Amazon, Google, and Meta combined for over $90 billion in quarterly capex on AI infrastructure. The spending binge confirms a structural shift: growth is now gated by power, chips, and data center construction timelines.
Key facts
- Amazon AWS revenue: $37.6B, up 28% year over year.
- Google Cloud backlog: $460B+.
- Meta quarterly capex: $19.8B.
- Microsoft capex declined sequentially to $31.9B.
- Google Cloud revenue grew 63% year over year.
First-quarter earnings from Amazon, Google, and Meta confirmed what Microsoft signaled last month: AI demand is growing faster than the industry can build capacity. Combined cloud revenue hit $77.4 billion, with Google Cloud growing 63% year over year to approximately $20 billion and AWS posting $37.6 billion—up 28% [According to Data Center Knowledge].
Meta's $19.8 billion quarterly capex, with expectations for further increases, underscores the scale required. Amazon CEO Andy Jassy pointed to sustained demand, while Google's cloud backlog swelled to more than $460 billion—a backlog larger than most companies' annual revenue.
The unique take: hyperscalers are now pre-selling capacity they haven't built yet. Backlog growth across providers reflects demand that has yet to be delivered, effectively turning infrastructure buildout into a forward-selling engine.
Microsoft offered a counterpoint: Azure grew 40%, but capex declined sequentially to $31.9 billion from $37.5 billion. Analysts called it discipline, not retreat. "That looks like discipline, not retreat," said Steven Dickens of HyperFrame Research [According to Data Center Knowledge].
Patrick Moorhead of Moor Insights & Strategy noted in an April 30 LinkedIn post that "the AI trade had three tests," and all three held up—capex, cloud growth, and real demand versus hype.
The Bottleneck Shifts from Silicon to Substations
The earnings cycle reveals a new constraint: power availability. Microsoft's validation of Nvidia's Vera Rubin NVL72 system and Google's $5B+ Texas data center investment for Anthropic signal that chip supply is less the bottleneck than grid interconnection timelines. PJM recently reported 220GW in interconnection requests, with Google-backed AI processes queued [As previously reported].
Meta's 8,000 job cuts to fund $145B in 2026 AI capex further illustrate the trade-off: headcount reductions finance infrastructure scale [Per our prior coverage].
What the Numbers Don't Say
Notably absent from earnings calls: ROI timelines for AI workloads. While cloud revenue grows, the ratio of capex to AI-driven revenue remains opaque. Analysts like Dickens noted the earnings cycle "hit two of three cleanly" on capex, cloud growth, and real demand—but the third metric, profitability, remains unaddressed.
Key Takeaways
- Amazon, Google, Meta spent $90B+ quarterly on AI infra; Google Cloud backlog hit $460B.
- Demand outstrips supply, pre-selling capacity not yet built.
What to watch
Watch for Q3 capex disclosures: if Microsoft's sequential decline continues while Amazon and Google accelerate, it signals a divergence in infrastructure strategy. Also track grid interconnection approval timelines for Google's Texas data center and Meta's new builds.
