Five companies are on pace to spend more than $700 billion on AI infrastructure in 2026 — while the clearest revenue number in the group, Microsoft’s $37 billion AI run rate, is smaller than Meta’s capex guidance alone. That gap, not any single earnings beat or miss, is now the central data story in AI markets. This is a living tracker: we’ll update the figures below as hyperscalers report and as litigation and GPU-pricing datasets refresh.
Why it matters
For two years, the market gave Big Tech a pass on AI spending because the revenue story was young. That grace period is ending. Microsoft, Alphabet, Meta, and Amazon have all disclosed or reinforced full-year capex plans in a short window, and three of the four raised their numbers rather than holding steady. Combined 2026 capex across the five biggest spenders — including Apple — is now estimated above $700 billion. Investors are no longer asking whether AI generates revenue; they’re asking whether it generates enough revenue, fast enough, to justify infrastructure bills of this size. That’s the crux of the AI capex vs revenue debate that will define 2026 earnings season.
The numbers, side by side
| Company | Metric | Figure |
|---|---|---|
| Microsoft | AI business annual revenue run rate | $37B, up 123% YoY |
| Microsoft | Azure growth | ~40% YoY |
| Microsoft | Intelligent Cloud revenue | $34.7B |
| Microsoft | CY2026 capex guidance | ~$190B (incl. ~$25B from higher component costs) |
| Meta | FY2026 capex guidance | $125B–$145B (up from $115B–$135B) |
| Big 5 combined | Estimated 2026 capex | >$700B |
The most striking single line: Microsoft’s infra spend guidance of roughly $190 billion for calendar 2026 is more than five times its disclosed $37 billion AI revenue run rate. Azure growth at ~40% is genuinely strong, and the 123% year-over-year jump in AI revenue is not a rounding error — but the ratio of spend to disclosed AI revenue is the number bulls and bears are both citing, just to opposite conclusions.
Key developments to track
- Capex keeps climbing, not plateauing. Meta lifted its full-year guidance from $115B–$135B to $125B–$145B — a $10 billion bump at the midpoint in a single revision cycle.
- Component costs are now a named line item. Microsoft is flagging roughly $25 billion of its 2026 capex increase specifically to higher hardware component prices, not just added capacity — a signal that supply-chain inflation is becoming its own budget category.
- The market’s patience is thinning. Reaction to earnings is increasingly sensitive to the spend-versus-monetization gap rather than to headline revenue growth alone.
- Litigation is a parallel, fast-moving dataset. One live tracker counts 108 major AI lawsuits; another logs 219 cases, 448 claims, and 263 separate enforcement actions. These aren’t static roundups — they’re dashboards updated as filings land.
Builder note: cloud GPU economics matter more than the megacaps’ balance sheets
For teams actually shipping AI products, the hyperscaler capex war is background noise; cloud GPU economics is the number that hits your invoice. As of August 2026, on-demand H100 pricing carries a median around $2.49 per GPU-hour — but the spread is wide. Specialist clouds cluster between $2.15 and $2.85 per hour, while hyperscaler on-demand pricing runs $3.70 to $3.98 per hour, roughly 50–80% higher for the same chip. If your team is optimizing training or inference cost, that spread is often bigger than any efficiency gain from a new model release. The practical takeaway: benchmark specialist providers before defaulting to your existing cloud vendor, and treat GPU pricing as a leaderboard worth checking quarterly, not once.
The honest counter-argument
The bear case — spend is outrunning revenue, full stop — has an obvious rebuttal: cloud infrastructure has always been a multi-year bet, and Azure’s 40% growth plus a 123%-YoY AI revenue line are not weak signals in isolation. Depreciation schedules and long-lived data centers mean this year’s capex isn’t meant to be justified by this year’s revenue. The counter to the counter: at some point the ratio matters, and a $190 billion outlay against $37 billion in disclosed AI revenue is a ratio investors will keep scrutinizing every quarter until it visibly narrows.
What we’re watching next
- Whether Alphabet and Amazon follow Meta and Microsoft with further capex raises this cycle
- Whether the 263 tracked enforcement actions produce any settlements that reset industry norms
- Whether the hyperscaler-vs-specialist GPU pricing gap narrows or widens as new chip generations ship
This tracker will be updated as new capex disclosures, GPU pricing data, and litigation counts come in.