--- title: "Meta (Trans): AI LT vision is not productivity tools, but broad empowerment of everyday users." type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/40322034.md" description: "The following is a transcript of the FY26Q1 earnings call for $Meta Platforms(META.US), compiled by Dolphin Research. For our earnings take, please see '《单腿走路...》'." datetime: "2026-04-30T11:26:34.000Z" locales: - [en](https://longbridge.com/en/topics/40322034.md) - [zh-CN](https://longbridge.com/zh-CN/topics/40322034.md) - [zh-HK](https://longbridge.com/zh-HK/topics/40322034.md) author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)" --- # Meta (Trans): AI LT vision is not productivity tools, but broad empowerment of everyday users. **Dolphin Research compiled the following** $Meta Platforms(META.US) **FY26 Q1 earnings call Trans. For the earnings recap, see** [**One-Legged Walk Limits Meta AI’s Near-Term Upside**](https://longbridge.cn/topics/40321449?channel=SH000001&invite-code=355628&app_id=longbridge&utm_source=longbridge_app_share&locale=zh-CN&share_track_id=980a8167-bdc0-4681-8555-900ad96354b3)**.** **I. Key Takeaways** 1\. **Q2 guide and full-year outlook**: Q2 2026 revenue guided to $58.0–61.0bn (FX tailwind ~2%). Full-year expense outlook unchanged at $162–169bn; OP expected to exceed 2025 levels. 2\. **CapEx raised**: 2026 CapEx lifted to $125–145bn (prior $115–135bn), driven by higher component costs (esp. memory) and data center spend. Contractual commitments rose $107bn this quarter (multi-year cloud and infra procurement agreements). 3\. **Q1 core results**: Revenue $56.3bn, +33% YoY (+29% cFX). OP $22.9bn with OPM of 41%. Adj. net income $18.7bn and EPS $7.31, excluding an $8.03bn one-time tax benefit. 4\. **CF and balance sheet**: Q1 FCF $12.4bn. Cash and marketable securities $81.2bn at quarter-end; debt $58.7bn. CapEx $19.8bn. 5\. **Headcount and tax**: Plans to reduce headcount in May. Q1-end headcount ~77,900, down 1% QoQ. Remaining quarters’ tax rate expected at 13%–16%. **II. Call Details** **2.1 Management Commentary** 1\. **AI models and Meta AI** a. Launched the Muse family and first model MuSpark via Meta Super Intelligence Labs (MSL), going from team formation to a strong model in just 10 months. b. Meta AI, powered by MuSpark, is now a world-class assistant, leading in visual understanding, health, shopping, social content, local services, and game creation. c. Usage of Meta AI surged, with double-digit growth in sessions per user. The Meta AI app remains near the top of app store rankings. d. More advanced models are in training as we climb the scaling ladder. 2\. **Personal and business agents** a. Vision: not just an assistant, but a 24/7 personal agent that understands user goals and helps execute them. In parallel, building business agents to help owners reach and serve customers. b. Personal and business agents will form an ecosystem, expected to drive entrepreneurship at scale. c. Early business AI tests show weekly conversations up from 1mn at the start of the year to 10mn+. Expanded to WhatsApp SMBs in LatAm and Indonesia, and to Messenger in APAC. d. Zuckerberg’s AI philosophy: amplify human capability rather than replace people, empowering individuals to pursue their goals. 3\. **Content recommendation systems** a. Instagram ranking gains lifted Reels time spent by 10%. Facebook global video time rose over 8%, the largest QoQ increase in four years. b. In the U.S. and Canada, Facebook video watch time rose 9% in Q1. c. Training data upgrades: doubled Instagram user interaction sequence length and enriched interaction descriptions. d. Same-day posts now account for 30%+ of recommended Reels (2x YoY). AI-translated/dubbed videos reach 500mn+ weekly viewers each on FB and IG. e. Next steps: continue to scale model size and complexity, incorporate LLMs for deeper content understanding. Building next-gen foundation models for recommendations, with this year focused on validating architecture and techniques. 4\. **Ads** a. Q1 ad impressions +19% with healthy growth across regions; global Avg. price per ad +12%. b. Lattice modeling and GEM architecture upgrades drove 6%+ lift in landing-page conversion rates. c. Adaptive ranking models (LLM-scale, 1T parameters) extended to offsite conversions, lifting conversion on core FB and IG surfaces by 1.6%. d. Meta AI business assistant rolled out to all eligible advertisers, improving resolution of common account issues by 20%. e. Launched Meta ads AI connectors (open beta) this week, enabling advertisers to connect ad accounts directly to AI agents. f. 8mn+ advertisers used at least one GenAI creative tool; early video generation tests show 3%+ lift in conversion. g. Value Optimization Suite ARR surpassed $20bn, doubling YoY. Partner ads revenue ARR reached $10bn, also doubling YoY. h. Threads ads expanded to more markets. Hundreds of millions view WhatsApp Status ads daily. 