Read the Story: Meta Platforms, Inc.

Management Narrative Analysis — FY2021 to FY2026

Generated 8 September 2026

Evidence base: prepared remarks from 19 Meta earnings calls (Q3 2021 through Q2 2026) and the Item 1 / Item 7 narrative sections of the FY2021–FY2025 Forms 10-K. Meta has never published a shareholder or CEO letter, so the prepared remarks are the primary management voice. No transcript exists in the collection for the Q1 2025 call; that quarter is covered by the 10-Q narrative and the surrounding calls.

1. Core Narrative Across Years

What management always comes back to, in five years of prepared remarks, regardless of what the quarter looked like.

• Scale is stated first, always. Every call for five years opens with the number of people using at least one app daily — 2.8 billion in December 2021, 3.1 billion at Q4 2023, 3.3 billion at Q4 2024, 3.5 billion at Q4 2025, 3.6 billion at Q2 2026. The sentence structure never changes. Scale is the one asset management never has to argue for.
• A two-wave investment frame: one technology monetising now, one platform bet for later. Zuckerberg has used this structure since Q4 2022 ("the two major technological waves driving our roadmap are AI today and over the longer term the metaverse"). The frame survived; the contents of both slots were swapped — AI moved from the near slot to the far slot as superintelligence, and the metaverse was removed entirely.
• Recommendation and ranking AI is the engine of both engagement and ad price. This is the most consistently evidenced claim in the whole period and the one where management volunteers the most falsifiable numbers — 20%+ more conversions year-over-year (Q4 2022), 5% more time spent on Facebook and 6% on Instagram in a single quarter (Q2 2025), Reels at a $50bn+ annual run rate (Q3 2025).
• "We have made this transition before." Web-to-mobile, feed-to-Stories, Stories-to-Reels, and now recommendations-to-LLMs. The playbook analogy is the standing justification for accepting near-term financial pain, and it is redeployed almost verbatim in each new cycle.
• Front-load capacity ahead of proven demand. Introduced in Q4 2023 as an explicit lesson from under-building GPU clusters for Reels, retold in some form in every subsequent year, and now carrying a capital programme roughly four times its FY2025 size.
• Own the distribution layer rather than rent it. Own apps, own devices, and from FY2026 own silicon (1GW+ of Broadcom-developed custom chips, plus AMD and MTIA). This has been the quiet constant beneath every product pivot.

2. What Changed

Meaningful shifts in tone, priority, emphasis or strategic framing. Cosmetic wording differences are excluded.

2.1 Capex moved from a variable to be minimised to the central strategic claim

When: FY2023 → FY2026

Evidence. Q4 2022 (FY2023 guide): capital expenditures "$30-33 billion, lowered from our prior estimate of $34-37 billion," presented as an efficiency achievement. Q4 2025 (FY2026 guide): "$115-135 billion." Raised to $125-145bn at Q1 2026, then narrowed to $130-145bn at Q2 2026. Total expense guidance moved from $90-95bn (FY2022) to $162-169bn (FY2026).

Why it matters. The same management team that treated a $4bn capex reduction as a headline win three years ago now treats a $70bn single-year increase as table stakes. Nothing in the narrative marks that reversal as a change of view.

2.2 Cost discipline changed owner and changed altitude

When: FY2023 vs FY2026

Evidence. Q4 2022: the CEO opens with "my management theme for 2023, which is the 'year of efficiency'" and spends the first third of his remarks on it. Q1 2026: the headcount reduction appears in the CFO's expense paragraph — "we recently shared internally that we plan to reduce the size of our employee base in May… a leaner operating model will allow us to move more quickly while also helping to offset the substantial investments we're making." Q2 2026 books $1.2bn of severance for it.

Why it matters. A cost action of comparable magnitude has been demoted from a company-defining theme to a financing mechanism disclosed in passing. Investors watching only the CEO's remarks would not have registered it.

