In 2023, Meta cut costs to build a war chest. Starting in 2024, it began betting that war chest on AI — and three years later, it keeps raising the stakes, calling each new figure more than it thought it would need.
The story
2023: building the war chest
In early 2023, after large layoffs, Meta called that year its "Year of Efficiency." On the February 2024 earnings call, Zuckerberg looked back and framed it as preparation "to become a stronger technology company and to execute efficiently towards our ambitions around AI and the metaverse" — the efficiency drive wasn't the goal, it was the fundraising step for the next bet.
The pivot: "build ahead of need"
The turning point came in July 2024. Zuckerberg said the compute needed to train the next model (Llama 4) would be roughly 10 times that of its predecessor, and declared: "I'd rather risk building capacity we don't end up needing... than build too late." From that point, capex guidance rose on every single call through the rest of 2024 — from $30–37B to $35–40B to $37–40B to $38–40B — landing at an actual $39.23B, near the top of the final range.
How Meta frames its own risk
The FY2023 10-K's risk factor about margin-reducing investments named only Reality Labs. The FY2024 10-K added "our artificial intelligence initiatives" alongside it for the first time. Then, in the FY2025 10-K (filed January 2026), a closing sentence template previously used only for the Reality Labs risk factor — "our AI strategy and investments may not be successful in the foreseeable future, or at all" — was applied to the AI risk factor for the first time. Meanwhile, in the opposite direction, Reality Labs loss language actually softened over the same period: "increase meaningfully in 2024" (FY2023 10-K) → "increase in 2025" ("meaningfully" dropped, FY2024 10-K) → "remain similar to 2025" (FY2025 10-K, for 2026). The company's newly-serious language has shifted from Reality Labs to AI investment itself.
"Superintelligence": a word that vanished, then came back
In July 2025, "Meta Superintelligence Labs" was announced as a new organization, and the word "superintelligence" appeared 6 to 20 times on every call that followed. Then, on the April 2026 call, the word was used zero times — and in the same call, 2026 capex guidance was raised again, from $115–135B to $125–145B. Even though revenue beat analyst estimates, the stock fell as much as 10% intraday. Three months later, on the July 2026 call, guidance narrowed to $130–145B (the floor moved up) — and "superintelligence" language returned.
CapEx and revenue guidance scorecards
CapEx guidance: three years, never cut
| Target year | Initial guidance | Final guidance | Actual | Pattern |
|---|---|---|---|---|
| 2024 | $30–37B (Q4 2023 call) | $38–40B (Q3 2024 call, 4 straight raises) | $39.23B | Raised only |
| 2025 | $60–65B (Q4 2024 call) | $70–72B (Q3 2025 call, 3 straight raises) | $72.22B | Raised only, actual exceeded even the final guide |
| 2026 (ongoing) | $115–135B (Q4 2025 call) | $130–145B (Q2 2026 call, 2 raises so far) | TBD | Still rising |
Across three years and nine quarterly updates, not one was ever a cut. Actual spending both years landed near or above the top of the last guidance given — the company has consistently underestimated its own AI investment pace.
Revenue guidance: nearly always beaten
| Guided | Target quarter | Range | Actual | Result |
|---|---|---|---|---|
| Q4 2023 call | Q1 2024 | $34.5–37.0B | $36.5B | Top of range |
| Q1 2024 call | Q2 2024 | $36.5–39.0B | $39.1B | Above high end |
| Q2 2024 call | Q3 2024 | $38.5–41.0B | $40.6B | Top of range |
| Q3 2024 call | Q4 2024 | $45.0–48.0B | $48.4B | Above high end |
| Q4 2024 call | Q1 2025 | $39.5–41.8B | $42.3B | Above high end |
| Q1 2025 call | Q2 2025 | $42.5–45.5B | $47.5B | Well above high end |
| Q2 2025 call | Q3 2025 | $47.5–50.5B | $51.2B | Above high end |
| Q3 2025 call | Q4 2025 | $56.0–59.0B | ~$59.9B | Above high end |
| Q4 2025 call | Q1 2026 | $53.5–56.5B | $56.3B | Top of range |
| Q1 2026 call | Q2 2026 | $58.0–61.0B | $60.8B | Top of range |
All 10 quarters landed at the top of the guided range or above it — never once in the middle or low end.
Source: earnings call transcripts, Q3 FY2023 through Q2 FY2026 (roic.ai), prepared remarks from Mark Zuckerberg and Susan Li; stock price reaction from Yahoo Finance (Apr 30, 2026).
Timeline
- Feb 2024"Year of Efficiency" declared complete, framed as preparation for AI and metaverse investment.
- Jul 2024"Build ahead of need" philosophy declared; capex guidance begins climbing every quarter.
- Jan 2025FY2024 10-K filed: "AI initiatives" named alongside Reality Labs as a margin-reducing investment for the first time.
- Jul 2025Meta Superintelligence Labs launched; "superintelligence" becomes the company's core vision language.
- Oct 2025$15.93B one-time tax charge disclosed; adjusted net income figure pre-announced defensively.
- Jan 2026FY2025 10-K filed: new closing language warns the AI strategy "may not be successful... or at all," mirroring language previously reserved for Reality Labs.
- Apr 2026Results beat estimates, but the stock falls as much as 10% intraday after 2026 capex guidance is raised again to $125–145B. "Superintelligence" isn't mentioned once on this call.
- Jul 2026Capex guidance narrows to $130–145B (floor raised); "superintelligence" language returns.
Our read
Revenue beat guidance in nearly every one of these quarters — but spending grew even faster, every time. Neither Meta nor the market has ever said "this is enough." Reality Labs' losses stopped getting worse for the first time in this period, and an unproven new worry — AI capex payoff — has taken its place. Whether "superintelligence" briefly disappearing in April 2026 was a deliberate tone adjustment for a quarter that landed badly with the market, or just a coincidence of prepared remarks, isn't something we can say for certain — but the timing, a cooler market reaction paired with more measured language, and a warmer one paired with the word's return, is at least suggestive.
What we still don't know
- Whether this scale of AI capex actually converts to proportional revenue in 2026–2027 isn't something these filings settle — the company hasn't given a specific ROI timeline.
- Whether the "zero superintelligence mentions" quarter was a deliberate tone shift or coincidence needs more quarters to judge with confidence.
- Whether Reality Labs losses truly hold flat near 2025 levels, or get revised upward again, will only be clear once 2026 results are in.
- How analysts' repeated questions (e.g., on AI payback timing) have evolved across calls wasn't analyzed in depth in this pass — a full Q&A review could sharpen this further.
Frequently asked questions
What was Meta's 'Year of Efficiency'?
In 2023, Meta cut costs and headcount to build up a cash war chest, branding it the "Year of Efficiency" — a deliberate pause before betting that same cash on AI infrastructure starting in 2024.
Has Meta's AI spending kept rising?
Yes — starting in 2024, Meta has raised its AI capital-expenditure guidance nearly every quarter, each time framing the new figure as more than it originally thought it would need.
What sources does this analysis draw from?
This piece is built from Meta's 10-K filings for FY2023 through FY2025 and 12 quarters of earnings call transcripts from October 2023 to July 2026.