Five Reasons Meta's Layoffs Do Not Fix the Capex Problem

Capital expenditure went from $37.3 billion in 2024 to $69.7 billion in 2025. That is an $32.4 billion increase in a single twelve-month window. The layoff announcement is sized to save approximately 5-7% of operating expense over 12-18 months; on a total opex base of approximately $117 billion, that is $6-8 billion of cumulative savings, or $2-3 billion annually on a steady-state basis.

The mismatch is what the Risk Desk reads. Even in an optimistic layoff-savings scenario, the annual capex step is 10-15x larger than the annual opex benefit. The layoffs are directionally positive for reported operating income but they do not meaningfully alter the capital-intensity trajectory of the business. That trajectory is the one that has compressed free cash flow in 2025 and will continue to shape the FY2026-FY2027 cash generation profile.

The historical parallel is cloud infrastructure build-outs from 2014-2019, where Alphabet and Microsoft grew capex faster than opex for multiple years before the infrastructure investment began contributing to operating income at scale. The lag between capex peak and incremental operating income contribution ran four to six years in those cases. Meta's 2025 capex step is the start of an analogous cycle, not the end of one.

One further detail on the capex comparison. The majority of the 2025 capex step was directed toward AI infrastructure (specifically GPU-heavy compute and data centre power expansion). That spend category has a longer depreciation life than prior capex cycles and typically begins contributing to operating income 18-24 months post-deployment. That timing gap is embedded in the 2026 and 2027 free cash flow trajectories.

Operating income grew 20% in 2025. Free cash flow fell 15%. That divergence is entirely explained by the capex step. In a stable capital-intensity regime, operating income and free cash flow would compound together. In a rising capital-intensity regime, they diverge. Meta is firmly in the second regime.

For investors, the distinction matters. Earnings multiples price operating income. Cash flow yields price free cash flow. Meta's 29x trailing P/E looks reasonable on operating income. The implied free cash flow yield of 2.6% on trailing FCF looks more demanding. Against AI-infrastructure peers, Meta's FCF yield is now tracking below Alphabet and Microsoft on comparable revenue bases.

The Risk Desk's read is that the FCF yield compression is a structural signal, not a cyclical one. Every dollar of incremental capex produces a multi-year return profile rather than an immediate revenue contribution. The market is paying for the revenue first and hoping the cash flow follows. That is a sequencing assumption worth testing against the history of large infrastructure investments in tech.

The Reality Labs segment recorded an operating loss of approximately $17-19 billion in 2025, broadly in line with prior years. The cumulative operating losses from the segment since 2020 now exceed $75 billion. Revenue from the segment has grown but remains small relative to the loss rate.

The AI push is being positioned as a partial substitute for the Reality Labs framing. That framing obscures the cost structure. Reality Labs continues to consume capital, the AI infrastructure build-out is layered on top, and the combined investment envelope is substantially larger than the corresponding revenue stream. The argument for continued funding is that either or both segments produce a winner-takes-all outcome with sufficient optionality to justify the current capital cost.

That is a defensible argument. It is also a high-variance argument. Historical examples of large-cap tech companies funding parallel moonshot segments with this capital intensity for more than five consecutive years are scarce. The cases that worked (Google Cloud, AWS) each required a structural demand shift to justify the capital commitment. The cases that did not (various VR-AR bets over the last two decades) required write-downs that absorbed years of accumulated investment.

The Risk Desk does not predict Reality Labs is a failed bet. The position is that the combined investment envelope across Reality Labs plus AI infrastructure is approaching a scale where disappointment on either segment produces a material downside to the multiple. That risk is not priced.

Meta's advertising revenue grew approximately 22% in 2025 against 25% in 2024. The deceleration is modest but consistent with broader digital advertising trends; the post-reopening reset that drove the 2023-2024 cycle's peak growth rates is rolling off, and the forward comparable base is higher.