2.3 The unit of ambition moved from products to a civilisational claim

When: FY2021 → FY2025

Evidence. Q4 2021: "Today I'm going to discuss our seven major investment priorities for 2022: Reels, community messaging, commerce, ads, privacy, AI, and of course the metaverse." Q2 2025: "Developing superintelligence — which we define as AI that surpasses human intelligence in every way — we think is now in sight."

Why it matters. Product-level priorities are checkable within a year. A superintelligence timeline is not. The narrative has migrated toward claims that cannot be marked to market on any earnings call.

2.4 A genuine business-model addition: enterprise and compute

When: FY2026

Evidence. Q2 2026, CEO: "we see a large enterprise opportunity to sell to businesses, including APIs, business agents, potentially selling compute directly." And: "we're getting a lot of offers for compute at a significant premium over what we paid for it." CFO the same quarter: enterprise offerings may take the form of "agentic tools, our API, or monetizing compute directly given outsized market demand."

Why it matters. This is the first time in the period that management has narrated non-advertising revenue as strategic rather than incidental. It also implies a customer set, a sales motion and a margin profile Meta has never operated.

2.5 Financing entered the story only after capex crossed $100bn

When: FY2026

Evidence. Q1 2026: "infrastructure purchase agreements drove a $107 billion step up in our contractual commitments this quarter." Q2 2026: "the strength of our balance sheet gives us the ability to attract capital from a wide range of markets to supplement the cash flow generated by our business," alongside a BlackRock joint venture for a 1GW data centre in El Paso. Q4 2025 introduced Dina Powell McCormick as President and Vice Chairman to lead partnerships with "governments, sovereigns, and strategic capital partners."

Why it matters. For most of the period Meta funded everything from operating cash flow and still bought back stock. The arrival of third-party structures and sovereign capital is a change in the company's financial character, not just its spending.

2.6 The CFO reordered the script

When: Q3 2025

Evidence. Every call from Q1 2022 to Q2 2025 began "Let's begin with our consolidated results." From Q3 2025 to Q2 2026, every call begins "Let's begin with our segment results."

Why it matters. Observation, not inference: the change is exact and coincides with the quarter in which the Reality Labs operating loss stopped being read aloud. Segment-first ordering places Family of Apps revenue growth ahead of the consolidated expense picture.

2.7 Downside language appeared for the first time — at peak spend

When: Q3 2025 → Q2 2026

Evidence. Q3 2025: "in the worst case, we would just slow building new infrastructure for some period while we grow into what we build." Q2 2026, CFO: the distribution advantage "should be true regardless of whether our models are on the frontier."

Why it matters. Two quarters earlier the stated goal was "establishing Meta as the leading frontier AI lab" (Q3 2025). The hedge is new and it is load-bearing.

3. What Disappeared or Faded

Silence is a finding. Each item below was prominent in management's own voice and is now reduced or absent. The last column records whether the change was ever acknowledged.

3.1 The Reality Labs quarterly operating loss

Was. Read aloud by the CFO in prepared remarks every quarter from Q4 2021 through Q2 2025 — $3.3bn (Q4 2021), $4.3bn (Q4 2022), $4.6bn (Q4 2023), $5.0bn (Q4 2024), $4.5bn (Q2 2025).

Faded. Q3 2025. Zero mentions in the Q3 2025, Q4 2025, Q1 2026 and Q2 2026 prepared remarks. Only Reality Labs revenue is now stated.

Acknowledged? Not acknowledged. The FY2025 10-K discloses Reality Labs loss from operations of $19,193m, up from $17,729m in FY2024 and $16,120m in FY2023, and states that Reality Labs "reduced our 2025 overall operating profit by approximately $19.19 billion." The number did not shrink; the sentence did.

3.2 "Metaverse"

Was. 12 mentions in the Q4 2021 CEO remarks; the FY2022 10-K used the word 36 times and opened Item 1 with "All of our products, including our apps, share the vision of helping to bring the metaverse to life."

Faded. Steadily from FY2023. Zero mentions in the Q4 2025 and Q2 2026 CEO remarks; one passing mention in Q1 2026. The FY2025 10-K uses it 9 times, all in risk-factor or definitional context.

