Token Economy Bubble Update
Hanchen Li’s analysis of whether the token economy is in a bubble, distinguishing between “validated demand” (measurable ROI) and “exploration demand” (experimental spending without validated business value).
Key Claims
- Revenue numbers are real: Cursor 14B ARR in 14 months, OpenAI $20B in 2025
- But revenue growth ≠ healthy economy — much is exploration demand, not validated demand
- 42% of companies scrapped most AI initiatives in 2025 (up from 17%)
- Only 5% of AI pilot programs achieve rapid revenue acceleration (MIT, Aug 2025)
- Average sunk cost per abandoned AI project: $4.2M
- Anecdote: engineer burning >$1M/year in tokens with zero validated business value
Dark Fiber Analogy
- 1990s: 80M+ miles of fiber laid, 85-95% remained dark years later
- Technology was right, timeline was wrong, investors went bankrupt
- Tokens are this generation’s broadband — needed eventually, but current demand inflated by exploration budgets
- The “danger zone”: gap between exploration demand peak and productivity demand catching up
Jevons Paradox Critique
- Per-token cost dropped 280x in two years (0.07 per million tokens for GPT-3.5-equiv)
- Enterprise AI spending surged 320% in 2025 despite cost drops
- But Jevons only works when underlying utility is real and elastic
- AI usage ceiling limited before people become AI-native
Takeaways
- Updates/extends Token Economy Bubble with 2026 data
- The vibe coding problem: “speed without direction is just expensive wandering”
- “Ordering from the entire menu because you just got a raise” — not sustainable
- We’re somewhere between “Euphoria” and “Profit Taking” on the cycle
Connections
- Token Economy Bubble — original analysis
- AI $600B Question — related investment analysis
- Stop Pretending AI Economy — complementary critique
- LLM Inference Economics — cost structure analysis