Study · AI Infrastructure · July 2026
The AI Capex Reality Check
Data Center Alley covers the supply side of AEA's AI-infrastructure thesis — the physical corridor, the power constraint, the companies building it. This is the demand-side companion: are the four hyperscalers actually paying for all of it out of cash they're generating, or out of a widening gap between what they're spending and what they're bringing in? Two of AEA's own holdings, GOOGL and META, are inside this exact number.
1. The four numbers
Every one of the four hyperscalers raised full-year 2026 capex guidance at least once already this year. Combined, they're now guiding to roughly $725 billion for the year — up about 77% from approximately $410 billion in 2025, per Tom's Hardware's aggregation of company guidance. That $725B figure covers only these four companies; when 2026 AI-infrastructure spending sector-wide is counted — including projects like Stargate and other cloud/chip players — ValueAdd VC's tracker puts total 2026 spending above $1 trillion already, not as a 2027 projection.
| Company | 2026 capex guidance | Note |
|---|---|---|
| Amazon | ~$200B | CEO Andy Jassy's own figure, CNBC |
| Microsoft | ~$190B | CFO Amy Hood attributes ~$25B of the raise to rising memory/component costs, CNBC |
| Alphabet (Google) — AEA holding | $180–190B | Raised from $175–185B in Q1, Alphabet's own Q1 2026 earnings release |
| Meta — AEA holding | $125–145B | Raised twice this year, most recently by $10B, Meta's Q1 2026 8-K, Exhibit 99.1 |
2. The cash flow squeeze, in two real companies
Guidance is a forward-looking promise. What's already happened to cash flow is not. Both AEA holdings in this basket reported Q1 2026 results this year, and both show the same underlying pattern at different intensities: capex is growing meaningfully faster than the cash being generated to fund it.
Alphabet
Q1 2026 capex more than doubled to $35.7B, versus $17.2B in Q1 2025 — roughly 60% into servers, 40% into data centers and networking. Trailing-12-month capex is up 90.4% year over year to $109.9B, while trailing-12-month free cash flow fell 14.0%, from $74.9B to $64.4B. Q1 2026 quarterly free cash flow compressed to $10.1B — $0.28 generated for every $1 of that quarter's capex.
Meta
Q1 2026 capex was $19.8B against $12.4B of free cash flow the same quarter — $0.65 generated per $1 of capex, a materially better coverage ratio than Alphabet's, though still capex-negative on a pure cash basis. Meta has no public cloud business to monetize excess capacity into, which makes its capex bet a more direct, less-hedged wager on its own AI roadmap than Alphabet's cloud-diversified spend.
| Company | Q1 2026 capex | Q1 2026 free cash flow | FCF generated per $1 of capex |
|---|---|---|---|
| Alphabet | $35.7B | $10.1B | $0.28 |
| Meta | $19.8B | $12.4B | $0.65 |
Neither company is self-funding its current capex pace out of free cash flow alone this quarter. Both have large cash balances, investment-grade access to debt markets, and (in Alphabet's case) a growing cloud business to eventually monetize the spend — a capex-negative quarter is not automatically a warning sign. It is, however, the honest starting point for asking how long the gap can widen before it matters.
3. The bull case and the bear case, stated honestly
A promotional version of this study picks one side. Both of the following are defensible readings of the same numbers above.
This is growth capex, not maintenance capex
Cloud and AI-services revenue backlogs at both companies are growing faster than the capex funding them, according to management commentary on both earnings calls. If that backlog converts to revenue on schedule, today's cash-flow compression is a temporary, self-correcting phase of an investment cycle, not a structural problem.
Capex is a hard commitment; revenue conversion isn't guaranteed
Data centers and chips are ordered years before they're needed and depreciate on a fixed schedule regardless of utilization. If AI-services revenue growth disappoints even modestly, the capex doesn't unwind — it becomes a depreciation drag on margins for years, exactly the "physical limits of compute" risk this site has argued elsewhere.
The two readings aren't actually in conflict
Both can be true at once: the spend can be rational at the company level given a real backlog, and the aggregate $725B bet across four companies can still be a source of real, correlated drawdown risk for a portfolio holding several of them if the market's patience for the gap runs out before the revenue does.
4. What would actually change my mind
Rather than assert a conclusion, here's what I'm actually watching, stated as falsifiable markers rather than a vibe. If Alphabet's or Meta's free-cash-flow-per-dollar-of-capex ratio keeps compressing for two more consecutive quarters without a matching acceleration in cloud or AI-services revenue growth, that's the point where "growth capex" starts to look more like the bear case. If it stabilizes or improves alongside continued revenue growth, that's the point where the bull case is doing exactly what it's supposed to do. I don't have a strong prior on which happens — I have a number I'm tracking, which is a different thing.
Related research
This is the demand-side half of AEA's AI-infrastructure thesis. The supply side — the physical corridor, the power constraint, and the companies building it — is covered in Data Center Alley. AEA's own concentration in this exact theme (60.93% of the book) is quantified on the Thematic Exposure page and examined structurally in The Concentration Premium.
Methodology & sources
2026 capex guidance figures. Amazon: CEO Andy Jassy's own figure, via CNBC. Microsoft: CFO Amy Hood's guidance, via CNBC. Alphabet: Alphabet's own Q1 2026 earnings release (also filed as an SEC 8-K exhibit). Meta: Meta's Q1 2026 8-K, Exhibit 99.1. The $725B combined total and the 77%/$410B year-over-year comparison are Tom's Hardware's aggregation of these four companies' guidance, cross-checked against ValueAdd VC's AI Spending Tracker. All of these are guidance ranges, not audited actuals, and several companies have raised guidance more than once this year, so figures here may already be stale by the time you're reading this — check the linked primary sources and trackers for the current number.
Alphabet Q1 2026 figures. Q1 2026 capex ($35.7B) and free cash flow ($10.1B and $64.4B TTM) are stated directly in Alphabet's own Q1 2026 earnings release. The trailing-12-month capex figure ($109.9B, +90.4% year over year) is not stated directly by Alphabet in a single release — it's a rolling four-quarter sum calculated by Asymco's analysis of Alphabet's free cash flow trajectory, cited here as a derived figure rather than a company-reported one.
Meta Q1 2026 figures. Capex ($19.8B) and free cash flow ($12.4B) are stated directly in Meta's Q1 2026 8-K, Exhibit 99.1.
FCF-per-dollar-of-capex ratio. This study's own calculation: quarterly free cash flow ÷ quarterly capex, for each company's most recently reported quarter. A simple coverage metric, not a formal corporate-finance ratio with a standard name.
Limitations. This is a two-company, one-quarter-plus-trailing-twelve-month snapshot, not a full four-company, multi-year capex/cash-flow model. Microsoft and Amazon are discussed only at the guidance level here, not the same cash-flow depth as the two AEA holdings. Free cash flow is a noisy, quarter-to-quarter figure sensitive to working-capital timing; a single weak or strong quarter is not, on its own, a trend.
Not investment advice. Nothing here is a recommendation to buy, sell, or avoid any security.