Acknowledged? Never retired on the record. The word was removed from the mission sentence in the FY2024 10-K without comment.

3.3 Llama

Was. The named model franchise from Q1 2023, peaking at 12 mentions in Q3 2024. Q4 2024: "our goal for Llama 4 is to lead."

Faded. Last mentioned Q2 2025 — "we're making good progress towards Llama 4.1 and 4.2." Zero mentions in Q3 2025, Q4 2025, Q1 2026 and Q2 2026. The Muse family (Muse Spark, Muse Image, Muse Video) appears in Q1 2026 as "our first model."

Acknowledged? Silent. No statement was made about whether Llama 4.1/4.2 shipped, were abandoned, or were folded into Muse. Q4 2025 says only that "in '25 we rebuilt the foundations of our AI program."

3.4 Open source as strategy

Was. Q4 2023 devoted eleven mentions and roughly a full page to defending it — safety, industry standards, recruiting, and the claim that open-sourcing "doesn't reduce our main advantages."

Faded. One generic mention in Q3 2025 ("our approach of advancing open source AI means that when Meta innovates, everyone benefits"), then nothing. Zero mentions in Q4 2025, Q1 2026 and Q2 2026 — the same period in which Meta launched a paid public API, the Meta One subscription, and discussed selling compute.

Acknowledged? Not acknowledged. A stated multi-year doctrine was reversed in practice without being withdrawn in words.

3.5 Business messaging as "the next pillar of our business"

Was. CEO language from Q4 2022 ("Facebook and Instagram are the first two pillars of our business, and in the next few years we hope to bring messaging online as the next pillar") through Q4 2023.

Faded. The pillar framing vanishes from CEO remarks in 2024 and 2025, surviving only as a CFO explanation for Family of Apps other revenue.

Acknowledged? Worth separating from the other silences. The revenue kept compounding — Family of Apps other revenue went from $155m in Q4 2021 to $1.0bn in Q2 2026, up 73% year-over-year. This looks like crowd-out by the AI narrative rather than a failed initiative, and the theme returns in FY2026 rebadged as "business agents."

3.6 Young adults as a stated north star

Was. Q3 2021: "we are retooling our teams to make serving young adults their north star, rather than optimizing for the larger number of older people… This shift will take years, not months." Echoed once in Q4 2021.

Faded. Effectively immediately. Two passing mentions across 2024, none since.

Acknowledged? Never mentioned again, and no cohort metric was ever introduced to measure it. A stated multi-year reorganisation of product goals simply left the narrative.

3.7 Avatars and Horizon

Was. Avatars: six mentions in Q4 2021, with an SDK, digital clothing and an NFL partnership. Horizon: eight mentions in Q1 2022; "Horizon is core to our metaverse vision."

Faded. Avatars: zero mentions from Q4 2023 onward. Horizon: near-zero from FY2024, reduced by Q4 2025 to "making Horizon a massive success on mobile."

Acknowledged? No acknowledgement. Horizon survives as a mobile ambition; avatars are gone.

3.8 Reels as a headline priority

Was. 24 mentions in Q2 2022; the single most-discussed product of FY2022–FY2023.

Faded. Two to five mentions per call in FY2026.

Acknowledged? This one is benign and partially explained: Reels was absorbed into the broader "recommendation systems" narrative, and Q3 2025 disclosed a $50bn+ annual run rate. The theme faded because it worked, not because it failed — a useful control case for reading the other silences.

4. New Themes and New Emphasis

4.1 Superintelligence

First appeared. Q2 2025, with nine mentions in its first appearance and a definition supplied on the spot: "AI that surpasses human intelligence in every way."

How fast it became central. Immediately central. Meta Superintelligence Labs was announced the same quarter with three named leaders (Alexandr Wang, Nat Friedman, Shengjia Zhao). By Q3 2025 the stated corporate goal was "establishing Meta as the leading frontier AI lab." The FY2025 10-K uses the word nine times; no prior 10-K uses it at all.

