META (Follow-up call Trans): Most capex translates into post-2027 capacity

Dolphin Research compiled the key takeaways from the post-earnings analyst huddle with $Meta Platforms(META.US). The discussion was a Q&A.

Q: The European Commission views Meta's paid option as non-compliant. Are there new EU compliance standards for product updates, launch timing, pricing, etc.? Also, how should we think about revenue and cost impacts into 2026, and potential mitigations such as reducing personalized ads?

A: In Dec 2025, we reached agreement with the Commission on further changes to consent for personalized ads in Europe. Over the next few days, European users will see changes in how personalized ads are presented. We continue to believe our solution goes well beyond what the DMA requires. We will keep challenging the Commission's decision in court, while acknowledging regulators or courts could still require further adjustments to less-personalized ad services.

As a result, if that happens, user experience in the EEA — and especially in Switzerland — could materially worsen. We are engaging with regulators and do not yet have a timeline for a final resolution. We have baked the impact of the new UI rolling out this quarter into our outlook. The topic is reflected in our long-range plan.

Q: On revenue, can you frame expectations for 2026? If OP is roughly flat this year vs. last and revenue grows 20% in 2026, that would imply OPM of 13–30% in 2026, vs. 41% in 2025.

A: We do not manage to a margin target. That said, we think we are well positioned: through 2026 we expect strong revenue growth and a healthy overall setup. We still see revenue compounding at a rapid pace. We are reinvesting most of our revenue back into priority areas — primarily AI infrastructure and talent, as noted on the call. Both the ramp of infra and the pace of hiring into 2026 are variable, including when planned capacity goes live and how fast we can secure new capacity.

This is an inherently dynamic planning process. Our approach is to gradually expand consolidated operating profit, recognizing there will be volatility in the path. Over the long run, margin expansion alone will not make us compelling vs. other equity opportunities.

Historically we have operated above that bar, and we believe we can do so over time. But it will not be linear, as significant investment opportunities can create periods of pressure.

Q: How much of 2026 growth can carry into 2027, and for how long, especially considering innovations funded by this year's capex and opex?

A: The macro backdrop has been very favorable heading into 2026. As mentioned on the call, holiday performance was exceptionally strong, and that momentum has continued into the first weeks of 2026. Investments we made in 2025 — mostly into ads performance and engagement — have worked well.

We will make similar investments in 2026, but it's hard to precisely place each initiative on the overall return curve. We can estimate ROI for individual programs with reasonable confidence, yet uncertainty grows when you sum them up. Based on 2025, aggregate returns were solid. So we think the upper end of 2026 outcomes looks attractive, with steeper benefit realization than we initially modeled. Still, macro can be volatile and there are many variables.

We feel good about the business fundamentals and the work underway, but uncertainty remains into 2026. And it only increases as we look into 2027.

Q: On Reality Labs losses: you said 2026 losses should be roughly flat vs. 2025, then start to narrow in 2027. What drives that change — a shift toward wearables, faster revenue growth, or something else?

A: We remain optimistic on VR and will keep investing, primarily in future headsets. Consumer adoption of VR has been slower than wearables overall, so we are rebalancing Reality Labs to reflect that.

This year we have significantly reduced investment in VR and Horizon, while leaning more into wearables to capitalize on our leadership position. That is the broad direction of the portfolio shift.

From 2026 onward, we expect Reality Labs operating losses to trend lower, supported by more mature supply chains as wearables scale within the VR ecosystem, a higher mix of higher-margin businesses, and ongoing efficiency gains over time. That is our base case, but roadmaps can change and markets may evolve differently, leading to non-linear outcomes. We therefore expect losses to decline off 2026 levels, but the exact slope is hard to predict.

Q: On advertisers' cost per action. I know it's tricky to measure, but can you share the latest trend?

A: Broadly, we continue to improve ads performance. Our key metric is conversion rate, though it's nuanced since advertisers optimize to different conversion values — from lower-value goals like views to higher-value goals like sales. We control for that by tracking a value-weighted conversion rate. We saw a strong YoY increase that outpaced impression growth, and the YoY uplift accelerated into Q4, driven by our performance optimization gains.

We also look at cost per action (CPA). It is complex for similar reasons — goals differ, so CPA varies. Adjusting for mix, we observed a steady decline in advertisers' CPA in Q4.

Q: On 'Meta Compute': with tens of GW expected over the next decade, is capacity growth tracking roughly with what we currently envision?

A: 'Meta Compute' reflects our view that, given model scale trends, we will need substantially more compute. The precise level is difficult to forecast.

This is a highly variable planning domain, even just looking to 2026, let alone beyond. Our focus is to ensure we can scale compute and do so at lower cost than today, building flexibility and optionality at multiple points in the plan.

On reducing compute cost, silicon is a major driver. We are pursuing multiple paths, including optimizing chip design to maximize cost-efficiency across workloads. For diverse workloads with varying compute, memory, and networking needs, we aim to match each with the most suitable chip to maximize perf/W and minimize total cost of ownership. In parallel, we are expanding our MTIA custom silicon program to support ranking and recommendation inference and training.

Silicon development is a heavy lift. Over the long run, we will also look at other areas, such as lowering energy production costs — not a near-term bottleneck, but ultimately dependent on compute scale. Meta Compute is about taking a long view on where our compute needs could face the tightest supply.

Q: On impression growth. What are the main drivers of sustained impression gains and how durable are they?

A: Global impressions grew 18% YoY in Q4, driven by user growth and higher engagement per user. Ad load contributed as well, but to a lesser extent on the YoY delta.

On engagement, video was the biggest driver in Q4, especially on Instagram, where Reels time rose over 30% YoY globally. Facebook video watch time also delivered healthy double-digit growth.

On ad load, we increased load in Facebook and Instagram Feed and Video in Q4, continuing the adjustments from Q3. Net-net, the 300bps step-up in global impression growth was supported by ad load optimization in FB/IG Feed and Video and stronger engagement on IG. Engagement was the primary driver, with ad load secondary.

Q: Mark noted higher employee effectiveness, a shift toward higher-caliber talent, a flatter org, and less work in flight. That seems to imply headcount growth slows, flattens, or declines. Is that right?

If so, does your 2026 opex guide imply other offsets to headcount, such as higher per-capita comp or other costs?

A: As of now, we expect to continue growing headcount into 2026, targeted to priority areas including infrastructure, monetization, and AI. These roles skew toward higher-comp tech positions, which will keep shifting the talent mix. We therefore expect comp to grow faster than headcount in 2026.

On AI's impact on hiring, AI will materially shape our workforce over the next few years, but the exact effect will be situational. If each engineer can deliver more value, we may hire more, not less. The outcome is uncertain. We are focused on broad adoption of productivity tools, finding people who can use them well, and continuously adjusting org design to how people work in a more AI-native environment.

Q: How should we think about reliance on third-party cloud providers — is it a bridge until 'Prometheus' (the 1GW DC slated for 2026) is live, or a durable component of your compute strategy with material annual spend beyond 2026?

A: We plan to keep investing heavily in data center capacity we own and lease, as reflected in capex plans.

But much of that capex builds capacity that comes online in 2027 or later. So we have been signing cloud agreements to bring capacity online faster this year and relieve current tightness. The build-vs.-public cloud decision weighs cost, timing, scalability, and the ability to customize and configure to our needs.

Owning data centers enables more customization and operational efficiency, and greater long-term supply assurance. Cloud has advantages too: when providers have ready inventory, they can activate capacity much faster.

Q: Capex guidance spans a wider-than-usual range. What's driving the uncertainty — DC construction costs, memory and component pricing, or the challenge of forecasting compute needs this early in the year?

A: We are only four weeks into the year. As you recall, last year's infra build had a lot of moving parts, and this year is similar. We are trying to budget with that variability in mind. We are also working to expand capacity to meet potential needs in 2027 and beyond.

Q: Once Prometheus reaches steady state with full utilization, can margins expand further by shifting more to owned capacity? Today we are in a mixed state between leased capacity and self-build DCs.

How do you think about that steady state and the scope to move more demand on-prem? Would that support further margin gains?

A: We do not think about this as a steady state — the environment is dynamic. Our priority is securing enough capacity across time horizons to meet planned needs. In the near term, we will light up cloud capacity where it makes sense. Longer term, owning and operating gives us more flexibility to allocate resources to specific workloads.

That connects to our long-term compute strategy. Our silicon strategy targets the best performance per watt and per dollar, matching each workload with the optimal chip. The two are complementary.

Q: On Agentic Commerce. As people start using agents, will you charge partners, or will these products be free? How do you plan to monetize these agent capabilities?

A: Commerce remains a core focus. We already offer tools to connect businesses with consumers and drive sales via ads — including AI-powered Advantage+ shopping campaigns.

We are investing in new experiences to make product discovery and purchase easier across our apps. Part of your question is how we monetize new AI revenue streams. We will explore subscriptions for high-value AI features, including attributes of Manus, as discussed. We plan to expand its current subscription offering to more businesses, integrate Manus across our products, and bring leading agents to billions of people.

We also see traction in Business Messaging, where our commercial AI in Mexico and the Philippines is gaining interest. We already see over 1 million weekly conversations across people-to-people and people-to-business AI. These are early examples, but we see ample opportunity in commerce, agentic commerce, and new revenue streams under exploration.

Q: On per-engineer productivity. How close is Meta to the ceiling, and how has this changed product development velocity vs. the past?

A: We have seen ~30% higher output per engineer, with power users who go deeper on AI coding tools up even more — about 80% YoY. The big change in H2 2025, especially in Q4, was improved efficiency of agentic coding tools and broader internal adoption as the tools became more capable.

Those percentages refer to code output per engineer. This year and next, we will learn a lot about the efficiency frontier. The tools themselves have room to improve, and our engineers are still early in learning to use them maximally. We are also seeing non-engineering functions — PMs and designers — take on more technical work with these tools.

Looking ahead, we will integrate coding tools more deeply into our dev environment. We will open up more internal resources and related tools, and deploy more customized models trained on our internal codebase.

Q: Reels launched a TV experience in Q4. How big is the opportunity in time spent and ad monetization?

A: IGTV is a new collaboration with select partners and is still early. We are excited to bring creator Reels to the big screen so people can enjoy them together, and early feedback is that co-viewing is more fun.

We are testing to learn what features make the TV experience work best. Future explorations include using the phone as a remote, intuitive channel switching, and sharing with friends. It's early days, and it's too soon to quantify impact on time spent or ads.

Q: Any impact from OBBB on your 2026 capex plans? Does it matter?

A: Given our heavy investment in infrastructure and R&D, we expect to save substantial cash taxes under the new U.S. tax rules, which will be favorable for our 2026 tax profile.

Q: On the Manus acquisition, which assets are uniquely valuable and strategically important?

A: We are excited about Manus on several fronts. First, we believe the deal accelerates development of new technologies that benefit businesses and users globally, while both sides maintain Manus' existing operations.

We also plan to leverage Manus technology to advance our platform efforts, opening new monetization opportunities. We expect it to materially improve our AI tools, and we are excited to welcome a talented team. We see general-purpose agents as a key technology and product to enhance consumer and enterprise experiences.

Risk disclosure and statements:Dolphin Research disclaimer and general disclosure