The Margin Break Is Opex, Not Depreciation, and Depreciation Is Still Ahead

Results vs. Consensus

Quality of the miss. The 14% EPS shortfall is arithmetically the two charges and nothing else. Adding back the $2.4B legal accrual and $1.2B of severance at the quarter's 16% effective tax rate implies EPS of roughly $7.36, which would have beaten consensus by about 2%. That is the honest read of the headline. It is also where the honest read stops: the operating deterioration is not a charge story. Excluding both items, operating income rose 9% on revenue that rose 28%, and the ex-charge margin of roughly 37% is still more than six points below last year's 43%. Two-thirds of the margin compression this quarter was clean.

Year-over-Year Comparison

Sequential Comparison

The sequential table needs one caveat before it is read as a collapse. Q1's $10.44 of EPS carried an $8.03B one-time income tax benefit, which is why the first half shows a net tax benefit of $2,113M rather than an expense. Strip that and the underlying sequence is roughly $7.31 to $6.18, a 15% decline on 8% sequential revenue growth. That is still the wrong direction, but it is a decline of a different order than the headline suggests.

Revenue

The top line is not the problem and should not be treated as one. Revenue of $60,801M grew 28%, or 27% on a constant-currency basis, so foreign exchange contributed roughly a point rather than the four points it contributed in Q1. Underlying demand held up almost entirely on its own. Within that, the composition shifted in a way worth watching: impressions grew 14% against 19% in Q1, while pricing held at 12% for a second straight quarter. Management attributed impression growth to "growth in engagement and users as well as ad load optimizations," and the last of those three is a finite lever. When volume growth decelerates five points in a quarter and price does not move to compensate, the revenue growth rate follows, which is exactly what the Q3 guide encodes.

Margins

Twelve points of operating margin disappeared in a year, and the decomposition matters more than the number. Gross margin fell 76 basis points, from 82.1% to 81.4%, which means the infrastructure build has so far cost Meta less than one point of gross profitability despite cost of revenue growing 33%. Everything else happened below the gross line. R&D rose $8,714M, and even removing the $1.2B of severance it grew roughly 58%, taking R&D from 27.2% of revenue to 35.6%. G&A more than doubled, though $2.4B of that is the legal accrual. The CFO named the drivers plainly: technical hires "particularly AI talent," plus third-party AI token costs. Those are cash operating expenses that hit the quarter they are incurred. The depreciation on $130–145B of 2026 capex is a separate, later, and larger wave.

EPS

Three things took EPS from $7.14 to $6.18, and only one of them is the charges. The legal accrual and severance together cost about $1.18 per share after tax. Tax cost roughly another $0.33: the quarter's $2,908M provision is about $845M more than the prior-year quarter's 11% rate would have produced on the same pre-tax income, because that 11% was itself unusually low. Offsetting those, revenue growth and the buyback that used to exist did not contribute what they normally would: with zero repurchases, basic share count actually rose 1% year over year even as diluted shares were flat, meaning stock-based compensation dilution is now flowing through unabsorbed for the first time in years.

The Cost Line

This is the section that would normally be a footnote and this quarter is the thesis. Meta grew revenue by $13,285M and grew costs by $14,951M. A company cannot do that for long and describe itself as operating-leverage-positive.

"Q2 total expenses were $42 billion, up 55% compared to last year and included $2.4 billion in charges related to legal proceedings and $1.2 billion in severance expenses in connection with the May 2026 head count reduction." — Susan Li, CFO

Read the table by column rather than by row. R&D alone accounts for 58% of the entire dollar increase in costs. Cost of revenue, the line most investors are watching because it is where depreciation lands, contributed only 19% of the increase and grew slower than the 55% total. Marketing and sales actually fell as a share of revenue. The single biggest change in Meta's cost structure this year is that it is paying dramatically more for AI engineers and for third-party model inference, and neither of those is capitalized.

Assessment: The market spent this print litigating capex. The income statement says the near-term issue is operating expense, which is worse in one respect and better in another. Worse, because opex hits immediately and is harder to throttle than a data-center build. Better, because it means the depreciation step-down in gross margin that bears have modeled has barely begun, so the margin trough is demonstrably not in. Anyone underwriting a recovery in operating margin over the next four quarters is underwriting two separate reversals, not one.

Segment Performance

Family of Apps

The segment that funds the company grew 28% and lost 14 points of its own operating margin, from 53.0% to 38.8%. Both facts are true and the second is the one nobody expected. Advertising revenue of $59,363M grew 27.5%, and the Other line crossed $1B for the first time on WhatsApp paid messaging and subscriptions, up 72.7%. Neither the legal accrual nor the severance sits in Reality Labs, so attributing both charges to Family of Apps would put the segment's clean margin somewhere near 45%, still eight points below last year.