Real shift or framing? Both. The organisational change, the named hires and the multi-gigawatt cluster commitments (Prometheus, Hyperion) are real and expensive. But the word also arrived precisely when the Llama franchise stopped being named, which makes it a replacement narrative as well as a new one.

4.2 Agents

First appeared. Scattered mentions from FY2023, but the frame arrives at Q1 2026 (10 mentions) and dominates Q2 2026 (28 mentions).

How fast it became central. One quarter from periphery to primary product frame. Q2 2026 splits it into personal agents, business agents, and an eventual "business-in-a-box."

Real shift or framing? Partly evidenced. Q2 2026 discloses more than one million businesses using Meta Business Agents weekly and 10x growth in weekly business-AI conversations since the start of 2026. Personal agents, by contrast, are still described in the future tense.

4.3 Compute as an asset class and a product

First appeared. Gigawatt language enters at Q4 2024 ("almost 1GW of capacity this year"); "Meta Compute" as a named initiative arrives at Q4 2025.

How fast it became central. Two years from a cost line to a stated source of strategic advantage — "being the most efficient at how we engineer, invest, and partner to build our infrastructure will become a strategic advantage" (Q4 2025) — and then to a potential revenue line by Q2 2026.

Real shift or framing? Real business shift. It comes with custom silicon (1GW+ with Broadcom, plus AMD and MTIA), external financing structures, and a new senior executive owning sovereign and strategic capital.

4.4 AI glasses

First appeared. Two mentions in Q3 2021, then near-silence; sustained presence from Q3 2023 (8 mentions) onward.

How fast it became central. Gradual and then dominant. By Q4 2025 glasses are the stated hardware endgame — "it's hard to imagine a world in several years where most glasses that people wear aren't AI glasses" — having displaced headsets. The FY2025 10-K uses "glasses" 14 times; the FY2021 and FY2022 10-Ks use it zero and once.

Real shift or framing? Real, and among the better-evidenced new themes. Q4 2025: sales "more than tripled last year." Q1 2026: daily users tripling year-over-year. Reality Labs revenue growth is now attributed to glasses offsetting declining Quest sales.

4.5 Revenue outside advertising

First appeared. FY2026. Meta One subscription (Q2 2026), WhatsApp paid messaging, a public API for Muse Spark, and enterprise/compute sales.

How fast it became central. One year.

Real shift or framing? Early but not rhetorical. Family of Apps other revenue reached $1.0bn in Q2 2026, up 73%. The enterprise and compute pieces remain claims rather than disclosed revenue.

5. Tone Shift

Overall arc: More defensive (FY2021–FY2022) → more operationally focused (FY2023) → more confident (FY2024) → more promotional (FY2025) → confident but newly hedged (FY2026).

FY2021 is cautious and explanatory. The Q4 2021 call names competition, Apple's iOS changes and a self-imposed shift toward short-form video as headwinds, and asks investors to accept "the right short-term tradeoff… in order to get long-term gains."

FY2022 is the low point and the most defensive. Total revenue fell for the first time in the company's history, from $117,929m in 2021 to $116,609m in 2022. Q4 2022 opens with "2022 was a challenging year" and pivots to cost — "the year of efficiency," fifteen uses of the word in one call, alongside an 11,000-person layoff and a lowered capex guide.

FY2023 is operational and self-congratulatory in a controlled way. Q4 2023: "not only did we achieve our efficiency goals, but we returned to strong revenue growth." The most substantive strategic argument of the entire period — the open-source defence — is made here, in a tone of explanation rather than assertion.

FY2024 is confident and deadline-driven. Q4 2024: "we have about 48 weeks to get on the trajectory that we want to be on," with specific falsifiable predictions about Meta AI reaching a billion users, Llama 4 leading, and an AI engineering agent at mid-level-engineer capability.

FY2025 is the most promotional. Q2 2025 opens with "we have begun to see glimpses of our AI systems improving themselves" and a superintelligence definition. Talent language spikes (nine mentions in Q2 2025) and the remarks are structured around vision rather than results.