5\. **AI glasses and hardware** a. Daily active users of AI glasses tripled YoY, making it one of the fastest-growing consumer electronics categories. b. Launched Ray-Ban Meta optical glasses in Q1 (all-day wear design). Following Oakley, more brand partnerships and styles are coming. c. Consumers show strong interest in display-enabled Meta Ray-Bans and Meta Neuralbands. d. Reality Labs Q1 revenue $402mn, -2% YoY, with AI glasses growth partially offsetting lower Quest units. e. Remain the largest investor in VR, but sharpening focus on sustainable VR while shifting more resources to AI and glasses. 6\. **Infra and compute** a. Custom in-house chips (with Broadcom) deployed at over 1GW, alongside significant AMD deployments and the latest NVIDIA systems. b. Meta Compute aims to lead in build efficiency and create a durable strategic edge. c. Signed multi-year cloud agreements rolling out in 2026–2027. d. Teams have consistently underestimated compute needs, and compute will be even more central. 7\. **Other** a. AI is changing work: small teams can do in a week what once took dozens of people months, reorganizing around high-output individuals. b. FoA other revenue was $885mn, +74% YoY, driven by WhatsApp paid messaging and subscriptions. c. In the creator economy, rolled out affiliate partnerships on Facebook to more test partners last month, enabling creators to tag retailer products and earn commissions. Similar tests are underway on Instagram. **2.2 Q&A** **Q: With large-scale AI investments, what KPIs ensure ROI?** A: I focus on whether we are building leading models and leading products. Our formula has always been: first build experiences that reach billions, then lean into monetization once at scale. We are in the upfront investment phase, building leading models and turning them into leading products, which we believe will define the next decade. Specifically, I track three milestones. First, technical quality sufficient to power great products. Second, product scale-up. Third, monetization and efficiency gains toward higher profitability.We do not plan monthly growth with precision, but we have a clear view of the shape of progress. Based on model quality, product usage, and usage trends for frontier models elsewhere, I am very confident we can build a top lab globally, and MuSpark is a very high-quality model powering a world-class Meta AI assistant. In the coming quarters, we will track the next training runs, product expansion, and gradually ratchet up monetization. **Q: How is MSL splitting focus between model training and product launches?** A: The roadmap has been consistent. Research focuses on scaling increasingly capable models and developing capabilities for our emphasis on business and personal agents. We just released the first model, and more advanced ones are in training; this is an ongoing loop, not something we will finish quickly.The product teams are now truly unlocked by having a powerful model. We used to prototype on older models or third-party APIs; now we can build on our own and scale. These tracks will iterate in tandem: keep improving intelligence, keep building and expanding products. Once we reach product-market fit, we will focus more on building the business around them and lowering costs. That is how we have run the company for 20 years, and it remains the plan. **Q: How to think about 2027 CapEx?** A: We are not providing specific 2027 CapEx outlook. Candidly, we are going through a highly dynamic planning process, evaluating compute needs over the next several years. Our experience so far is that we continue to underestimate our compute needs even as we ramp capacity materially, because AI keeps advancing and teams keep surfacing attractive new projects and internal use cases.We expect compute to become even more central, determining model quality, product scope, and organizational productivity. We will flexibly expand infra, and if demand ends up lower than expected, we can slow bring-up or reduce future-year spend. **Q: How do you see agentic compute opportunities across consumers and businesses?** A: Near-term priorities include deepening engagement in our communities, making ads more personalized and efficient, and helping SMBs find and serve customers on our platforms. These are intuitive and closely tied to the current biz.As agentic capabilities mature, agents will boost individual productivity and serve businesses, with personal and business agents interacting to form a vibrant ecosystem. That is the outcome we aim for. The latter opportunity is on a longer timeline. We are focused now on a personal super-intelligence, a consumer agent that works for you and gets things done. While it is a consumer experience today, we see clear monetization paths, such as commissions or premium subscriptions.On the business side, weekly business-AI conversations grew from 1mn to 10mn+, and we will continue global rollout in Q2. It is free for most businesses today, but we will develop durable monetization over time. **Q: What constrains expanding AI glasses across more Essilor-Luxottica brands, and what defines success in 2026?