Meta Platforms Q2 FY26 Earnings Call | Full Recording | META
Alfa finder USA
Call Opens With Standard Formalities 6:40
The operator welcomes participants to Meta's second quarter 2026 earnings conference call, and Chad Heaton, Meta's vice president of finance, introduces the call. He notes that Mark Zuckerberg and Susan Lee will present results, and reminds listeners that the discussion will include forward-looking statements subject to risks described in the company's SEC filings, along with both GAAP and non-GAAP financial measures available on the investor website.
Zuckerberg Highlights Community Milestones 8:31
Mark Zuckerberg opens by reporting that 3.6 billion people now use at least one Meta app daily. Instagram has reached two billion daily actives, Threads has crossed 500 million monthly actives, and Facebook has held above two billion daily actives for some time. WhatsApp hit an all-time record of 30 million messages sent per second during the World Cup final, and Kunal Shah, who built one of India's major payment companies, has joined as the new head of WhatsApp. Zuckerberg frames the quarter around three big opportunities: AI accelerating the core business, new personal agents becoming a foundation for future products, and a growing enterprise business selling APIs, business agents, and compute.
AI Reshapes Recommendations and Ads 10:00
Zuckerberg describes integrating large language models, or LLMs, into recommendation systems so they understand content and user intent more deeply, making feeds more relevant. New Muse image and video models will let people discover a nearly infinite range of personalized content beyond what friends or followed creators post. On the ads side, LLMs have expanded the context used to judge an ad's relevance, and Meta's ad business is now growing faster year-over-year, on a dollar basis, than any other reported ad business. Nine million small businesses are already using at least one AI ad creative tool, and Meta 1, a new subscription offering with AI features, has just launched.
New Apps and Meta Superintelligence Labs 12:31
Zuckerberg mentions recent launches including Instagram Instance, Forum, a standalone groups app, and Seller, a standalone marketplace app, saying AI is making it easier to ship new products quickly. He also updates on Meta Superintelligence Labs, active a little over a year, noting the recent release of Muse Spark 1.1 and Muse Image. Since Meta AI was rebuilt around Muse Spark, daily interactions with the assistant have risen 60%. Muse Spark 1.1 is described as a strong, efficient agentic coding model good at computer use and tool use, now available through a public API.
Personal Agents and Business Agents 13:31
Zuckerberg says Meta is working toward personal agents that can operate around the clock to help with goals like health, relationships, and finances, and that messaging apps like WhatsApp will be central to this, with a new incognito mode launched this quarter for private conversations. On the business side, Meta business agents are now live globally on WhatsApp and Messenger, with over one million businesses using them weekly, and rollout is beginning on Instagram. Zuckerberg describes ambitions for these agents to eventually summarize customer conversations, offer competitive insights, and evolve into a broader "business in a box" service, monetized through subscriptions, volume pricing, and eventually auction-based models similar to advertising.
Infrastructure Investment and Hardware Push 16:00
Meta announced a new strategic venture with BlackRock to build a 1 gigawatt data center in El Paso, Texas, part of its broader compute buildout. Zuckerberg expects much of this compute to support model training and new products, while also serving a growing business selling compute and tools to large customers at a premium. On hardware, Meta's glasses, including a new line made with EssilorLuxottica and a style designed with Kylie Jenner, are described as the first to ship with Muse Spark built in, with early sales exceeding expectations. More glasses news is promised at the Connect Conference on September 23rd. Zuckerberg closes his remarks referencing an op-ed on distributing superintelligence widely rather than centralizing it, framing this as both a values statement and a business strategy.
Susan Lee Reviews Financial Results 19:32
Susan Lee reports Q2 total family of apps revenue of $60.44 billion, up 28% year-over-year, with ad revenue at $59.4 billion, up 27%. Ad impressions rose 14% and average price per ad rose 12%, aided by performance gains and currency tailwinds but offset by growth in lower-monetizing surfaces. Other revenue, driven by WhatsApp paid messaging and subscriptions, topped $1 billion for the first time, up 73%. Reality Labs revenue was $431 million, up 16% on AI glasses strength despite softer Quest headset sales. Total revenue reached $60.8 billion, up 28%, while total expenses rose 55% to $42 billion, including $2.4 billion in legal charges and $1.2 billion in severance tied to a May 2026 headcount reduction that affected about 8,000 employees. Operating income was $18.8 billion, an 8% decline year-over-year, though it would have risen 9% excluding one-time charges. Net income was $15.88 billion, or $6.18 per share, with capital expenditures of $31.1 billion and free cash flow of $784 million.