The monetization machinery underneath is genuinely working, which is what makes the margin line so uncomfortable. Meta introduced what it calls Meta Generative Recommender this quarter, deploying its first generative model into the ads retrieval system rather than scoring each candidate ad individually. The measured results are specific and large.

"Combined with our GEM model for ads ranking and sequence learning, these advancements generated an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook." — Susan Li, CFO

Advantage+ end-to-end solutions reached over $75B in annual revenue run rate, and 9 million small businesses now use at least one generative creative tool. On the organic side, a single Reels ranking release drove a 15 basis point increase in Instagram sessions, and more than half of recommended Instagram Feed content is now less than a day old, double the share of a year ago.

Assessment: The ad stack is not the problem, and an investor waiting for evidence that the AI spend touches revenue got it again this quarter. What changed is that the spend required to produce those gains is now growing faster than the gains themselves. A 27.5% advertising growth rate against a 55% expense growth rate is not a sustainable configuration, and the segment margin is the place it shows up first.

Reality Labs

Revenue grew 16.5% to $431M on AI glasses strength, partially offset by lower Quest headset sales, and the operating loss widened to $4,619M from $4,530M. That is worth stating plainly because last quarter we described the loss as no longer expanding: it is expanding again, by $89M year over year and by $591M sequentially from Q1's $4,028M. On a Q2 run rate the segment is now an $18.5B annual drag, and the first half loss of $8,647M against $8,739M a year ago means the improvement Meta banked in Q1 was given back entirely in Q2.

The glasses line remains the one place in this segment with a real curve. Meta launched its own line with EssilorLuxottica during the quarter, the first to ship with Muse Spark on the device, and management put a date on the next reveal.

"Early sales have been strong, exceeding our expectations, and we're going to have more to share on our glasses lineup at this year's Connect conference on September 23." — Mark Zuckerberg, CEO

Assessment: Reality Labs has now absorbed roughly $18B a year for several years and the loss is drifting wider, not narrower, while the one product with genuine momentum generates a few hundred million dollars a quarter inside a segment losing $4.6B. September 23 is a real catalyst for the glasses narrative. It is not a catalyst for the segment's arithmetic.

Key KPIs

Cash Flow and Capital Returns

If the income statement is where the quarter got uncomfortable, the cash flow statement is where it became a different company.

Operating cash flow grew a healthy 24.7%. Capex including finance-lease principal grew 82.7%, to $31,078M, which is 51.1% of revenue against 35.8% a year ago and consumed 97.5% of the quarter's operating cash flow. What was left, $784M, did not cover the $1,353M dividend. The shortfall came off the balance sheet.

Assessment: Zero buybacks in a half-year is the single most concrete signal management sent about how tight the funding position has become, and it went almost unmentioned on the call. Meta returned $22.9B to shareholders through repurchases in the first half of 2025 and $0 in the first half of 2026. Net cash fell by $16.3B in six months even after $24.9B of incremental debt. None of this is distress: $90B of cash against $84B of debt on a business generating $64B of operating cash flow in six months is a strong balance sheet by any normal standard. But it is no longer a balance sheet that funds a buyback, and investors who owned this stock partly for the share-count reduction now own a different instrument. Property and equipment rising $49.3B in six months is also the clearest possible preview of the depreciation schedule to come.

Key Topics & Management Commentary

Overall Management Tone: Expansive on opportunity and notably unapologetic about the cost of pursuing it, with essentially no acknowledgement that the quarter's margin or cash-flow prints required explanation. The prepared remarks opened three new revenue narratives (enterprise, API, selling compute directly) before the CFO reached the expense line, and the framing throughout treated capacity as the binding constraint and profitability as a later question. Where management was least convincing was on sequencing: asked directly which of the new opportunities would scale first and against what return threshold, the answer was that all of them look promising. That is now two quarters running in which the specificity of the spending has outrun the specificity of the return.

Where the Twelve Points Went

Management did not volunteer a margin bridge, so the useful disclosure came in fragments. The CFO identified the cost drivers as "increases in employee compensation, infrastructure costs, legal-related costs and third-party AI token costs," and separated the compensation growth from severance explicitly.

"Excluding the previously mentioned severance expense, growth in employee compensation was driven by technical hires we've added over the past year, particularly AI talent. The growth in infrastructure costs was driven by higher depreciation, data center operating costs and third-party cloud spend." — Susan Li, CFO

The only margin framing offered was the ex-charge comparison: operating income "would have increased 9% year-over-year." True, and also a 19-point gap against 28% revenue growth that went unaddressed.