FY2026 is confident on the core business and, for the first time, hedged on the thesis. The specificity is now concentrated where the evidence is strongest — ad conversions, time spent, glasses units — while the largest claims (personal agents, superintelligence) carry the least disclosure. The hedges are new and precise: "in the worst case, we would just slow building" (Q3 2025) and "regardless of whether our models are on the frontier" (Q2 2026).

Tone by fiscal year, scored on the prepared remarks:

6. Reframed Problems

The same underlying reality, described in materially different language across years. This section carries the most weight for an investor because reframing is how narrative and business reality diverge.

6.1 "Efficiency"

Before — Q4 2022 — earnings call

"we've taken some additional steps like working with our infrastructure team on how to deliver our roadmap while spending less on capex… we're working on flattening our org structure and removing some layers of middle management."

Frame: Efficiency = spend less.

After — Q1 2026 — earnings call

"we are increasing our infrastructure capex forecast for this year… That said, we are very focused on increasing the efficiency of our investments." And: "one of the primary goals of our Meta Compute initiative is to lead the industry in efficiency of building compute."

Frame: Efficiency = better cost per unit of compute, while total spend rises.

Likely underlying reality. The word survived a complete inversion of its cash-flow meaning. In FY2023 efficiency meant a capex guide cut to $30-33bn. In FY2026 efficiency accompanies a capex guide of $130-145bn. Both uses are internally coherent; used across years without comment, the continuity of the word obscures the discontinuity of the policy.

Does the reframing obscure or clarify? Obscures.

6.2 The metaverse

Before — FY2022 Form 10-K, Item 1

"All of our products, including our apps, share the vision of helping to bring the metaverse to life… Meta is moving our offerings beyond 2D screens toward immersive experiences like augmented and virtual reality to help build the metaverse, which we believe is the next evolution in social technology."

Frame: The metaverse is the company's organising purpose.

After — FY2025 Form 10-K, Item 1

"Our products enable people to connect and share through mobile devices, personal computers, virtual reality (VR) headsets, and AI glasses. We are innovating in artificial intelligence (AI) technologies… and to advance our vision to deliver personal superintelligence for everyone."

Frame: AI and glasses are the organising purpose; the metaverse survives only in risk factors.

Likely underlying reality. The FY2024 10-K performed the swap at sentence level — the identical construction "to help build the metaverse" became "to help build the next computing platform." The same 10-K also rewrote the corporate mission, from "give people the power to build community and bring the world closer together" (FY2016–FY2023) to "build the future of human connection and the technology that makes it possible." Reality Labs, the segment created to build the metaverse, has since been redirected — Q4 2025: "we're directing most of our investment towards glasses and wearables going forward."

Does the reframing obscure or clarify? Clarifies the direction, obscures the cost. The strategy genuinely changed; what is missing is any accounting of what roughly $77bn of cumulative Reality Labs operating losses from FY2021 to FY2025 bought ($10.2bn, $13.7bn, $16.1bn, $17.7bn and $19.2bn respectively).

6.3 Open source and the model franchise

Before — Q4 2023 — earnings call

"our long-standing strategy has been to build and open source general infrastructure while keeping our specific product implementations proprietary… we find that there are mostly advantages to being the open source leader and it doesn't remove differentiation from our products much anyway."

Frame: Open source is a durable competitive strategy with named benefits in safety, standards and recruiting.

After — Q2 2026 — earnings call

"Muse Spark 1.1 is a strong agentic and coding model… It's available through our new public API, and we're ramping up distribution through partner channels… We're also building out features to make it easier for enterprises to adopt Muse Spark."

Frame: The model franchise is a renamed, commercialised, API-gated product.

Likely underlying reality. Llama was last named in Q2 2025; open source was last named in Q3 2025; Muse appeared in Q1 2026. Across four calls a stated strategic doctrine and a named product line both left the record without a single sentence marking their departure. Q4 2025 offers the only oblique reference: "in '25 we rebuilt the foundations of our AI program."