** A: AI glasses continued strong growth in Q1. Demand is robust across an expanded lineup, and we see sales shifting from prior-gen Ray-Ban Meta to the latest generation, showing features like longer battery life and higher-res video are valued.Consumers are also keen on display-enabled Meta Ray-Bans and Meta Neuralbands, the next steps in the product evolution. We remain constructive and will keep investing here. **Q: Are May layoffs driven more by AI-driven efficiency or by staying lean? How will headcount trend vs. revenue?** A: Frankly, we do not know the optimal size for the company yet. AI capabilities are advancing quickly and changing the calculus. We are very focused on using AI tools to drive big productivity gains, and engineering output is already accelerating.Our bias is to use these tools to build more products and services than before. At the same time, we are investing heavily in infra and remain very focused on operating efficiency. We will keep evaluating org design to best support priorities over the next few years. **Q: What does the Muse Spark model unlock for products, and what is the cadence over the next nine months for consumer and biz launches?** A: The field is moving extremely fast. I am proud we went from forming the lab to releasing a widely recognized strong model faster than anyone, validating team execution, infra operations, and our broader plan.On timing, it is hard to give precise dates. We do not want to disclose competitively sensitive details, and on the research side we prioritize quality over hitting specific dates. Research involves trying new things; you cannot predict breakthroughs precisely.Product is similar. There are many agents on the market, but I want a quality bar high enough that I would recommend it to my mom, which trumps any launch date. That said, the team is making tangible progress daily, not just quarterly, which is what makes AI development exciting—small teams can move very fast. The first releases of Muse Spark and Meta AI show we are on track, and the picture should be much clearer over the next few quarters. **Q: How do emerging consumer AI apps like OpenClaw affect Meta AI and the overall agentic strategy?** A: OpenClaw and other agents showcase exciting possibilities. But today they are still quite rough—you need local compute, a terminal, and lots of config, which only hundreds of thousands to a few million people can manage. We are talking about delivering personal super-intelligence to billions globally.So we are focused on a more refined, easier-to-use version with infra ready out of the box. If we can deliver something far better than current systems yet simple enough for everyone, we jump from hundreds of thousands or millions to billions of reachable users. That has been the top goal since day one of the lab.On the business side, the same applies. Many people want to build—websites, products, growth—and good agents can help. Today we help people achieve big goals: staying connected, learning about the world, and entertainment. But that is not all people care about.I want products that understand your specific goals, execute on them, and only come back for confirmation when needed. Whether personal or business goals, everyone will want some version of this service, and there is a natural upgrade path—more use and more ambition justify paying for premium or higher-compute tiers. **Q: Will MSL stay consumer-focused, or also pursue code/recursive self-improvement?** A: We have two primary goals. First, the agent vision—delivering capabilities to users. Second, self-improvement, which is critical: you cannot build leading AI products without leading models, and future-leading models must self-improve.Models still learn from humans today, but at some point they must improve themselves; without that, you are not a leading lab. So self-improvement is a foundational focus. Does that mean becoming a developer tools company? Not necessarily. We are not against APIs or coding tools, but they are not the core focus. I also think people over-associate coding with self-improvement—coding is one element, not the only one. We focus on all ingredients needed for self-improvement, ultimately in service of the personal super-intelligence vision. **Q: When will personal agent products be visible—near, mid, or long term?** A: There will be near-term versions of agent workflows, but the bigger upside comes from continually improving model intelligence and capabilities. The industry trend is clear: each generation gains more capabilities, and people internalize them to gain superpowers.I see agents as the product vehicle to deliver those capabilities to users. This year is pivotal for establishing agents as the core interface for AI, but model progress will continue for a long time. So there is heavy lifting in the near, mid, and long term. **Q: How do ranking changes like doubling interaction sequence length drive usage? How much runway remains in recommendations?