Assessment: Third-party AI token costs are the most interesting item in that list and the least discussed. Meta is paying other model providers at scale while simultaneously building the largest private compute fleet in the industry and telling investors it could sell that compute at a premium. Those two facts sit awkwardly together, and no analyst pressed on the size of the token bill.

The Capex Guide Narrowed Rather Than Rose

This is the quarter's most underappreciated positive and it cuts directly against what we expected. Coming out of Q1, our base case was another raise, on the reasoning that the CFO had conceded chronic underestimation of compute needs and had just moved the range up $10B at the midpoint. Instead the 2026 range moved to $130–145B from $125–145B: the floor came up, the ceiling did not, and the midpoint moved $2.5B. After two consecutive quarters of upward revisions, the guide effectively stopped moving.

The framing around 2028 was also new and more disciplined than prior calls. Management described a strategy of laying down land, power, and network foundations now while deferring the expensive server decisions.

"Therefore, our longer-term capacity strategy aims to give us the flexibility to continue growing compute in 2028 and beyond by laying down data center and network foundations to accommodate future server decisions. The long-lived nature of these assets inherently provides the flexibility that will make it possible to adjust our investment to the pace of AI adoption." — Susan Li, CFO

Assessment: Separating the long-lived, low-obsolescence portion of the build from the short-lived, expensive silicon portion is the right structure and the first genuine capital-discipline argument management has made in three quarters. It does not reduce 2026 or 2027 spending. It does mean the 2028 commitment is more optional than the headline capex trajectory implies, which is worth something to a discounted cash flow that has to guess at terminal capital intensity.

The Legal Charge Stopped Being Hypothetical

In our Q1 note we called the regulatory tail the most underpriced overhang on the stock and sketched a scenario in which youth-safety settlements took $3–10B out of free cash flow. One quarter later, $2.4B of it is in the income statement, and management raised the low end of the full-year expense guide specifically to accommodate it. The forward language did not soften.

"For example, we continue to see scrutiny on youth-related issues in several markets and have a number of youth-related trials scheduled for this year in the U.S., which may ultimately result in a material loss." — Susan Li, CFO

The phrase "may ultimately result in a material loss" is identical to the language used last quarter. What changed is that it is now accompanied by an actual accrual, which converts it from boilerplate into a disclosed estimate of a subset of the exposure.

Assessment: A company takes a $2.4B accrual when it can estimate a loss, not when it has finished losing. With trials still scheduled this year, the base case should be further charges rather than this one clearing the docket. This is the thesis point that moved furthest against us this quarter, and it moved from a risk we flagged to a cost we can measure.

The Enterprise and Compute Pivot

The genuinely new strategic content of this call was Meta describing itself, for the first time, as a prospective seller of compute and enterprise services. The CEO listed it third among the company's three opportunities, alongside the core business and personal agents, and returned to it repeatedly.

"And third, we see a large enterprise opportunity to sell to businesses, including APIs, business agents, potentially selling compute directly and other services that we're building for large customers." — Mark Zuckerberg, CEO

The supporting evidence is concrete but early: a public API for Muse Spark, distribution via OpenRouter for U.S. developers, the Meta Business Agent platform for enterprises, and a claim that inbound demand for capacity is priced above cost. Management was careful to rank the margin profile of the options.

"The question in thinking about this is we believe that there will continue to be a significantly higher margin on selling intelligence rather than selling compute directly. But we think that there is a big opportunity, obviously, to sell compute as well." — Mark Zuckerberg, CEO

Assessment: Treat this as real optionality with zero current revenue and a meaningful signalling cost. A company that can rent its compute at a premium has a floor under the return on its capex, which materially changes the downside case if AI product monetization disappoints. But announcing a compute-rental option in the same quarter that free cash flow fell 91% invites the reading that the spend needs a monetization story faster than the products can supply one. Both readings are available and the market took the second.

LLMs Move Into Ranking and Retrieval

The technical disclosure this quarter was more substantial than the ad-stack detail in Q1, and it is where the capex argument is strongest. Meta described replacing per-candidate ad scoring with a generative approach that reasons over ad content and user preference jointly, and reported the first production results.

"This quarter, we introduced Meta Generative Recommender, a paradigm shift in how our ad system works. Rather than scoring every possible ad individually, we are now using LLMs to reason about ad content and user preferences together and predict the best ad for each person." — Susan Li, CFO

On the organic side, every public Reels and Feed post on Instagram is now processed through an LLM and analyzed across dimensions from topic to tone, with those signals feeding ranking, recommendations, and policy enforcement. Management reported hitting a first research milestone on continuously pretraining a large-scale model on recommendations data and observing healthy scaling laws.

