Source video: *OpenAI's Finances Just Leaked. We're Cooked* by MonkeyExplains, published 2026-07-06: https://www.youtube.com/watch?v=eXbZrx5XW_k
This post is a structured English summary of a long-form Chinese analysis of OpenAI's leaked financials and the broader enterprise AI ROI problem.
Key points
- The coconut stand parable: The author compares OpenAI to a stall that loses $3.40 for every $1 of sales in year one, $2.60 in year two, and runs a -122% operating margin in Q1 2026 — it isn't "not yet profitable," it's losing *more* money as it grows.
- Leaked financials (June 2026, confirmed by The Motley Fool and others):
- 2024 net loss: $5.09B; cash burn ~$22B
- Annualized 2026 loss projected at ~$28B (~$150M/day)
- $600B in data center commitments signed through 2030
- Break-even forecast: no earlier than 2029; cumulative 2023–2028 losses projected at $44B
- "Narrative engineering": Altman emphasizes 253% revenue growth while costs grew 272%. The loss-to-revenue ratio improved (2.37 → 1.60) but absolute losses grew by $12.1B. The post contrasts this with Amazon's deliberate market-share investment: OpenAI's loss is structural subsidy, not strategy.
- Fundraising magic: Despite the $20.9B loss, OpenAI raised mega-rounds in 2025 (SoftBank, Microsoft) and targets a $852B–$1T IPO underwritten by Goldman Sachs, Morgan Stanley, and JPMorgan — using the growth story itself to justify losses.
- The 5% curse (MIT GenAI Divide): MIT Media Lab's Project NANDA report (*The GenAI Divide: State of AI in Business 2025*, July 2025; 300 deployments, 52 executive interviews, 153 leader surveys) found 95% of enterprise GenAI pilots produce no measurable financial return. Causes: learning gaps (AI can't adapt to workflows), tools disconnected from processes, budget misallocation (50% spent on sales/marketing instead of back-office automation), and in-house builds failing at twice the rate of partnerships.
- Corroborating data:
- The four-step doom loop: (1) Layoffs replaced by AI → public failures (Gannett's AI sports reporting, Pizza Hut's chatbot, VW's inspection system); (2) management panics and mandates AI usage KPIs → metric-gaming, not productivity; (3) budgets exhaust, CFOs find negative ROI; (4) enterprises demand price cuts, but OpenAI loses $1.22 per $1 of service — a race to the bottom where survival depends on outlasting competitors in burn rate.
- Cost passed to the public: Data center electricity, water, tax incentives, and property devaluation are shifting AI's costs onto residents, fueling recalls of supportive officials, state-level moratoriums, and community protests.
- The closing question: "If AI were truly efficient, why would you need to force employees to use it?" Excel, email, and search engines needed no mandates. The post concludes the current AI boom resembles a justification operation for massive ineffective investment — either the greatest example of repackaging losses as growth (via the trillion-dollar IPO) or the most expensive tech bubble in history.
- MIT Media Lab Project NANDA, *The GenAI Divide: State of AI in Business 2025* (July 2025)
- RAND Corporation, *Meta-analysis of Enterprise AI Initiatives* (2026)
- Forrester, *AI ROI Measurement Survey* (2026)
- IBM, *CEO Study: AI Adoption and ROI* (2026)
- Gartner, *Predicts 2026: Agentic AI* (2025)
- McKinsey, *State of AI 2025* (2025)
- The Motley Fool, *OpenAI's Financials Were Just Leaked* (2026-06-17)
- The Information, *OpenAI 2026 Loss Forecast* (October 2025)
| Metric | 2024 | 2025 | 2026 Q1 | |---|---|---|---| | Revenue | $3.7B | $13.07B | ~$5.7B | | Total costs | $12.48B | $34B | — | | Operating loss | $8.78B | $20.92B | — |
| Source | Finding | |---|---| | RAND | 80.3% of enterprise AI projects fail | | Forrester 2026 | 79% perceive productivity gains; only 29% measure ROI | | IBM 2026 | Only 25% hit expected ROI; 16% scale enterprise-wide | | Gartner | 40%+ of agentic AI projects abandoned by 2027 | | McKinsey 2025 | 73% of pilots never pass proof-of-concept |
Of $684B in 2025 enterprise AI investment, an estimated $547B produced zero measurable results. Meanwhile a "shadow AI economy" thrives: only 40% of firms have official LLM subscriptions, but 90% of employees use personal ChatGPT/Claude accounts for work.