** A: We see significant runway through the rest of the year and expect continued engagement gains on Facebook and Instagram. Several areas stand out.First, keep improving data infra to train on more data. We are enriching descriptions of past user interactions and scaling model complexity to leverage larger datasets, including longer interaction histories, all of which improve recommendation quality. Second, make recs more personalized and relevant. We are redesigning retrieval systems to show content matching the full spectrum of user interests and tuning topic diversity based on interest concentration—more tightly focused content for narrow interests, and broader topics for wide interests.Third, keep improving LLM-based tuning so users can express finer-grained preferences in natural language. Sequence length was just one of many Q1 improvements, and there is a large roadmap ahead. **Q: Moving from smaller models to Spark and future LLMs, what unlocks in ads?** A: The ads ranking architecture is advancing. Historically, we avoided using GEM-scale models for inference due to cost at that size and complexity, opting instead to distill knowledge into lighter runtime models.Inference is bound by strict latency—you need to find the right ad in milliseconds—which has constrained us from scaling inference model size and complexity. In H2 last year, we introduced adaptive ranking that leverages LLM-scale complexity (1T parameters).We made architectural progress and co-designed systems with underlying chips to keep sub-second speed while serving large-scale ad delivery. We also built intelligent routing—sending requests to more compute-heavy inference paths when conversion probability is high—improving both performance and inference ROI. Much work is underway before we bring even more LLM workloads into core ad ranking. **Q: Any update on the Manus acquisition?** A: We are still working through details and have no update. **Q: What are the shopping and commerce opportunities for Muse Spark, and any lessons from the 2021–22 e-commerce push?** A: AI agents perform best when you optimize the entire stack. That is why we believe we must build frontier models, agent products, and the infra to support them.Our investment thesis is essentially a bet that the things individuals care about will matter even more in the future. Much industry commentary imagines a single centralized system doing all productive work for society. That is not our worldview. We believe society advances through individuals pursuing their own goals. Some aim to cure disease; many care about finding the right thing for their child. While empowering individuals and building consumer products may sound obvious, our execution differs meaningfully from others. Shopping is a concrete, commercially relevant example—consumers will love it, yet I do not hear other labs talking about building an AI truly great at shopping.Not because shopping is the most important thing, but because enabling people to do what matters—local services, social understanding, shopping, personal health, visual perception (critical for glasses)—are all parts of the personal super-intelligence vision. When assessing Meta’s long-term investment value, start from these values: what do we want AI to do in society? If the answer is to empower individuals and build AI that serves personal goals, that is what we are building—and I believe it will be highly valuable. **Q: Core ads continue to grow at 2x industry. How visible is the growth trajectory?** A: We provided a Q2 guide that embeds a range of macro scenarios and our ongoing work to drive app usage/engagement and improve ad performance.At a higher level, the ads roadmap trajectory remains strong. I have been in this field a long time, and the team’s ability to push the frontier consistently impresses me. Our planning process is now very rigorous—the ROI-based budgeting I have mentioned on prior calls ensures we fund all ads initiatives expected to drive future growth.The process is both mature and reliable, and our measurement of its impact is robust. It has been a key driver of ads revenue growth and proved out again in the latest budget cycle. Based on today’s visibility, we are optimistic about the opportunity set ahead. **Risk Disclosure & Disclaimer:**[**Dolphin Research disclaimer and general disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [AMD.US](https://longbridge.com/en/quote/AMD.US.md) - [AVGO.US](https://longbridge.com/en/quote/AVGO.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [META.US](https://longbridge.com/en/quote/META.US.md) ## Comments (2) - **闲庭信步纳斯达 · 2026-05-03T07:16:13.000Z**: Now AI can even start its own company. Things have really changed a lot. - **新用戶Tiffany · 2026-05-02T04:51:35.000Z**: Assuming I invest 5000 yuan monthly in 03466 and 3145 for long-term dividend income, what would the dividend yield of this portfolio be like?