Assessment: This is the mechanism by which compute becomes revenue, and it is producing measurable output on a quarterly cadence: 8.3% more ad clicks and 15.7% higher conversions on Facebook this quarter, following comparable gains in Q1. The bull case does not require any new product to work. It requires this to keep compounding at roughly this rate for long enough to outrun the cost curve. So far it is compounding and the cost curve is winning.

Engagement Milestones Against Slowing Impressions

The engagement disclosures were the strongest in several quarters: 3.6 billion daily users, Instagram crossing 2 billion daily actives, Threads passing 500 million monthly, and a WhatsApp record of 30 million messages per second during the World Cup final. Instagram time spent grew double digits, Facebook video time grew 9% globally and more than 10% in the U.S. and Canada.

Set against that, DAP growth of 3% is a point slower than Q1, and impression growth of 14% is five points slower. Both can be true: engagement per user is rising while the user base matures and the ad load lever approaches its useful limit.

Assessment: The engagement data resolves the Q1 scare, when a sequential DAP decline driven by Iran and Russia disruptions gave bears a user-growth narrative. That narrative is dead. The replacement concern is arithmetic rather than existential: with users growing 3% and load optimization finite, revenue growth increasingly has to come from price and conversion quality, both of which depend on the AI spend continuing to deliver.

The Lab: Two Model Releases and a Return to Open Weights

Meta Superintelligence Labs shipped Muse Spark 1.1 and Muse Image in the month before the call, and the assistant metrics moved with them: a 60% increase in daily Meta AI users since the rebuild and Muse Spark integration, still growing weekly. Muse Spark 1.1 was positioned as an efficient agentic coding model strong at computer use and tool use, now available through a public API.

Asked about open source, the CEO signalled a reversal of the closed posture MSL adopted during its rebuild, while explaining why the pause happened.

"And I just wanted to make sure that the MSL team was uninhibited in building out the most intelligent models that we could. And we expect that we will get back to releasing some open source models at some point soon." — Mark Zuckerberg, CEO

Assessment: Shipping two models in a month is a real change in cadence from a lab that produced one in its first ten months, and it partially answers the product-velocity question that went unanswered in Q1. The open-weight signal is strategically consistent with selling inference and enterprise services rather than model access alone. What is still missing is any disclosure of what the API earns.

Meta One and the Subscription Line

The company launched Meta One during the quarter, a subscription bundling additional tools and AI features across the apps, with tiering promised as demand grows. It sits alongside the Family of Apps Other line that crossed $1B for the first time this quarter on WhatsApp paid messaging and subscriptions, growing 72.7%.

Assessment: Other revenue is now growing more than twice as fast as advertising off a base that just reached $1B, which makes it the fastest-compounding line in the company and still only 1.7% of revenue. Meta One is the first consumer subscription with a real surface area behind it. Neither is close to mattering to 2026 earnings, and both matter a great deal to the 2028 argument for why this capex was rational.

Headcount Down 8,000, Compensation Still Up

Meta ended the quarter at 75,472 employees, down 3.2% sequentially, with roughly 8,000 people affected by the May reduction and the majority expected to leave headcount by the end of Q3. The $1.2B severance charge is the cost of that action. And yet employee compensation still grew materially year over year, driven by AI hiring.

Assessment: This is a barbell, not a cut. Meta is removing roughly 8,000 general roles while paying premium compensation for AI researchers, and the net effect on the compensation line is upward. The severance is genuinely one-time; the elevated comp structure that survives it is not. Investors modelling a post-restructuring opex step-down should model the opposite.

The BlackRock Venture and How the Build Gets Financed

Meta announced a strategic venture with BlackRock to develop a 1 gigawatt data center in El Paso, Texas, described as an example of partnership structures that complement the company's own balance sheet. The CFO placed it within a broader shift toward long-duration external capital, and long-term debt rose $24.9B in six months.

Assessment: Partnership structures for data centers are the most consequential unquantified item on this call. Depending on consolidation treatment, a joint venture can move capacity additions outside reported capital expenditures, which would make the headline capex guide a less complete measure of what Meta is actually committing. No detail was given on Meta's contribution, the economics, or the accounting. Until it is, treat the $130–145B guide as a floor on total capacity spending rather than a ceiling.

Guidance & Outlook

The Q3 revenue guide is the proximate cause of the sell-off. A $62.5B midpoint against roughly $63.2B of consensus is not a large shortfall in dollars, but it lands on a decelerating series and it embeds a currency reversal: Q1 carried a four-point FX tailwind, Q2 carried about one point, and Q3 is guided to a one-point headwind. On the underlying trend, the guide is materially softer than the optics suggest.