The consensus FY2026 revenue growth forecast sits at 18-19%. If the deceleration continues at the current pace, FY2027 growth lands at 15-16%. The implied mix of growth and capex intensity then looks different: at 15% revenue growth against capex of $80-90 billion, the FCF yield compression extends another two to three years before stabilising.

The advertising revenue line is the anchor for the entire investment case. If advertising revenue growth decelerates faster than the consensus base, the capex envelope becomes materially harder to justify. Advertiser spending is broadly sensitive to macroeconomic conditions; a recessionary cycle would compress revenue growth into the single-digit range and immediately turn the capex story from optional to burden.

Historically, large-cap advertising-revenue businesses have entered three to five year deceleration phases at various points in their maturity curves. Each deceleration phase has been accompanied by a multiple compression of 100-300 basis points. The Risk Desk reads the current set-up as the entry point of such a phase, though the probability and timing are naturally uncertain.

Advertising revenue is also sensitive to election-cycle advertising spend, which was elevated in 2024. FY2025 grew despite losing that tailwind, which is modestly encouraging. FY2026 faces a similar base-effect drag and a continued deceleration in comparable-quarter growth. The combination makes the 18% consensus for FY2026 growth look modestly ambitious; the Risk Desk's base case is 16%, with the downside case at 14%.

The AI push is the single largest strategic initiative in the current capex programme. Meta's positioning centres on AI-augmented advertising targeting, creator tools, and platform personalisation. The incremental revenue contribution from these initiatives has been referenced in commentary but has not yet been quantified in the reporting package at a level of detail that supports independent modelling.

The Risk Desk read is that the incremental AI revenue is partially baked into the 2025 print through advertising revenue acceleration attributable to targeting improvements. The challenge is isolating the AI-specific contribution from the broader advertising trend. Until management provides explicit AI-revenue disclosures, the investment case requires the investor to accept the narrative at face value.

Historical base rates across the last cycle of major AI-related capex in large-cap tech (Azure AI, Google Cloud AI, Amazon Bedrock) suggest that the incremental revenue from AI investments typically matures 3-5 years after the initial infrastructure build. On that timeline, Meta's 2025 capex contributes to FY2028-FY2030 revenue rather than FY2026-FY2027 revenue. The intervening period is one of compressed free cash flow without commensurate revenue benefit.

That timing mismatch is the bearish part of the AI capex story. The bullish part is that the eventual revenue contribution can be large enough to justify the capital cost. Both parts are simultaneously true. The question is how the market prices the intervening period.

Compare this set-up against Amazon's 2015-2019 AWS investment cycle. Amazon absorbed four years of heavy capital investment before the market fully re-rated the stock in recognition of the incremental cash flow stream. The intervening period produced a trading range of $30-40% around the eventual cycle-trend. Meta's current trajectory implies a similar intervening period is likely, though the outcome distribution is harder to characterise because the AI infrastructure payoff timing is itself uncertain.

Compare the capital intensity profile against Alphabet and Microsoft across 2025. Alphabet's capex was approximately $75 billion on a revenue base of roughly $350 billion, a capex-to-revenue ratio of 21%. Microsoft's was approximately $65 billion on revenue of $305 billion, a ratio of 21%. Meta's $69.7 billion on $201 billion revenue gives a ratio of 35%.

The 14 percentage point gap is the core risk argument. Alphabet and Microsoft both have cash flow-generating cloud businesses that absorb the capital investment with identifiable revenue streams. Meta's AI infrastructure build-out serves an advertising revenue base that is already well-monetised; the incremental revenue is an incremental improvement on an existing base, not a new business line that compounds independently.

That structural distinction is worth pricing. Both Alphabet and Microsoft trade at forward multiples similar to or lower than Meta's (22x and 28x respectively). The fact that Meta's multiple is close to Microsoft's while its capex-to-revenue ratio is 14 percentage points higher is the optical mispricing the Risk Desk reads. Meta either needs to bring capex down faster than peers or accept a multiple compression as the market normalises the capital intensity comparison.