Engagement Gains From Better Ranking 23:30
Lee explains that revenue depends on both engagement and monetization efficiency. Instagram global time spent grew double digits year-over-year, and Facebook video time spent rose 9% globally and over 10% in the US and Canada, driven by ranking improvements. LLMs are increasingly used to understand content and improve engineering workflows, with every public Reels and feed post on Instagram now processed through an LLM for topic and tone analysis. A major Reels ranking release drove a 15 basis point increase in sessions, and over half of recommended Instagram feed content is now less than a day old, more than double the share from a year earlier. New tools like the "your algo" page and Facebook's "shape your feed" let users directly tune recommendations, with the latter seeing over 80% retention among engaged users.
Monetization Efficiency and Ad Tools 27:02
Lee describes work to place ads at optimal moments, including full ad rollout on Threads and expanded ad support on WhatsApp. A new system called meta generative recommender uses LLMs to jointly reason about ad content and user preferences, contributing to an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook, alongside a 1% rise in app event conversions on Instagram from early pilots. Advantage Plus, Meta's AI-powered end-to-end ad solution, now runs at over $75 billion in annual revenue run rate; an Indian apparel brand using it saw a 13% lift in purchases and 16% increase in cart conversions. Generative ad creative tools are used by over 9 million small businesses, with image generation adoption more than doubling this quarter. The new Meta Business Agent Platform gives enterprises tools to build agents on WhatsApp, illustrated by Mova, a Brazilian rental car company, which saw a 44% increase in daily WhatsApp bookings, with 85% of conversations resolved entirely by AI.
Subscriptions, API Access, and Capacity Strategy 31:30
Lee highlights Meta 1 as an evolution of the subscription portfolio offering more AI tools for users, businesses, and creators, alongside a newly launched, competitively priced high-intelligence model API. Muse Spark is now available on OpenRouter for US developers, with broader distribution, more countries, and enterprise access planned soon. On infrastructure, Lee says the industry has historically underbuilt for AI demand, making existing capacity, including Meta's own, highly valuable, and that current planning is focused on maximizing capacity through 2026 and 2027. She notes confidence that added capacity will prove valuable for scaling experiences, much as it did for Reels, and that Meta's distribution advantages should let it serve AI products valuably regardless of whether its models are strictly at the frontier, though frontier capability is still expected to unlock new markets.
Capacity strategy and infrastructure flexibility 34:01
Meta's longer-term capacity strategy aims to keep growth options open into 2028 and beyond by building data center and network foundations now, while leaving actual server purchase decisions for later. Because these physical assets last a long time, they give Meta room to adjust investment pace as AI adoption becomes clearer. Investments in custom internal silicon are meant to add long-term strategic flexibility and supply chain leverage, improving returns on these commitments. Management believes industry-wide compute capacity will stay tight for the foreseeable future, and sees its own models, consumer products, and enterprise offerings as the highest-return use of that infrastructure.
Financial outlook and guidance 36:00
Meta expects third quarter 2026 total revenue of 61 to 64 billion dollars, with foreign currency acting as roughly a 1 percent headwind to year-over-year growth. Full year 2026 total expenses are now guided to 165 to 169 billion dollars, raised at the low end to reflect a 2.4 billion dollar legal charge recognized in the second quarter. Operating income for 2026 is still expected to exceed 2025 levels. Capital expenditures, including finance lease principal payments, are now expected between 130 and 145 billion dollars, narrowed upward from the prior 125 to 145 billion range. The expected tax rate for the rest of 2026 has moved up to 15 to 17 percent from 13 to 16 percent. The company also flagged ongoing legal and regulatory scrutiny, including several youth-related trials scheduled in the US this year that could result in a material loss.
Financing the AI buildout 40:32
Asked about financing this multi-year infrastructure buildout, Susan explained that strong operating cash flow gives Meta a position of strength, and the company has been shifting its capital structure toward more debt to lower its overall cost of capital. She noted a preference for cost-efficient, long-duration financing that matches the long time horizons of AI infrastructure projects, and pointed to the newly announced partnership with BlackRock as an example of broadening the range of capital sources available. On 2027 capex specifically, she said no formal outlook is being given yet, since infrastructure planning remains highly dynamic and current plans are focused on maximizing capacity through 2026 and 2027 while preserving flexibility for decisions further out. She added that Meta generally sees near-term capacity as more valuable than long-term capacity.