Implied Q3 growth: against Q3 2025 revenue of $51,242M, the $61–64B range implies 19.0% to 24.9% growth, with a 22.0% midpoint. Set against 33.1% in Q1 and 28.0% in Q2, that is a five-to-six-point step down in each of two consecutive quarters. Growth was still accelerating into Q1, which grew 33.1% against Q4 2025's 23.8%, so this roll-over is two quarters old rather than a long-running trend.

What the operating income commitment now requires. FY2025 operating income was $83,276M, the sum of the four reported quarters. Holding management to "above 2025" at the $167B midpoint of the expense guide means full-year revenue must exceed roughly $250.3B. The first half delivered $117,111M and the Q3 midpoint adds $62,500M, which leaves Q4 needing about $70.7B, or 18.0% growth against Q4 2025's $59,894M. That sounds demanding until it is expressed sequentially: it requires a 13.1% step up from the Q3 midpoint, against actual Q3-to-Q4 sequential gains of 16.9% in 2025 and 19.2% in 2024. The commitment holds comfortably on normal seasonality.

What the commitment does not mean. On those same assumptions, full-year operating income lands only a few percent above 2025 on revenue growing roughly 26%, with a full-year operating margin near 34% against 41.4% in 2025. "Operating income above 2025" is a floor that happens to be satisfied by arithmetic, not a promise of operating leverage. Investors should read it as the weakest commitment management could make while still making one.

Guidance style: historically conservative on revenue, and the Q1 print landing at the top of its guided range was consistent with that. The tax-rate revision is the item most likely to be underweighted: two points on roughly $22B of quarterly pre-tax income is a real second-half EPS headwind that has nothing to do with operations.

Analyst Q&A Highlights

Which Opportunity Scales First, and Against What Return

The opening question of the call went straight at the gap between the breadth of announced opportunities and the absence of any prioritization or return threshold. It asked, in effect, for one thing management would commit to showing quantifiable returns on. The answer declined to rank them.

Q: "My question is, as you look at these opportunities and the state of the current offerings, the compute capacity, which of them do you expect to be able to scale first sort of in '26 and '27 to sort of showcase quantifiable material ROIC for investors?"
— Brian Nowak, Morgan Stanley

A: "I think you asked what one area would likely scale the most. But I mean I'm actually quite optimistic that we're going to see meaningful growth in all of these areas. And I think we will have more to share soon on a number of them."
— Mark Zuckerberg, CEO

Assessment: This is the same non-answer the same question received last quarter, and it is now a pattern rather than an instance. Three consecutive quarters of accelerating spend without a single published return threshold is the strongest argument the bears have, and it is entirely self-inflicted. Management could defuse it with one disclosed metric and continues to choose not to.

2027 Capex: A Second Consecutive Refusal

The follow-up sought any framing at all on next year's capital envelope, explicitly offering to accept philosophy in place of numbers. The response reiterated that no outlook would be given, while confirming that planning is oriented toward maximizing capacity in exactly the years for which no figure exists.

Q: "Any early comments on '27 CapEx, even the philosophy around sources of upside or sources of downward pressure as you sort of think through different ways to finance this multiyear build and have '27 CapEx."
— Brian Nowak, Morgan Stanley

A: "Brian, on your second question, we aren't providing a specific outlook for 2027 CapEx at this time. Infrastructure planning remains highly dynamic. And even this year, there are a range of outcomes embedded in our outlook."
— Susan Li, CFO

Assessment: The refusal is defensible on its own terms and corrosive in aggregate. Sell-side models now carry 2027 capex assumptions ranging widely enough that the resulting free-cash-flow estimates are not comparable to one another, which widens the dispersion of price targets and raises the risk premium on the equity. The company is paying for this silence in its multiple.

Demand-Constrained, Not Supply-Constrained

The most clarifying exchange of the call came last. A question pressed on whether the stated focus on maximizing 2026 and 2027 capacity was a statement about available supply or about internal demand, and whether Meta intends to consume all of what it builds. The answer was unusually direct and is the single best argument available for the capex program.

Q: "You noted that in your prepared remarks that you plan to maximize '26 and '27 capacity. Is that a demand or supply comment? Meaning, are you suggesting that Meta plans for internal use of all the capacity built through '27? Or are you saying that you will evaluate '28 and beyond builds based on demand for Meta products and services?"
— Ken Gawrelski, Wells Fargo

A: "When we refer to focusing on 2026 and '27 capacity, there are really 2 factors. One is we are today and expect to be in the sort of foreseeable future demand constrained. And that really includes our core business, too, where there are -- we still have numerous ROI-positive places that we would put compute toward if we had it."
— Susan Li, CFO

Assessment: "We still have numerous ROI-positive places that we would put compute toward if we had it" is the closest thing to a return argument anywhere on this call, and it is stronger than it sounds because it is grounded in the core advertising business rather than speculative products. If the marginal GPU is going into ad ranking, the return is measurable today and the ad-stack results suggest it is positive. That this appeared as an aside in the final answer rather than as a framing in the prepared remarks is a communication failure.

