meta has no specific financing plans to announce, but the timing is notable given negative free cash flow, capex of up to $145 billion and a balance sheet that is now net debt. Equity issuance is a new option on the table, and investors should treat it as one given the scale of the build-out.