CITIC Securities: North American Big Four Cloud Providers Raise 2026 Capex Guidance to $720 Billion
Claire Weston
Amazon, Microsoft, Google and Meta spent a combined $171.2 billion on capex in Q2, lifting full-year guidance to a midpoint of $732.5 billion — up from $710 billion last quarter. This means the AI infrastructure arms race is still accelerating, with no peak in sight.
Where is the $720 billion going?
Full-year guidance for the four totals roughly $720 billion to $745 billion, midpoint $732.5 billion — $22.5 billion above the Q1 midpoint of $710 billion.
Amazon leads at $220 billion, raised from ~$200 billion; Google's midpoint sits at $200 billion; Microsoft at $175 billion; Meta at $137.5 billion.
This means → none of the four is easing off. The money flows into AI servers, GPU clusters — large arrays of GPUs networked for parallel computing — data centers, and power infrastructure.
Are cloud and AI revenues keeping pace?
Q2 revenues from AWS, Microsoft Intelligent Cloud, and Google Cloud totaled roughly $106.3 billion, up ~43% year-on-year and ~15% quarter-on-quarter.
AWS growth hit an 18-quarter high; its AI business now runs at over $25 billion annualized with triple-digit growth. Azure revenue grew 43%.
In plain terms = the hyperscalers are spending aggressively, but the revenue side is running in step. For now, the "invest → earn back" loop looks healthy.
Is Meta spending its money differently?
Meta posted Q2 total revenue of $60.8 billion, up 28% year-on-year; ad revenue hit $59.4 billion, up 27%.
CITIC notes that Meta's AI capex is not aimed at selling cloud services directly. Instead, it monetizes indirectly — through better recommendation algorithms and more efficient ad targeting.
This means → even though Meta is building AI infrastructure at a similar scale, it follows a "sell smarter ads" playbook, not the "sell compute directly" model of AWS and Azure. Two very different monetization paths.
Why is the market selling off anyway?
CITIC judges that AI-compute sector momentum remains intact, but the market has pulled back sharply in recent weeks, with signs of overselling.
The report flags four watch items: ① whether large-model annualized recurring revenue (ARR — subscription revenue annualized) hits a near-term ceiling, especially in the coding segment where recent price cuts may signal pressure; ② progress in non-coding use cases; ③ pricing trends along the compute-inflation chain — cost pass-through from chips to power across the stack; ④ fundraising conditions and risk appetite across AI sub-sectors.
In plain terms = the spending numbers look impressive, but what the market really worries about is whether all that capex converts into sustainable revenue growth — and that remains the core validation checkpoint for this AI investment cycle.
What should investors watch now?
CITIC observes that Q2 positioning in AI names was heavy, and the market is showing signs of a high-to-low rotation — money moving from high-flying momentum stocks into lower-valued, quieter names.
The report recommends focusing on low-valuation, high-dividend names as a defensive play during the correction.
This reflects a deeper signal: when the market splits on whether AI capex will convert to revenue, capital retreats first to assets with more certain payouts.
Content is for reference only, not financial advice.