Does the reframing obscure or clarify? Obscures. This is the clearest instance in the period of a commitment being abandoned in practice while remaining unretracted in words.

6.4 Headcount reduction

Before — Q4 2022 — earnings call, CEO

"I want to discuss my management theme for 2023, which is the 'year of efficiency'… even if our business outperforms our goals, this will stay our management theme for the year."

Frame: A company-defining commitment, owned by the CEO, announced first.

After — Q1 2026 — earnings call, CFO

"As we grow our infrastructure spend, we remain committed to operating efficiently, and we recently shared internally that we plan to reduce the size of our employee base in May… a leaner operating model will allow us to move more quickly while also helping to offset the substantial investments we're making."

Frame: A funding offset, disclosed by the CFO inside the expense paragraph.

Likely underlying reality. Q2 2026 booked $1.2bn of severance for the May 2026 action — a real, material event. The change is not in what was done but in who said it, where it sat in the script, and what it was said to be for. Headcount reduction is now explicitly framed as a mechanism for funding capex.

Does the reframing obscure or clarify? Obscures. An investor tracking only the CEO's remarks would have missed a billion-dollar action entirely.

6.5 Frontier leadership

Before — Q3 2025 — earnings call, CEO

"I'm very focused on establishing Meta as the leading frontier AI lab — building personal superintelligence for everyone."

Frame: Winning the model race is the goal.

After — Q2 2026 — earnings call, CFO

"our distribution advantages will give us the opportunity to serve AI products that are valuable for everyone… This should be true regardless of whether our models are on the frontier."

Frame: Distribution, not model leadership, is the durable advantage.

Likely underlying reality. Three quarters separate these two statements. The second is a materially lower bar and was introduced without reference to the first. It may well be the more defensible position — Meta's distribution advantage is real and its model position is not obviously leading — but the shift was made quietly and by the CFO.

Does the reframing obscure or clarify? Clarifies the economics, obscures the retreat.

7. Investor Interpretation

Observations — directly stated or disclosed

• Revenue grew from $134,902m (FY2023) to $164,501m (FY2024) to $200,966m (FY2025). Q2 2026 revenue was $60.8bn, up 28%.
• Reality Labs loss from operations rose in every year of the period: $16,120m (FY2023), $17,729m (FY2024), $19,193m (FY2025). Management stopped stating the quarterly figure in prepared remarks after Q2 2025.
• Capex guidance rose from $29-34bn (FY2022) to $130-145bn (FY2026). Total expense guidance rose from $90-95bn to $162-169bn over the same span.
• Six distinct themes — metaverse, Llama, open source, business messaging as a pillar, young adults, and the Reality Labs quarterly loss — left the narrative without being formally retired. No theme in the period was ever explicitly withdrawn on the record.
• New themes enter loudly and with numbers. Superintelligence arrived with an org chart and named hires; agents arrived with a business-count; glasses arrived with growth rates.

Interpretation — inference from the pattern

The evidence supports a mixed reading rather than a single verdict. There is real business progress, a real strategic transition, and material narrative management running simultaneously — and they are easy to confuse for one another.

The progress is not in doubt. The advertising claims are the most specific and most repeated statements in the entire corpus, they are made quarter after quarter with different numbers, and the reported revenue line is consistent with them. Glasses and Threads have disclosed traction. Management's most checkable claims are the ones that have held up.

The transition is real too. Meta is becoming a different kind of company: a consumer advertising business bolted to a capital-intensive compute platform with enterprise ambitions. The evidence for this is structural rather than rhetorical — off-balance-sheet financing structures, a new President for sovereign capital, a custom silicon programme, a reordered CFO script, and headcount reductions explicitly framed as a capex offset.

The narrative management is the part that should trouble a long-term holder, and it is a pattern rather than an incident. Themes that stop working exit in silence; themes that start working enter with fanfare. Over five years and nineteen calls, management has not once said on the record that a prior commitment was wrong. That asymmetry is a durable feature of how this company communicates, and it means the absence of a theme carries more information than its presence.

