Summary
Amazon, Alphabet, Meta, and Microsoft all reported Q1 2026 calendar earnings on April 29 and lifted their 2026 capex guides. The combined number for the four pushed past $725B (up 77% YoY from $410B) - about $95B higher than the $630B figure used in Episode 4. Microsoft now plans $190B in CY2026 capex ($25B of which the CFO attributed to rising memory chip and component costs). Amazon led the quarter at $44.2B in Q1 capex alone, with AWS revenue +28%. Alphabet spent $35.67B in Q1 (more than 2x YoY), with Google Cloud orders backlog over $460B and Cloud revenue +63%. Meta raised full-year guidance to $125-$145B, citing higher component and data center costs.
Per-Company Numbers
| Company | 2026 Capex Guide | Q1 Capex | Q1 AI/Cloud Read |
|---|---|---|---|
| Amazon | (Q1 run-rate ~$176B annualized) | $44.2B | AWS +28%; chip business at $20B run rate |
| Alphabet | (Q1 run-rate ~$143B annualized) | $35.67B | Google Cloud +63%; cloud backlog >$460B |
| Microsoft | $190B (CFO; vs $152B consensus) | $30.88B (FQ3, +84% YoY) | AI revenue at $37B annual run rate |
| Meta | $125-145B (raised) | n/a | Cited component + data center cost inflation |
| Combined | ~$725B |
Market Reaction
- Meta: -6% on the print
- Microsoft: -2.5%
- Alphabet: gained on cloud strength
- Market focus shifted from top-line beats to payback timeline and capital intensity
Conclusions
The $725B figure is ~4.7x the entire US electric utility industry’s 2024 investment. In one quarter, the hyperscaler capex envelope expanded by $95B - more than the entire 2026 capex of either Microsoft or Meta as recently as 2024.
Microsoft’s $25B component-cost attribution is the cleanest tell: the $190B figure is not just buying more compute, it is buying compute at higher unit prices because the equipment supply chain (chips, memory, transformers, switchgear, turbines) is supplier-favored. Meta’s “higher component and data center costs” comment says the same thing in fewer words.
Our Thinking
Three things matter most:
-
The capex commitment is now structurally above $700B/year and the binding constraint is no longer financial. The political constraint was resolved by the March 4 White House pledge (covered in Ep4). The financial constraint is now demonstrably absorbed by hyperscaler balance sheets without market pushback at the top-line level. Market concerns are about ROI, not affordability.
-
The capex bump is partly inflation-driven, not pure volume growth. Microsoft’s $25B for memory chips and components is a direct read on the supply-side pricing power problem. This is the same dynamic GE Vernova called out (“pricing rising faster than inflation”) and Siemens called out (“excellent pricing conditions”). The order book is now repricing.
-
The capex-to-physical-delivery gap is now empirical. The hyperscalers are committing dollars at $725B/year against a physical delivery ceiling of 12 GW of new US data center capacity in 2026, of which only ~5 GW is actually under construction (see us-data-center-delays-12gw). The financial book and the build book are diverging. That gap is what the rest of this episode’s research covers.
Watch
- Q2 2026 hyperscaler prints (late July/early August) - does the capex bump hold or rise?
- Disclosed allocations between AI training infrastructure vs cloud expansion
- AWS Bedrock / Azure OpenAI / Google Cloud AI ARR
- Any new direct hyperscaler investment in equipment manufacturers (precedent: Microsoft-Chevron-Engine No. 1)
- Whether memory chip pricing keeps absorbing the marginal capex dollar