Selling Compute While Buying It

A question probed the apparent contradiction between claiming premium inbound offers for capacity and simultaneously purchasing third-party capacity, asking whether the difference was timing, or training versus inference. The answer reframed both as consequences of a single condition.

Q: "And then, Mark, just in terms of the number of offers to monetize your compute externally, you also, at the same time, are purchasing capacity from a number of third parties. So just hoping you can help us understand some of the differences here. Is it just timing and stopgap issues? Or is it training versus inference and leveraging the chips that are best suited for each?"
— Doug Anmuth, JPMorgan

A: "I mean, look, the high-level observation is that there's just nowhere near enough compute for all the demand. So that is why we see that like basically, we are getting a large number of offers for the compute that we have, but also we have a lot of internal uses that we think are going to be quite valuable."
— Mark Zuckerberg, CEO

Assessment: The answer is coherent and it dodges the specific question, which was about the economics of the arbitrage rather than the existence of scarcity. Buying third-party capacity at market while claiming the ability to sell owned capacity at a premium implies either a spread worth quantifying or a distinction between fungible and non-fungible capacity worth explaining. Neither was offered, and the third-party token and cloud spend showing up in the cost lines suggests the flow is currently running in the expensive direction.

The Enterprise Muscle Meta Does Not Yet Have

A question separated the enterprise opportunity into the portion reachable through the existing advertiser relationship and the portion requiring a genuinely new go-to-market capability. The answer conceded the second category directly, which was the most candid moment of the call.

Q: "When you frame up the enterprise opportunity, how much of that opportunity do you think is available to you today based on what you've built out in terms of go-to-market strategy as extensions of the advertising and the marketing business you have already versus go-to-market strategies that might have to be built to capitalize on the opportunity?"
— Eric Sheridan, Goldman Sachs

A: "And now that we have those, we feel like there's a large opportunity to serve whether that's small businesses or larger businesses, that is a somewhat different muscle than we have historically had, and we will share more soon on how we're planning to build that out."
— Mark Zuckerberg, CEO

Assessment: Admitting that enterprise sales is "a somewhat different muscle" is correct and underrated as a risk disclosure. Meta has never run a direct enterprise sales motion; its advertising business is self-serve and auction-mediated at the long tail and agency-mediated at the top. Building enterprise distribution takes years and carries a sales-and-marketing cost line that does not currently exist in this P&L. The opportunity is real; the timeline implied by "we will share more soon" is not.

Funding the Build: Debt, Equity, and Partnerships

A question asked how management philosophically reconciles ambition on spending with the capital required to sustain it. The answer confirmed a deliberate, ongoing shift toward debt and external partnership structures rather than a one-time financing event.

Q: "We continue to get questions from investors about how to think about the mix of debt and equity and sources of capital. Philosophically, how are you guys thinking about wanting to be ambitious on the spend, but then marrying that with the need for capital?"
— Eric Sheridan, Goldman Sachs

A: "But we've also been evolving our capital structure in recent years to include a greater mix of debt as we work to bring down our cost of capital, and we have generally found it prudent to continue adding cost-efficient long-duration sources of capital as we make investments in initiatives that themselves have long time horizons, especially AI infrastructure projects."
— Susan Li, CFO

Assessment: Read alongside $24.9B of incremental debt in six months and a buyback at zero, this is confirmation that the capital structure change is structural rather than opportunistic. Matching long-duration liabilities to long-lived assets is textbook correct. It also means the equity is now levered to the outcome of the AI build in a way it was not eighteen months ago, and the appropriate multiple on a levered outcome is lower than on an unlevered one.

Personal Agents: Still Unshipped

A question on whether consumer AI adoption is close to a breakthrough or still requires patience drew an extended answer on why consumer agents are harder than coding agents, ending in an unusually plain admission about where the product actually stands.

Q: "Consumer behavior is always pretty difficult to predict, but kind of reading your op-ed on AI for everyone, how do you think about whether consumer adoption can close this AI utility gap? Like are we on the cusp of something kind of breaking through? Or do we just need to be a bit more patient?"
— Mark Shmulik, Bernstein

A: "Now I get we're going to ship this at some point soon, and that's going to be very exciting. And we haven't done that yet. So this is -- there's kind of only so much that I can say on an earnings call about this."
— Mark Zuckerberg, CEO

Assessment: "We haven't done that yet" is the most honest sentence on the call and the most expensive. Personal agents are the product that justifies the training portion of the capex, and after four quarters of being described as imminent they remain unshipped. The diagnosis offered, that consumer agents must work out of the box for billions of people rather than for technical users willing to tinker, is correct and is also a description of a hard problem with an unknown completion date.