One interpretation of the Reality Labs silence is benign — the segment is being redirected toward glasses, which are working, and the quarterly loss is a distraction from that. The competing interpretation is that a $19.19bn annual loss became inconvenient to say aloud at the exact moment the company began asking investors to fund a much larger bet. The Q3 2025 script reordering is consistent with both, which is why it should be watched rather than concluded from.

The strongest evidence for management quality in the period is the Reels under-build story: a specific, admitted mistake, converted into a stated policy of over-building capacity, retold annually, and so far vindicated by ad-revenue growth. That is a genuine credibility asset. It is now carrying a capital programme roughly four times larger than the one it was earned on, with no comparable proof point yet — and the appearance of downside language in Q3 2025 and Q2 2026 suggests management knows it.

What an investor should weight most heavily, in order: first, whether the Reality Labs loss trajectory actually inflects in FY2026 as promised, since that is the one place where a stated commitment and a disclosed number can be checked against each other; second, whether enterprise and compute revenue becomes a disclosed line rather than a claim; third, whether the model franchise, having been renamed once, is named consistently going forward. The narrative record suggests all three will be reported enthusiastically if they succeed and quietly if they do not — so the discipline required is to look for the absence, not the announcement.

8. Follow-Up Questions

Five questions that arise directly from the narrative record and can be answered from disclosure, management access, or the next two reporting periods.

Q1. What happened in Q3 2025 that removed the Reality Labs quarterly operating loss from prepared remarks — and what exactly does "losses will peak" mean?

The loss was stated every quarter from Q4 2021 through Q2 2025 ($4.5bn) and has not been stated since, in the same quarter the CFO reordered the script from consolidated-first to segment-first. Meanwhile the FY2025 10-K discloses a $19,193m annual loss and guides to 2026 losses "similar to 2025." Q4 2025 called 2026 "likely the peak." Establish whether that means peak absolute dollars — which is checkable — or peak as a share of a much larger expense base, which is not the same claim.

Q2. Were Llama 4.1 and 4.2 shipped, abandoned, or absorbed into Muse?

Q2 2025 said Meta was "making good progress towards Llama 4.1 and 4.2." The name has not been spoken since. Muse Spark was introduced in Q1 2026 as "our first model" from Meta Superintelligence Labs. A named model franchise with a stated roadmap disappeared across two quarters with no explanation beyond "in '25 we rebuilt the foundations of our AI program." The answer bears on how much of the $19bn-plus of AI R&D in that window produced shippable output.

Q3. Has the open-source commitment been retired, and what replaces the benefits it was said to deliver?

The Q4 2023 defence rested on three specific claims: open source improves safety and compute efficiency through community scrutiny, it makes Meta's stack an industry standard, and it is a recruiting advantage. Meta now sells API access, subscriptions, and possibly compute. If the doctrine is gone, each of those three benefits needs a replacement, and the recruiting one matters most given how much of the FY2025 narrative was about talent density.

Q4. Is headcount reduction now a permanent funding mechanism for capex, and what is the operating expense trajectory excluding infrastructure?

The May 2026 action cost $1.2bn in severance and was framed by the CFO as "helping to offset the substantial investments we're making." With total expense guidance at $162-169bn and capex at $130-145bn, the split between infrastructure and everything else determines whether the core business is actually operating leaner or simply being harvested. This is checkable from the segment and expense-line disclosures and is not addressed in the narrative.

Q5. What is behind "offers for compute at a significant premium over what we paid for it," and how much of FY2026 capex is contractually committed versus optional?

Q2 2026 made an unusually strong claim about external demand for Meta's compute without naming size, duration or counterparties. In the same period, Q1 2026 disclosed a $107bn step-up in contractual commitments and Q2 2026 announced a BlackRock joint venture. Management describes the long-lived nature of the assets as providing flexibility "to adjust our investment to the pace of AI adoption" — the commitment schedule is the test of whether that flexibility is real.