Behind the AI spending wave

Microsoft told investors that fiscal-year 2026 capital expenditure was approximately $190 billion and that cash capital expenditure in its fourth quarter would exceed $40 billion. Meta separately reported $31.08 billion of capital expenditure for the second quarter of 2026. These are company disclosures with different definitions and reporting periods; they do not establish a verified $700 billion global AI-infrastructure total or a reliable industry-wide spending-to-revenue ratio.

What the disclosures establish

The disclosures confirm sustained investment in data centres, accelerators, networking, and supporting infrastructure. They also show that capacity planning remains a material constraint for major providers. They do not isolate every euro or dollar spent only on AI, and they do not prove the return generated by each investment.

What enterprise buyers should measure

Buyers should evaluate their own unit economics: cost per successfully completed task, utilisation, latency, data-transfer charges, operational staffing, and switching cost. Published provider expenditure is useful context, not a substitute for an internal business case.

Practical decision rule

Use short, measurable pilots; keep model and infrastructure interfaces portable; negotiate usage visibility; and compare hosted, reserved-capacity, and private-deployment options against the same workload. Treat forecasts about future price declines, capacity abundance, or provider consolidation as scenarios rather than facts.