Adaptation Curve

Lens · L1 · Demand & Capex

Is the spend being earned back? That's the whole ballgame.

Hyperscaler capex is a ~$340B/yr wager that AI demand is real. The bubble tell was never the spending — it's whether app-layer revenue is closing the gap. Here's that gap, over time.

Why this matters to you

Demand durability decides whether your compute budget keeps growing. The earn-back ratio is the leading indicator for next year's allocation.

Hyperscaler capex guidance

MSFT · GOOGL · AMZN · META · ORCL
2026-03-13600source →

Big-four 2026 capex ~$600B; full supply chain ~$1T. Amazon ~$200B, Google ~$180B (weighing ~$300B next yr, FCF->0)

Each reading is a dated, LLM-extracted take on hyperscaler capex guidance from EDGAR filings. The numeric_value is the headline aggregate across filers as the extractor finds it.

Software-layer monetization vs. capex

manual Tier C
2026-06-14source →

MANUAL ENTRY NEEDED — fill during quarterly review (Tier C signal).

2026-05-2011source →

Anthropic added ~$11B ARR in one month (March) — ~ the combined revenue Palantir+Snowflake+Databricks built over a decade

2026-05-20200source →

OpenAI and Anthropic each expected to exceed ~$200B ARR this year, aided by usage-based pricing

2026-05-20source →

Profits accruing to energy, data centers, chips, and models — not to applications; app layer net-destroyed value

2026-03-1320source →

Anthropic ARR ~$20B, gross margin sub-50%; adds ~$4B (Jan), ~$6B (Feb); reliability low due to compute constraint

2026-03-135source →

Google Gemini ~$5B Q4 ARR after Nano Banana / Gemini 3 inflected usage (from ~nothing Q1-Q3)

Reported capex — per quarter (XBRL)

MSFT · GOOGL · AMZN · META · ORCL
CompanyQuarter endCapex (B USD)Source
meta2026-03-3119.00filing →
microsoft2026-03-3130.88filing →
google2026-03-3135.67filing →
amazon2026-03-3144.20filing →
microsoft2025-12-3129.88filing →
microsoft2025-09-3019.39filing →
amazon2025-09-3035.09filing →
oracle2025-08-318.50filing →
amazon2025-06-3032.18filing →
meta2025-03-3112.94filing →
microsoft2025-03-3116.75filing →
google2025-03-3117.20filing →
amazon2025-03-3125.02filing →
microsoft2024-12-3115.80filing →
microsoft2024-09-3014.92filing →
amazon2024-09-3022.62filing →
oracle2024-08-312.30filing →
amazon2024-06-3017.62filing →
meta2024-03-316.40filing →
microsoft2024-03-3110.95filing →
google2024-03-3112.01filing →
amazon2024-03-3114.93filing →
microsoft2023-12-319.73filing →
microsoft2023-09-309.92filing →
amazon2023-09-3012.48filing →
oracle2023-08-311.31filing →
amazon2023-06-3011.46filing →
meta2023-03-316.82filing →
microsoft2023-03-316.61filing →
google2023-03-316.29filing →
amazon2023-03-3114.21filing →
microsoft2022-12-316.27filing →
microsoft2022-09-306.28filing →
amazon2022-09-3016.38filing →
oracle2022-08-311.72filing →
amazon2022-06-3015.72filing →
microsoft2022-03-315.34filing →
google2022-03-319.79filing →
amazon2022-03-3114.95filing →
meta2022-03-315.44filing →

Sourced directly from SEC XBRL company-facts (PaymentsToAcquirePropertyPlantAndEquipment and equivalents). Discrete quarterly frames only — no YTD estimates.

The earn-back chart — capex series loaded, revenue next

Per-quarter capex is now loaded from XBRL. The two-line capex vs. app-layer revenue chart needs the revenue side — a second ETL pass adding cloud segment revenue (metric_key='cloud_revenue_usd') to the same spine.