Four hyperscalers will spend north of $700 billion this year on the proposition that we'll benefit from computers sounding more confident. A figure that has grown so large it no longer alarms — just a comforting siren song calling on the stock market to spiral ever upwards. But it's not all one direction. Meta, having built more compute than it can consume, is now quietly shopping the surplus back to the industry: a merchant of what was, until recently, marketed as a scarce and transformative resource.
Anthropic and Blackstone reckon the next trillion-dollar AI business is explaining AI to the people who bought it ...
More RotD …As for payoff? Revenue does exist. OpenAI’s $25B ARR is real. Anthropic’s climb to $47B in three years is more so. But each new customer now arrives with a token meter that Tesla, Uber, and Walmart have begun to cap. Engineers buy by the sip while suppliers pour by the bottle. So profit remains elusive for the pure AI players: a claim on margins that must arrive before depreciation schedules and interest bills.
| Company | Revenue | Spend | Profit / (Loss) | 1 Year Ago |
|---|---|---|---|---|
| OpenAI | ~$25B ARR | ~$27B burn FY2026 | −$14B projected | ~$7B ARR; ~$5B loss |
| Anthropic | ~$47B ARR | Burn ~33% of revenue | Losses narrowing | ~$4B ARR |
| Cloud $20B Q1 (+63%) | $180–190B capex FY2026 | Profitable; EPS $5.11 | Cloud ~$50B ARR | |
| Microsoft | AI $37B run rate | $190B capex FY2026 | Profitable; 46% margin | AI ~$13B run rate |
| Meta | $56B Q1 (+33%) | $125–145B capex FY2026 | Profitable; 41% margin | ~$165B/yr |
| AWS | $32B Q1 (+40%) | ~$180B capex FY2026 | Profitable; 35% margin | ~$100B/yr |
| xAI | ~$1B ARR | ~$4B burn FY2026 | −$3B projected | Pre-revenue |
1 Spent — Cumulative capital expenditure on AI infrastructure by the major hyperscalers (Google, Microsoft, Meta, Amazon), plus operating losses at AI-native labs (OpenAI, Anthropic, xAI, and others with disclosed figures). Where AI-specific capex is not separately reported, an industry-consensus fraction of total capex is used. Figures run from ChatGPT’s public launch (November 2022) to the date shown.
2 Earned — Cumulative AI-attributable revenue across the same companies — AI cloud services, API access, and consumer subscriptions. Not all companies separately report AI revenue; figures are estimated from earnings calls and analyst disclosures, and are intentionally conservative: revenue is counted only where the AI contribution can be reasonably isolated.