Big Tech's AI Capex: $635B-$665B in 2026
The four hyperscalers - Microsoft, Alphabet, Amazon, and Meta - plan to spend between $635 billion and $665 billion on AI in 2026. That's a 67% to 74% jump from an estimated $381 billion in 2025.
Most of the money is earmarked for AI chips, servers, and data center buildouts. This is the most aggressive investment cycle the sector has seen.
Who's Spending What
- Amazon: ~$200 billion in 2026 capex
- Alphabet: $175-$185 billion
- Microsoft: ~$145 billion run rate for its FY2026
- Meta: $115-$135 billion
Market Reaction
Investors pushed back on the scale and timing. Amazon fell more than 8% after its update. Alphabet slipped 3% post-guidance. Microsoft dropped over 11% after its quarterly report, pressured by slightly slower Azure growth.
Why This Matters for Finance Teams
- Free cash flow: Capex intensity will suppress FCF near term. Watch gross vs. net capex, lease usage, and capitalized vs. expensed R&D.
- Depreciation step-up: Useful lives on data center assets and AI accelerators will drive GAAP EPS optics. Expect heavier D&A in 2H26-2027.
- Capacity risk: Returns hinge on utilization. If AI demand lags buildouts, ROIC compresses. If demand outpaces, expect more spend.
- Supply chain dependence: Chip availability and lead times can swing capex cadence and delivery schedules.
- Power constraints: Energy procurement and grid access are now core to execution; delays here can push revenue out.
Second-Order Effects to Track
- Vendors and contractors: Chipmakers, component suppliers, power equipment, and data center constructors stand to see multi-year order books.
- Energy and cooling: Electricity demand from data centers is rising, putting pressure on regional grids and pricing (IEA overview).
- AI unit economics: Pricing for AI services (training and inference) must improve to match the asset base being built.
Key 2026 Earnings Signals
- Quarterly capex cadence vs. full-year guidance
- Cloud revenue growth relative to incremental capex
- Data center depreciation schedules and useful life assumptions
- Comments on chip lead times, energy availability, and construction timelines
- Shift in spend mix between training and inference infrastructure
Actionable Steps for Portfolios and Planning
- Model scenarios at both $635B and $665B spend; stress-test FCF, net leverage, and ROIC.
- Sensitize utilization, chip pricing, and power costs; small deltas move returns materially.
- Map vendor exposure across your holdings; look for backlog visibility and margin durability.
- Track AI service pricing and attach rates in cloud units; monetization pace is the swing factor.
If you work in finance and need a quick scan of practical AI tools that actually save time, this curated list is a useful starting point: AI tools for finance.
Your membership also unlocks:
