Fastest Growth Since 2021 Gets Punished as Capex Ceiling Moves Up Again — Sell-Off Hands Long-Term Investors a Better Entry

Results vs. Consensus

Quality of the Beat

• Revenue: Revenue on a constant currency basis would have increased by 29% year-over-year, so roughly 4 points of the headline 33% came from FX — a real-but-temporary tailwind. Importantly, the print landed at the top of the prior $53.5–56.5B guidance range, meaning management called it tightly and execution delivered against the high end.
• Mix of ad growth is the most encouraging signature. Pricing and impressions both up double-digits implies the auction is firming, not just being loaded with more inventory. Susan Li attributed the price-per-ad lift to "ad performance improvements, better macro conditions versus Q1 of last year, and currency tailwinds in international regions" — a defensible cocktail.
• Margins: Operating margin held at 41% despite total costs rising 35% YoY — the operating leverage on a 33% topline absorbed a sharp infrastructure depreciation step-up, which is the more telling signal than the GAAP EPS jump.
• EPS: The headline $10.44 is largely a tax accounting reversal. The $7.31 underlying figure — up 14% YoY — is the right number to mark; it still beat consensus by ~8%, but at substantially less than the GAAP optics suggest.
• DAP miss: Exogenous and identifiable. The sequential pressure was partly attributed to internet disruptions in Iran due to ongoing conflict, as well as a restriction on WhatsApp access in Russia, not engagement deterioration.

Segment Performance

Family of Apps

This is the segment that pays for everything else, and it had the kind of quarter that justifies the whole AI investment cycle. Revenue came in at $56.3 billion, compared with the $55.557 billion FactSet analyst consensus forecast, and the ads engine produced double-digit growth on both volume and price simultaneously — a configuration we haven't seen consistently for over a year. The "Other" revenue line jumped 74% to $885M, almost entirely on WhatsApp paid messaging and subscriptions, validating the company's slow-but-real monetization of WhatsApp as a transactional surface rather than a pure messaging utility.

The mechanism underneath the topline is what matters for the forward thesis. Li was specific: Ad impression volume grew 19% across Meta's services, while the average price per ad increased 12% year-over-year, reflecting gains from ad performance improvements and a more favorable macro environment compared to Q1 2025. The conversion-rate gains the company called out — enhancements in Lattice and GEM architectures delivered a 6%+ conversion rate gain for landing page view ads; expansion of adaptive ranking to off-site conversions drove a 1.6% conversion rate increase on Facebook and Instagram — are the bridge from capex spend to monetization. They're directly measurable, attributable, and recurring.

"On Instagram, the ranking improvements that we made in Q1 drove a 10% lift in Reels time spent. On Facebook, total video time increased more than 8% globally in Q1, the largest quarter-over-quarter gain in four years." — Susan Li, CFO

Assessment: The FoA segment is converting AI infrastructure into measurable ad-stack gains in real time, and the engagement story (Reels time spent +10%, Facebook video +8% — the biggest quarterly gain in four years) is strengthening rather than rolling over. This is the segment that needs to defend a 41% operating margin while capex builds, and Q1 says it can.

Reality Labs

The segment continues to do exactly what it does: lose roughly $4 billion a quarter while the AI glasses business inside it builds quietly. Reality Labs posted $402 million in revenue and a $4.03 billion operating loss. The loss narrowed slightly from $4.21 billion a year ago, so the trajectory is fractionally better — but the segment is still a $16B/year drag on operating income. The bright spot, and it's a real one, is the glasses category: daily users of AI glasses tripled year-over-year, and Meta launched Ray-Ban MetaOptics designed for all-day wear during the quarter.

"Our AI glasses continue to perform well with the number of people using them daily tripling year over year. This continues to be one of the fastest growing categories of consumer electronics ever." — Mark Zuckerberg, CEO

Assessment: Treat RL as a roughly $16B/year R&D line item with optionality on glasses. The loss isn't expanding, the glasses curve is exponential off a small base, and Zuckerberg's framing of glasses as an agent-delivery surface ("I am also really excited to see the glasses evolve from being able to answer questions to being able to be a personal agent that is with you all day long") is increasingly credible.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Confident and unapologetic on strategy, more cautious in framing on returns. Management did not hedge on the size of the capex commitment — and Li explicitly conceded that the company has "continued to underestimate" its compute needs — but they would not be drawn on 2027 capex, on specific monetization timing for personal agents, or on product launch dates beyond Muse Spark. The posture is "we know what we're building, we don't know exactly when each product lands, and we're not going to pretend otherwise" — which reads as honest, not evasive, but it is unmistakably a less quantitative call than this stock's holders would prefer at this stage of the AI cycle.

The Capex Raise — Component Pricing, Not Demand

This is the line item that drove the entire stock reaction, so the framing matters. Management was explicit that the raise is largely on input costs, not on expanded ambition. Management raised its 2026 capital expenditure forecast to a range of $125 billion to $145 billion, up from a prior forecast for $115 billion to $135 billion. Showing how extreme this capital expenditure ramp is, the midpoint of the new range is nearly twice the company's $72 billion in capital expenditures last year. The official rationale: "reflects our expectations for higher component pricing this year and, to a lesser extent, additional data center costs to support future year capacity".

"On that note, we are increasing our infrastructure CapEx forecast for this year. Most of that is due to higher component costs, particularly memory pricing. But every sign that we are seeing in our own work and across the industry gives us confidence in this investment." — Mark Zuckerberg, CEO

The Q1 step-up in contractual commitments was even more striking than the capex guide itself: non-cancelable contractual commitments totaled $237.67 billion, with a $107B single-quarter increase from new multi-year cloud and infrastructure agreements running through 2027.

Assessment: "Memory pricing" is a defensible attribution — DRAM/HBM costs are real and observable — but it doesn't fully explain a $10B midpoint raise. The honest read is that Meta is locking in capacity to remove a future supply constraint, and the choice to confirm a planned long-dated debt issuance alongside this signals management views the spend as funded out of the next 5–10 years of cash generation, not the next 2. That is a different shape of bet than the market priced before the print.

Muse Spark and Meta Superintelligence Labs

The lab is roughly 10 months old and shipped its first frontier model, which is a faster cycle than most of Meta's peers. Zuckerberg framed Muse Spark as validation of the lab build-out, not the finish line. Zuckerberg highlighted that Meta AI engagement accelerated sharply following the rollout of Muse Spark, with double-digit percent increases in Meta AI sessions per user since the broad launch, and the standalone Meta AI app has held near the top of US app store rankings since release.

"This was the first release from Meta Superintelligence Labs, and it shows that our work is on track to build a leading lab... Spark is just one step on that scaling ladder, and we are already training even more advanced models." — Mark Zuckerberg, CEO

The strategic distinction Zuckerberg drew with the rest of the industry was the most pointed editorial of the call: he positioned Meta's AI vision around "personal superintelligence" for individuals, contrasting with what he characterized as the industry's pursuit of centralized productivity systems that "replace people." It's a values pitch, but it also doubles as a competitive moat argument — Meta's distribution surface (3.56B DAP across four apps) is uniquely suited to consumer-AI delivery in a way no competitor can match.

Assessment: The Muse Spark velocity is a genuine signal that the team-rebuild is producing output, but Zuckerberg's reluctance to commit on subsequent launch timing ("we care a lot about just having something I would give to my mother") is a watch item. Investors are paying for a product cadence the company isn't yet willing to set expectations on, and that asymmetry creates ongoing whipsaw risk on every print until a second model lands.

Ad Stack Compounding — The Lattice/GEM/Adaptive Ranking Story

This is where the bull case actually lives. The ad-system upgrades the company called out aren't theoretical — they're producing measurable conversion lifts in the quarter they ship. Lattice and the GEM model architecture delivered a 6%+ conversion-rate gain on landing-page-view ads; the new adaptive ranking model — described by Li as an LLM-scale recommender that routes high-value impressions to more compute-intensive inference — drove a 1.6% conversion uplift after expanding to off-site conversions. Combined with the Meta AI Business Assistant, GenAI creative tools (now in use by 8M+ advertisers), and the value optimization suite (revenue run-rate doubled to $20B+), the ad infrastructure is monetizing the compute build in real time.

"Even in the current landscape of the ads roadmap, we are advancing the architecture to allow us to leverage the abilities of larger models... in the second half of last year, we introduced a new adaptive ranking model, which enables us to leverage LLM-scale model complexity of trillion-parameter class, and we made advances in the model architecture and the system with the underlying silicon so it maintains the sub-second speed required to serve ads at scale." — Susan Li, CFO

Assessment: This is the cleanest "capex begets revenue" story in big-cap tech right now. Every conversion-rate basis point compounds across $220B+ of annual advertising revenue. If Meta can keep producing 1–5% conversion lifts per quarter from successive model deployments — which they have for at least four straight quarters now — the math on AI infrastructure ROIC works even if Muse Spark consumer products stay subscale.

The MTIA Silicon Roadmap with Broadcom

The custom-silicon story is the most underdiscussed positive of the print. The Meta MTIA Broadcom partnership committed to a 1 gigawatt deployment on a 2-nanometer process, scaling to multiple gigawatts by 2027. In-house silicon is typically 30% to 50% cheaper per unit of inference than merchant accelerators at retail. Zuckerberg explicitly framed the silicon program as a strategic-advantage lever: If Meta migrates ad ranking and recommendation inference to MTIA, the marginal cost of every Reels view drops, and that's the route to defending operating margin through the capex peak.

"We are rolling out more than one gigawatt of our own custom silicon that we are developing with Broadcom as well as a significant amount of AMD chips to complement the new NVIDIA systems that we are rolling out as well. One of the primary goals of our Meta compute initiative is to lead the industry in efficiency of building compute, and we expect that will be a strategic advantage over time." — Mark Zuckerberg, CEO

Assessment: MTIA at scale is a 2027 story, while the capex bill is a 2026 reality, so the market won't credit this fully yet. But the structural argument — Meta becoming dramatically less dependent on Nvidia retail pricing on incremental inference capacity — is the single most important offset to the bear case on long-term margin compression.

Personal & Business Agents — The Long Game

Zuckerberg used substantial airtime to outline a two-track agent product strategy: personal agents that "understand your goals and then work day and night to help you achieve them" and business agents for SMBs and entrepreneurs. The early data points are encouraging in scale terms: business AI weekly conversations grew from ~1M at the start of 2026 to 10 million weekly conversations through Business AIs on WhatsApp and Messenger — a 10x in roughly four months, with expansion to Latin America, Indonesia, and APAC during the quarter.

The monetization framing was deliberately patient. Li acknowledged business AIs are currently free for most SMBs and that monetization is a "longer-term" project. On the consumer side, Zuckerberg pointed to "commission structures or a premium offering" as future revenue vehicles without committing on timing.

Assessment: 10M weekly business AI conversations is a real engagement signal, but it's not yet a P&L event. The agent strategy is correctly framed as the long-cycle pay-off on the current capex spend; investors who can't wait 12–24 months for monetization clarity should not be in this stock at this size.

Headcount & the Coming May RIF

Li said in prepared remarks: "we recently shared internally that we plan to reduce the size of our employee base in May." Subsequent press reporting placed the figure around 8,000 roles, with another ~6,000 open requisitions closed. Zuckerberg framed the layoffs as necessary to offset rising infrastructure costs and improve organisational agility, signalling that headcount efficiency will remain a priority as AI spending accelerates. The company ended Q1 with 77,986 employees, down 1% from Q4 and up only 1% YoY.

"We are seeing more and more examples where one or two people are building something in a week that would have previously taken dozens of people months. And I want to make sure that Meta Platforms, Inc. is the best place in the world for these types of people to come and make an impact." — Mark Zuckerberg, CEO

Assessment: The RIF is the offset to keep FY2026 operating income above FY2025 even as depreciation builds. It's also a signal that internal AI productivity gains are real enough that management is willing to bet headcount efficiency on them. The risk is talent retention — Meta is simultaneously firing in some areas and paying premium comp to attract MSL hires, and that dual posture has historically introduced execution noise.

Guidance & Outlook

The Q2 revenue range of $58–61B brackets the consensus $59.5B midpoint precisely, and the FX assumption (~2% tailwind) is two points lighter than Q1 — implying the full-year underlying growth trajectory remains accelerating but lapping a tougher comparison set. The reiteration of the $162–169B expense range, combined with the commitment that operating income will be above 2025, is the strongest signal management gave on the call about margin defense through the capex peak.

The capex raise needs to be unpacked carefully. Per management, the bulk of the $10B midpoint increase is component pricing (memory in particular). The remainder is "additional data center costs to support future year capacity" — i.e., pulling 2027 capacity forward. That distinction matters: the first is a cost inflation hit; the second is a duration extension that doesn't change the steady-state spending envelope.

Implied Q2 growth ramp: The $59.5B midpoint implies ~25% YoY growth at the midpoint (~28% on the high end), a deceleration from 33% in Q1 but consistent with the FX tailwind compression from 4pp to 2pp. Underlying constant-currency growth is implied at ~23–25%, meaning core demand is holding even as the FX kicker fades.

Street at: $59.5B was the pre-print Q2 consensus per LSEG-style estimates; new guide brackets it.

Guidance style: Tight — Meta has historically guided revenue conservatively and landed at or above the high end. The Q1 print at the top of the prior $53.5–56.5B range is consistent with the pattern.

Analyst Q&A Highlights

AI Investment ROIC & Signposts

• Brian Nowak, Morgan Stanley: Pressed for the specific signposts management is watching to ensure ROIC on the AI buildout. Zuckerberg deflected on quantification, framing the metrics as model quality → product scale → monetization in sequence, with no specific dollar or share milestones offered.
  Assessment: The non-quantitative answer is the single biggest tell of the call. Management is not yet willing to commit to a measurable ROIC threshold, which is appropriate this early in the cycle but creates an ongoing transparency overhang.

2027 Capex Trajectory

• Mark Shmulik, Bernstein: Asked whether 2027 capex steps up further given peer commentary. Li was explicit that no 2027 figure would be provided and that internal planning is "very dynamic." Notable concession: Meta has "continued to underestimate" compute needs even as it has ramped capacity.
  Assessment: The honest read is that 2027 capex is likely to step up further unless component pricing materially deflates. Investors should model another raise on the Q4 2026 print as the base case.

Agentic Strategy & SMB Opportunity

• Eric Sheridan, Goldman Sachs: Asked how Meta is thinking about the agentic opportunity across consumers vs. SMBs/enterprises. Li pointed to the 10M weekly business AI conversations (up from 1M at year-start), confirmed business AIs are free today, and outlined "commission structures or a premium offering" as future consumer monetization vehicles.
  Assessment: The SMB agent opportunity is the most credible incremental revenue line in the 2027–2028 model. WhatsApp's installed base in LATAM and APAC is the unique distribution moat.

Headcount Optimization & AI Productivity

• Youssef Squali, Truist: Asked what share of the upcoming May RIF is AI-driven productivity vs. general efficiency, and how headcount scales relative to revenue. Li declined to attribute the cut to specific drivers and framed organizational size as an ongoing optimization with "a bias toward wanting to use these tools to build even more products."
  Assessment: The framing implies further selective hiring in AI/infra alongside cuts elsewhere — a "barbell" headcount strategy that should support gross-margin defense.

Muse Spark Cadence & Product Pipeline

• Justin Post, Bank of America: Asked what cadence of new product launches investors should expect over the next nine months. Zuckerberg explicitly declined to set a calendar, citing both competitive sensitivity and an internal quality bar ("we care a lot about just having something I would give to my mother").
  Assessment: This was the second of three non-quantitative answers on the call. Management is asking for patience on product cadence — a reasonable ask but one that constrains upside on near-term prints.

Lab Direction & Recursive Self-Improvement

• Ross Sandler, Barclays: Asked whether MSL will pursue developer-tools / recursive self-improvement work or stay consumer-focused. Zuckerberg was sharper here, distinguishing self-improvement (table stakes for a leading lab) from a coding-tools commercial product (not a primary focus): "I am not against having an API or coding tools, but it is not our primary focus."
  Assessment: This is a notable strategic clarification. Meta is explicitly not chasing the API/dev-tools revenue pool that several peers are monetizing. The bet is that consumer + business agents will be larger.

Larger Models in Ads & the Manus Deal

• Doug Anmuth, JPMorgan: Asked about the productivity unlock from moving the ad business from smaller models to LLM-scale architectures, and for an update on Manus. Li gave a substantive answer on the LLM-in-ads roadmap (adaptive ranking, trillion-parameter inference at sub-second latency) but explicitly declined to comment on Manus: "we are still working through the details, so we do not have an update right now."
  Assessment: The Manus non-update is notable — the deal was widely reported and the silence implies either negotiations are unresolved or there are integration complications. Watch for clarification at next quarter's print.

Shopping & Growth Visibility

• Ken Gawrelski, Wells Fargo: Asked about commerce/shopping (any application of 2021–22 lessons) and visibility into continued double-pace-of-industry growth. Zuckerberg used shopping as an example of the broader "personal superintelligence" thesis without committing to a specific commerce product. Li pointed to a "very ROI-based" budgeting process and "feel good about the investment opportunities ahead of us" as the visibility answer.
  Assessment: The shopping question was answered philosophically rather than tactically. If there is a near-term commerce surface (marketplace, integrated shopping in Meta AI), management is not telegraphing it yet.

What They're NOT Saying

1. 2027 capex envelope: Explicitly declined to give a range. Combined with Li's admission that Meta has "continued to underestimate" compute needs, the base case is another step-up — investors should not anchor on "$135B is the new normal."
2. Reality Labs operating income outlook: No specific 2026 RL loss guidance was offered. Q1's $4.0B loss annualizes to $16B, but management did not confirm whether 2026 losses will exceed, equal, or be below 2025's $19.2B.
3. Muse Spark 2 / next model timing: No commitment on when the next MSL model lands. Zuckerberg was deliberate that quality, not timing, sets the cadence.
4. Specific advertising contribution from AI-driven targeting: Management quantified individual conversion-rate gains (6%, 1.6%, 3%) but did not put a dollar figure on the cumulative AI contribution to Q1 revenue — leaving consensus to triangulate.
5. Manus deal status: Direct question, no update. Notable.
6. The exact size of the May RIF: Press reporting put it at ~8,000, but management did not confirm a number on the call. The 1% YoY headcount print already reflects some of the optimization.
7. Q2 operating margin guide: No quarterly margin commitment. The full-year "operating income above 2025" framing leaves substantial latitude on quarterly cadence.
8. The 40-year debt issuance scale: Pre-print reporting suggested up to $25B; the call did not address financing structure. Capital structure will be a Q2 topic.
9. EU regulatory crystallization: Li flagged "headwinds in the EU and the U.S." and youth-trial risk that "may ultimately result in a material loss," but gave no scoping on potential P&L impact or settlement framework. This is the most underpriced overhang on the stock.

Market Reaction

• Pre-print setup: META entered the print at ~$669, down from the ~$796 February 2026 high (already ~16% off the peak going in). The Nasdaq was tracking its best month since April 2020 (+14% MTD). Pre-print analyst consensus rating was 96% Buy with an average target near $837–$839. Options markets implied a typical 6–9% earnings move.
• After-hours April 29: Down approximately 6–7% in extended trading immediately after the print, with the capex raise — not the EPS miss/beat — driving the action.
• Opening April 30: Stock opened at $603.61, down 9.79% from the April 29 close, with intraday drawdowns approaching 10%. Volume was significantly elevated relative to the trailing 20-day average.
• Trailing performance through May 18: Shares have stabilized in the $610–625 range, down roughly 22–23% from the February high but still up materially on a one-year basis.

The market's reaction is fundamentally a duration debate, not a fundamentals debate. The Q1 print itself — 33% topline growth, 41% operating margin held flat, conversion-rate gains across the ad stack, a successful first MSL model launch — is unambiguously bullish for the underlying business. What spooked the tape was the combination of: (1) a $10B midpoint capex raise less than 90 days after the prior raise, (2) confirmation of a long-dated debt issuance that codifies the spend, (3) Li's explicit non-commit on 2027 capex with the tacit admission that compute needs have been chronically underestimated, and (4) a sequential DAP decline that — even with exogenous attribution — gave bears a "the user growth story is breaking" narrative to anchor on.

The price action is the textbook "beat-and-raise punished" pattern when the raise is on the spend line rather than the revenue line. In our view, the discount embedded in the post-print level is excessive relative to the ad-stack monetization trajectory and the MTIA cost-curve optionality, but the timing of when the multiple re-rates is genuinely uncertain — likely tied to the Q3 print, which will be the first quarter where the new capex guide is fully reflected in depreciation flow-through.

Street Perspective

Debate: Is the 2026 Capex Raise the Last One, or the First of Several?

Bull view: The raise is largely on memory pricing, which is a transient cyclical input cost rather than a structural ambition increase; with MTIA scaling in 2027, unit inference costs should fall materially even as gross capacity expands. Operating margin held at 41% in Q1 despite a 35% cost increase, demonstrating the underlying earnings power can absorb the build.

