Meta’s Road to $145 Billion, Told in 6 Earnings Calls. A genuine Capex cycle needed to stay competitive in the next 5 years.

An analysis of Meta’s official earnings call transcripts from Q4 2024 through Q1 2026 — six quarters, roughly 90,000 words of prepared remarks and Q&A — tracking not just what management said, but how they said it. All quotes and figures come exclusively from Meta’s official transcripts and SEC filings at investor.atmeta.com.

Numbers get revised once a quarter. Language drifts continuously — and it usually drifts before the numbers do. When a CFO stops saying “disciplined” and starts saying “aggressive,” when a flagship product name quietly disappears from prepared remarks, when the same hedge sentence gets copy-pasted three quarters in a row, these are signals worth tracking.

RiskNot one downward revision in six quarters. Every time management had the chance to update a full-year capex range, it went up. The stated reasons rotated — more compute demand, faster data center buildout, and most recently “higher component costs, particularly memory pricing” — but the direction never did.

Call	FY Capex Guide	Change vs. Prior Guide
Q4 2025 (Jan 28, 2026)	$115–135B	Raised
Q1 2026 (Apr 29, 2026)	$125–145B	Raised, +$10B both ends

Note: management disclosed guidance revisions across all six calls in the sample without a single downward move; only the two most recent figures are quoted verbatim here, since earlier-quarter numbers were shown only as a chart in the original source rather than stated in text.

The practical implication for anyone modeling Meta: treat each quarter’s capex guide not as a point estimate but as a floor with drift. Management’s own track record makes a further raise at the Q2 2026 call (July 29, 2026) the base case, not the surprise scenario.

In January 2025, Meta’s language still carried the DNA of the 2023 “Year of Efficiency”: server-life extensions, custom silicon savings, cost per unit of compute. Capex was big but framed as disciplined.

StrengthThen Q2 2025 happened. Mark Zuckerberg introduced two phrases that had never appeared on a Meta earnings call before: “personal superintelligence” and Meta Superintelligence Labs, complete with named star hires. From that call onward, the framing of spending shifted from ROI language to opportunity-cost language — by Q3 2025 the CEO was openly saying the right strategy was to aggressively front-load capacity given the uncertainty about when superintelligence arrives.

The right strategy is to aggressively front-load capacity.

Track one word across the sample: “disciplined.” Prominent in early 2025, nearly extinct by late 2025, then reincarnated in a completely different body in 2026 — no longer describing infrastructure spend, but describing headcount.

The Q1 2026 call (April 29, 2026) delivered, in a single hour, five things pointing in opposite directions at once:

• RiskThe largest capex raise of the entire sample: $125–145B for FY2026.
• RiskAn announced workforce reduction — “reduce our employee base in May” — framed as a move toward a leaner operating model.
• StrengthThe launch of Muse, the first shipped model family from Superintelligence Labs.
• RiskAn $8 billion tax benefit partially reversing the $15.9 billion charge from just two quarters earlier.
• RiskA disclosed $107 billion step-up in contractual commitments — a forward liability that got exactly one mention.

Record spending on machines, fewer humans, and a CEO simultaneously stating the company’s AI philosophy is about “empowering individuals.” Nobody on the call directly asked management to reconcile the layoffs with the previously-cited 30% productivity gain from internal AI coding tools — whether the reduction is AI-driven restructuring or plain cost discipline is, in our view, the single most interesting open question for the next call.

For years, Meta’s capital structure was the least interesting slide in the deck: massive net cash, metronomic buybacks. That ended in Q4 2025, and it ended fast.

• RiskNet debt jumped from a stable ~$28.8bn — a figure that hadn’t moved for four straight quarters — to $58.7bn.
• RiskShare repurchases for the quarter: zero, confirmed directly when an analyst asked.
• CFO Susan Li, for the first time in this sample, floated the idea that Meta could eventually maintain a positive net debt balance — a genuine break from a decade of net-cash identity.
• A new executive hire (Dina Powell McCormick, President & Vice Chairman) was announced with an explicit mandate around capital partnerships — sovereign wealth and strategic co-investors for data centers, following the Blue Owl joint-venture template from Q3 2025.

Individually, each item is defensible. Together, they describe a company transitioning from self-funded compounder to levered infrastructure builder — with remarkably little sell-side pushback on the calls themselves. Buyback yield, which mattered to the Meta bull case for years, is currently zero and unmentioned in prepared remarks.

Llama

RiskIn January 2025, Llama 4 was the centerpiece — the model that would “lead,” the answer to DeepSeek. By the Q3 2025 call, the Llama name had faded from prepared remarks; by Q1 2026 it was fully superseded by the new “Muse” branding. Crucially, the open-source commitment reaffirmed quarter after quarter in the Llama era was not reaffirmed in the Q1 2026 call. Whether Muse continues the open-weights strategy or marks a quiet turn toward closed models is, in our view, one of the most underpriced narrative questions in the stock.

Manus

RiskThe acquisition was flagged by the CEO as strategically relevant on the Q4 2025 call. One quarter later, the entire update was a variant of “still working through the details.” In six quarters of transcripts, it is one of the shortest answers given to any direct question. Deals that go from CEO-mentioned to CFO-deflected in ninety days have a way of resurfacing — restructured, repriced, or abandoned.

One more pattern worth flagging. Meta’s disclosure about the EU Digital Markets Act and its “Less Personalized Ads” offering used structurally identical language in three consecutive quarters (Q1, Q2, Q3 2025): the Commission could force changes, producing a “materially worse user experience” and a significant negative revenue impact “as early as” the next quarter.

Then in Q4 2025, the language flipped to “aligned” with the Commission — and the story ended. What never appeared, in any quarter: an actual euro figure for the impact that did or didn’t materialize. Three quarters of maximal legal hedging resolved into zero quantification. If European revenue growth in 2026 shows any unexplained softness, this is the first footnote to revisit.

Meta reports Q2 2026 on July 29. Our checklist, derived entirely from the language patterns above:

1. Capex: is $125–145bn raised again? (Pattern says yes.)
2. Buybacks: still zero? Does anyone force a real answer on the “positive net debt” comment?
3. Muse: open weights or closed? The phrase to listen for is “open source” — its presence or absence will be the tell.
4. The May layoffs: framed as AI-productivity harvesting, or plain cost cutting?
5. Manus: an update, or another eleven-word answer?
6. EU revenue: any quantification, at last, of the Less Personalized Ads impact?

The remarkable thing about Meta right now is that the advertising engine gives management a degree of narrative freedom — Q1 2026 ad impressions grew 19% and pricing 12%, a broad-based re-acceleration. As long as that machine runs, the market forgives the vocabulary shifts.