What They're NOT Saying

1. Any return threshold on the AI build. Asked directly in the first question of the call, for a second consecutive quarter, and declined again. No dollar target, no margin target, no timeline, no signpost.
2. 2027 capital expenditures. A second consecutive refusal, in a year when planning is explicitly oriented toward maximizing 2027 capacity. Management knows the number well enough to build against it.
3. The size of the compute-sale offers. "A significant premium over what we paid" appeared twice with no dollar value, no counterparty type, no contracted volume, and no indication whether any offer has been accepted.
4. Third-party AI token spending. Named as a driver of expense growth and never sized, despite being the one cost line that would tell investors how much of Meta's own AI usage is running on someone else's models.
5. Whether the buyback resumes. Zero repurchases in two consecutive quarters, $22.9B of first-half repurchases eliminated year over year, and not one word on the call about the suspension or the conditions for restarting it. No analyst asked.
6. Reality Labs 2026 loss guidance. Still none, now for a fourth quarter, while the loss widens both sequentially and year over year.
7. Any margin guidance. The company guides revenue, expenses, capex, and tax, which brackets margin without committing to it. In a quarter where margin fell twelve points, the omission is conspicuous.
8. The BlackRock venture's economics. No disclosure of Meta's capital contribution, ownership, consolidation treatment, or whether the associated capacity sits inside the $130–145B guide.
9. Remaining youth-litigation exposure. A $2.4B accrual was taken and the "material loss" language retained, with no range, no case count, and no framework for the rest.
10. Meta One pricing or contribution. Launched during the quarter, described as having tiers coming, quantified nowhere.
11. Muse Spark API pricing and revenue. A "competitive price" and "encouraged by the initial results" is the entirety of the disclosure on what is positioned as a new revenue line.
12. Manus. The open item from last quarter's call, promised an update, and not mentioned once.

Market Reaction

• Pre-print setup: META closed at $585.61 on July 29, down 11.3% year to date against the S&P 500's +6.9%, and down 16.3% over the trailing twelve months. The stock had recovered 4.1% over the trailing thirty days and sat well off the top of its 52-week closing range of $525.72 to $790.00. This was a de-rated, out-of-favour setup rather than a crowded long.
• After-hours July 29: down approximately 10% on the print, with two independent sources placing the move at roughly -9.6% to -10%.
• Reaction session July 30: opened at $526.00, a 10.2% gap down, traded a $524.49 to $539.88 range, and closed at $539.03, down 8.0% or $46.58. Volume of 42.3M shares was 2.3 times the 30-day average. The S&P 500 rose 1.7% the same day, so the relative move was close to ten points.
• Follow-through July 31: opened at $543.60 and closed at $556.71, up 2.4% on 24.3M shares, recovering more than a third of the reaction-day decline. Two-session net from the pre-print close: -4.9%.

The sell-off was not about revenue, which beat, and only partly about EPS, which missed entirely on charges that reverse. It was about the three things that arrived together: a Q3 guide whose midpoint sat below consensus on a decelerating growth series, an operating margin twelve points lower than a year ago, and a free-cash-flow line that rounded to zero after the dividend. Any one of those is survivable. Delivered simultaneously by a management team that declined to quantify a return on the spending causing them, they repriced the multiple rather than the estimates.

The July 31 bounce is worth reading carefully rather than dismissing. Recovering more than a third of the decline on lighter volume, in a stock that entered the print already down 16% over twelve months and now trades at roughly 18 times forward earnings, suggests the reaction-day move overshot on positioning rather than establishing a new fundamental level. It does not suggest the debate is resolved. The next real test is the Q3 print, where the first meaningful tranche of 2026 capex begins flowing into depreciation and the guide's own deceleration has to be either confirmed or beaten.

Street Perspective

Debate: Is the Margin Trough In?

Bull view: Q2 carried $3.6B of charges that do not recur, so the clean margin is roughly 37% rather than 31%. Severance is behind the company, the headcount reduction lowers the compensation base from Q3, and the ad-stack gains keep compounding. Margin stabilizes in the second half and recovers through 2027 as revenue outgrows a flattening expense base.

Bear view: Clean margin still fell more than six points year over year, and the largest cost pressure has not arrived. Property and equipment grew $49.3B in six months against a $130–145B annual capex program, and essentially none of that depreciation is in the run rate yet. Adding a depreciation wave to an elevated compensation base means 2027 margins are lower than 2026's, not higher.