Bear view: Susan Li explicitly conceded Meta has "continued to underestimate" compute needs. A capex raise less than 90 days after the prior raise establishes a pattern — 2027 capex will almost certainly step up further, and the cumulative spend through 2028 will exceed $600B before any meaningful new revenue line monetizes. Depreciation will weigh on EPS for years.

Our take: Both are partly right. The bears are correct that another raise is likely; the bulls are correct that the absorbed margin tells you the business can fund it. The mistake the bears are making is assuming the spend is unproductive — the Q1 ad-stack metrics show it isn't. We side with the bull case on a 2-year-plus horizon, but acknowledge the next print could re-litigate this debate.

Debate: Is GAAP EPS of $10.44 Real, or an Accounting Mirage?

Bull view: Even stripping the $8.03B tax benefit, underlying EPS of $7.31 beat consensus by ~8% and was up 14% YoY against a quarter with materially higher depreciation. That is genuine earnings growth on top of a 33% topline.

Bear view: The headline obscures the fact that adjusted EPS grew at less than half the rate of revenue, which is the first time in five quarters that has happened. Margin compression is the real story, the tax benefit just papered it over.

Our take: Both are misleading. Adjusted EPS growing slower than revenue is mathematically inevitable when depreciation is being front-loaded against capacity that isn't yet productive. The right metric is operating income, which grew 30% — slower than revenue but well above the rate of margin-compression peers. The bear framing overstates the issue.

Debate: Does Personal-Superintelligence Vision Compete or Lose to OpenAI / Anthropic / Google?

Bull view: Meta is the only company with 3.5B+ DAP across four apps to which it can ship consumer AI at zero CAC. Muse Spark on Meta AI has already produced double-digit session-per-user gains. The infrastructure stack is being built end-to-end (model + MTIA silicon + distribution surface). No competitor has all three.

Bear view: The lab is 10 months old and shipped one model. OpenAI is years ahead on consumer mindshare; Google has equivalent distribution and a longer foundation-model track record. Zuckerberg's product cadence non-commitments suggest the team is not yet confident in shipping at scale.

Our take: The distribution argument is structurally undervalued. Even if Muse Spark generation N is 6 months behind GPT-5/Gemini-Ultra-N on benchmarks, putting it inside WhatsApp, Instagram, and Facebook gives Meta a faster pathway to mass adoption than any standalone competitor. The bull case is correct over a 3-year horizon; the bear case is correct over a 6-month one.

Debate: Is the EU/US Regulatory Tail Material to the Investment Case?

Bull view: Every quarter for five years has had some version of "regulatory headwinds may impact results" boilerplate. The actual fines and settlements have been manageable in dollar terms. Youth-safety litigation will produce specific cases, not existential outcomes.

Bear view: Li elevated the language from "may impact" to "may ultimately result in a material loss" — a specific change in disclosure that warrants attention. Multiple US trials are scheduled in 2026, and the company already lost two such cases in March (per reporting). Cumulative damages could be in the multi-billion-dollar range.

Our take: The change in disclosure language is the most important signal investors are underpricing. We don't model a specific dollar impact, but we acknowledge a low-probability, high-magnitude scenario where 2026 youth-safety settlements take $3–10B out of FCF. That is reflected in our rating but does not change the Outperform call.

Model Update Needed

Valuation impact: Our DCF target moves from ~$780 to ~$800 on the revenue/margin revisions, partially offset by the higher capex profile. Against a current ~$615 print, this implies ~30% upside to fair value over 12 months, more if MTIA scale-out compresses unit inference costs in line with peer benchmarks.

Thesis Scorecard Post-Earnings

Overall: Thesis strengthened on the ad-stack monetization side, neutral on the AI-consumer-product side, and weakened on the regulatory tail. The capex bear case is more credible than it was a quarter ago, but the operating margin defense and conversion-rate evidence make it manageable rather than thesis-breaking.

Action: Initiate Outperform. The ~22% sell-off from the February high overdiscounts the capex story relative to the demonstrated ROIC trajectory. Build positions through the next two prints; the Q3 2026 release (where new capex guide fully reflects in depreciation flow-through) will be the next major test of the margin-defense thesis. A capex-driven re-rate lower from current levels would only improve the entry.