Citi: Global AI Capex Rises to $1.2 Trillion, Market Underestimates 2027 Revenue
Miles Bennett
Citi's latest report lifts 2026 combined capex for the Big Four cloud firms to $730 billion, with total global AI infrastructure spending reaching $1.2 trillion. The real mispricing, Citi argues, is not that spending is too high — it is that the market systematically underestimates how much revenue that spending will generate in 2027–2028.
How did cloud revenue perform in Q2 — did it beat expectations?
Combined cloud revenue growth for the Big Four accelerated from 39% year-on-year in Q1 to 48% in Q2, with all three reporting units beating consensus.
Specifically: Google Cloud grew 82%, beating estimates by 17 percentage points; AWS grew 37%, beating by 6 pp; Azure grew 43%, beating by 3 pp.
This means → the acceleration is not one outlier pulling the average — the entire cloud sector is speeding up in lockstep as AI demand converts into actual revenue.
Order backlogs are still rising — what does that signal?
Combined backlog for Google, Microsoft, and Amazon rose from $1.45 trillion in Q1 to $1.69 trillion in Q2; year-on-year growth climbed from 143% to 151%.
In plain terms = customers are placing orders faster than cloud providers can deliver — demand is still running ahead of supply.
This reflects a bottleneck that is not about weak demand but about constrained capacity.
With spending this high, are returns holding up?
The Big Four's trailing-twelve-month CROCI — cash return on capital invested, adjusted for debt and capitalised R&D — held at 28%, just 1 pp below last quarter's 29%. EBITDA margins stayed at 43% for a second straight quarter.
Free cash flow is under pressure, but Citi attributes this to capex timing running ahead of revenue timing, not to deteriorating project-level returns.
This means → the money is out the door but hasn't come back yet — the P&L looks strained for now, but the underlying ability to generate returns has not worsened.
Why does Citi say the market underestimates 2027 revenue?
Citi time-shifted capex growth against revenue growth and found that capex inflection points typically lead revenue inflection points by about three years.
That implies the accelerated AI infrastructure spending since 2024 will deliver its main revenue contribution from 2027 onward.
Citi estimates the market may be underpricing 2027 hyperscaler revenue growth by roughly 500 basis points, and 2028 by roughly 900 basis points. After Q2 results, the Street raised its 2027 revenue forecast by 2.3% — Citi sees this as the first sign that "revenue expectations are starting to catch up with capex," but says there is still a gap to close.
Why is each company still raising spending — and is the rationale the same?
Meta raised its 2026 capex guidance to $130–145 billion; Google raised to $190–205 billion, up ~116% year-on-year; Microsoft guided ~$175 billion; Amazon added ~$20 billion, reaching $220 billion, up ~67% year-on-year.
Citi stresses that the key point is not the dollar amounts but that every company cited supply constraints as the reason: Google said it is accelerating capacity delivery; Amazon flagged rising memory costs; Microsoft and Amazon both said demand exceeds supply.
In plain terms = this is not "spending more because we feel like it" — it is "chips, memory, power, and data centres are all bottlenecks, so we have no choice." As long as bottlenecks persist, higher spending does not equal oversupply.
What does Citi see as the next phase of the trade?
Citi sums up the cycle's logic as "CapEx + Time = Profits" — capital spending plus time equals earnings.
The next phase of the AI trade, Citi argues, may shift from "raising capex forecasts" to "raising revenue, profit, and long-term return forecasts."
This means → whether the market can complete that expectation shift before 2027 will determine if cloud valuations can re-rate further — and that, in Citi's view, is the single most important verification point ahead.
Content is for reference only, not financial advice.