Scaling compute across product lines 40:32
Mark was asked which of Meta's many opportunities, consumer agents, business agents, APIs, developer tools, or direct compute sales, might scale first to show clear returns. He said a substantial share of compute goes toward training frontier models, with the rest split across improving the core business, new consumer products, APIs, business agents, developer tools, and selling compute directly at a premium. He expects meaningful growth across essentially all of these areas rather than one clear leader, while noting that selling intelligence built on top of compute should carry a significantly higher margin than selling raw compute.
Enterprise opportunity and go-to-market 41:30
On the enterprise opportunity, Mark described it as partly an extension of Meta's existing advertiser and small-business relationships, now expanding into business agents across messaging apps that get paid based on delivered results, much like the ad auction system. He also pointed to a newer area, coding, productivity, and developer tools originally built for Meta's own internal use, which the company now sees as serving external businesses too, describing this as a different muscle the company is still building. He framed the enterprise opportunity as the sum of compute sales, API services, productivity tools, and business agents together, calling it a very large combined opportunity that Meta is now focused on developing.
Personal AI agents as a coming market 45:31
Responding to a question about whether AI has really broken through for everyday consumers, Mark said some categories already have, pointing to AI assistants and, over the past year, coding agents as the first real agentic market, helped along by technical users and coding's closed, digital nature. He argued that personal consumer agents, tools that understand a person's goals and work on their behalf continuously across health, hobbies, finances, productivity, home life, relationships, and career, will become an extremely large market within the next several years. He noted that consumer products need to simply work for billions of people rather than require fiddling like today's proto-agents, and said Meta's strength at scaling working products to huge audiences positions it well, though he stopped short of detailing unreleased products.
Recommendation systems and compute allocation 50:00
Susan described continued headroom to improve Facebook and Instagram recommendations through the rest of 2026 and into 2027, driven by more personalized models, deeper LLM-based content understanding, richer training data describing past user interactions, longer interaction sequences, and more complex model architectures. She also highlighted growth in LLM-based ranking agents that make engineers more productive. On the question of selling compute externally while also buying it from third parties, Mark said demand for compute far exceeds supply, so Meta balances near-term monetization offers against building intelligence on top of that same compute, which compounds its value over time, rather than treating it as a simple either-or choice.
AI lab progress and culture 55:33
Asked about the AI lab's performance about a year after new leadership joined, Mark said he is happy with its trajectory, citing early scaling-ladder model releases as impressive while larger, more advanced models are in progress. He stressed that beyond raw intelligence, the data and feedback flywheel from real usage is critical to serving people well, and that Meta's demonstrated ability to scale working products to billions of users gives it durable advantages in personal agents and business agents alike. He also emphasized building a research culture with low drama and consistent management as part of compounding progress over time.
Model scale, cost, and open source 59:00
Discussing Muse Spark 1 and 1.1, Mark explained that training progresses by scaling up from smaller to larger models, with each stage revealing new behaviors, and said he feels good about these models given their scale and the lab's current stage of development. He noted Meta wants both more advanced large-scale models and efficient, lower-cost models suitable for serving billions of consumers at scale, and the conversation touched on renewed interest in open source as part of the lab's broader strategy alongside monetization plans.
Balancing Open and Closed Models 1:01:00
Mark explains that Meta continues to value efficiency alongside solving the hardest problems for businesses, which means both smaller efficient models and frontier models matter. He reaffirms that open source has always been part of Meta's strategy because it benefits the wider ecosystem and creates positive feedback loops, drawing the community into Meta's infrastructure and contributing improvements. He notes that building open source models actually takes more work than closed ones, since an open model needs to be well-rounded for many uses rather than tailored to one company's needs. He says Meta Superintelligence Labs was given room to build the most intelligent models possible, and expects Meta will return to releasing open source models again soon, while continuing a mix of open and closed approaches.
Why Meta Builds Its Own Frontier Models 1:03:00
Asked whether the rise of open weight models reduces the need for Meta to build its own frontier systems, Mark says no, since open models are not yet as strong as frontier ones, and relying on other companies' choices carries real risk. He describes Meta as a full stack technology company that has always built its own data centers, infrastructure, and chips, which lets it create more personalized and efficient experiences than others can. He argues that having control over model building will matter for businesses that want trust and data control, and that different models, like people, have different strengths, so building models tailored to Meta's own use cases, like personal superintelligence or small business agents, is a lasting advantage.
Capacity Plans Reflect Demand, Not Just Supply 1:08:31
Susan clarifies that Meta's focus on 2026 and 2027 capacity reflects being demand constrained, with many ROI positive uses for compute even within the core business, alongside uncertainty about long-term supply chain limits. Planning for 2028 and beyond centers on flexibility, securing land and power without locking in major chip purchases too early, since internal demand and technology will keep evolving.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