Our take: The bears have the better of this and it is not close. The one-time charges are genuinely one-time, but they are the smaller half of the story, and the composition of the compression argues against a near-term recovery: opex arrived first and depreciation follows, so the two pressures stack rather than hand off. We would need to see either an expense guide that implies a flat second half or an explicit margin floor from management before underwriting a trough. Neither was offered.

Debate: Is the Enterprise and Compute Pivot a Real Business or a Deflection?

Bull view: Inbound demand for capacity at a premium to cost puts a floor under the return on capex that did not exist a quarter ago. If AI products disappoint, Meta can rent the fleet; if they succeed, it keeps the higher-margin outcome. That optionality is worth real money and is in nobody's model.

Bear view: The pivot was announced in the same quarter free cash flow collapsed, with no revenue, no pricing, no counterparties, and an admission that enterprise sales is a capability the company does not have. It reads as a narrative supplied to justify spending whose original justification, consumer AI products, has not shipped.

Our take: Both readings are defensible and the market chose the second, which we think slightly overshoots. The demand-constrained disclosure is the tell: if Meta genuinely has ROI-positive internal uses it cannot fund with current capacity, then the compute is not idle and the rental option is a genuine alternative use rather than a fallback for a failed one. But the option is worth nothing until something is contracted, and we will not capitalize it until it is.

Debate: Does Zero Buyback Change the Equity Story?

Bull view: Suspending repurchases to fund the highest-return investment opportunity in the company's history is exactly what a rational allocator does. The balance sheet still holds $90B of cash against $84B of debt. Buybacks resume when the build normalizes, and shareholders are better off owning the capacity than the retired shares.

Bear view: Free cash flow no longer covers the dividend, net cash halved twice over in six months, and share count is rising on a basic basis for the first time in years as stock compensation goes unabsorbed. A company financing its dividend from the balance sheet while adding $25B of debt has lost the capital-returns leg of its investment case, and management did not think the change merited comment.

Our take: The bulls are right about the logic and wrong about the disclosure. Redirecting capital to investment is defensible; doing it silently is not. The practical consequence is that a shareholder base that included capital-return buyers now holds a levered growth-capex story, and that constituency rotation is part of what the last two quarters of relative underperformance represents. We treat the buyback as suspended through 2027.

Debate: Is 18x Forward Earnings Cheap or a Value Trap?

Bull view: A business growing revenue 28% with an ad engine producing measurable conversion gains every quarter, trading at roughly 18 times forward earnings and 16% below where it started the year, is inexpensive against both its own history and the megacap peer group. The de-rating has already happened.

Bear view: The multiple looks reasonable only because it is applied to earnings that are about to stop growing. On our own numbers, 2027 EPS is roughly flat against 2026 as depreciation lands, which means the stock trades at 18 times a stagnant earnings stream, not a compounding one. That is not a discount; it is a fair price for a plateau.

Our take: This is the crux and it resolves to the margin question. If 2027 operating margin holds near 34%, the multiple is defensible and the stock works from here. If it compresses another three to four points as depreciation flows, earnings decline and 18 times becomes expensive. We do not have enough visibility to underwrite either outcome, which is precisely why this is a Hold rather than a call in either direction.

Model Update Needed

Valuation impact: Our target moves to $610 from approximately $800. The reduction is almost entirely the operating-margin revision, which removes roughly $17B of annual operating income relative to our prior path, compounded by a lower terminal multiple to reflect the levered balance sheet and the suspended buyback. Against the July 31 close of $556.71, the new target implies roughly 10% upside, which is a market-like return and consistent with a Hold. We would revisit constructively on either an explicit margin floor from management or a contracted enterprise or compute revenue line, and negatively on a 2027 capex figure above $175B without accompanying revenue commitments.

Thesis Scorecard Post-Earnings

Overall: Thesis materially weakened. Of the four bull pillars carried into this quarter, two confirmed, one slipped to neutral, and the one we identified in Q1 as the near-term load-bearing pillar broke. Of the four bear points, three moved against us and one improved. The revenue and engagement half of the thesis is intact and arguably stronger than it was in May; the profitability and cash-generation half is not.

Action: Downgrade to Hold. Our Q1 initiation argued that a 22% drawdown overdiscounted the capex story relative to demonstrated returns, and said the Q3 print would be the test of margin defense. The test arrived a quarter early and the answer was negative, so the honest response is to mark the call rather than defend it. This is not a thesis reversal: the ad engine is compounding, engagement milestones are real, and the capex guide finally stopped rising. It is a recognition that we cannot underwrite the margin path with the depreciation wave still ahead and management declining, for a second consecutive quarter, to publish any return threshold on the spend causing it. Hold the position, do not add, and revisit on either an explicit margin floor or the first contracted dollar of enterprise revenue.