Token Economics and Pricing
Definition
The emerging economic framework in which AI value is measured, priced, and captured via tokens rather than compute-hours or software seats — and the implications for business models, GPU monetization, and the broader AI value chain.
Key Points
- Shift from GPU-hours to token-dollars: same GPU yields 2-4x more revenue when monetized by token value vs hourly rental (clouded judgement per token pricing)
- Jensen's Pareto frontier: throughput vs latency tradeoff; pricing power lives at the premium end (low-latency agent tokens) (clouded judgement per token pricing)
- Credit-based pricing will win as abstraction layer over complex token economics; customer buys credits, operator manages model mix (clouded judgement per token pricing)
- In AI, pricing model = business model; unlike 80%+ margin SaaS, inference costs make pricing mistakes existential (clouded judgement per token pricing)
- Token cost deflation: Vera Rubin delivers 5x throughput of Blackwell, 10x cost reduction (clouded judgement per token pricing)
- Alchian-Allen effect: rising fixed GPU costs narrow the price ratio between premium and commodity tokens, pushing users toward the best models (dwarkesh dylan patel interview)
- Anthropic's revenue growth ($4B added in Jan, $6B in Feb) demonstrates token demand elasticity (dwarkesh dylan patel interview)
- 5-10x ROI on AI tools = massive pricing headroom before demand destruction (great gpu shortage rental capacity)
- "API tax starting to look like the cloud markup of 10 years ago" — observation that paying per-token to frontier labs may become uneconomical relative to training vertical models in-house on open-source bases (ai daily brief anthropic mythos vertical models)
- Multiple product companies (Cursor, Intercom, Decagon, Pinterest, Airbnb, Notion) finding it better/cheaper/faster to post-train open models in-house rather than rely on API calls to frontier models (ai daily brief anthropic mythos vertical models)
Open Questions
- Will per-token pricing cannibalize per-seat SaaS models, or will they coexist?
- How will tiered token pricing (premium vs commodity) actually be implemented in practice?
- At what token cost does AI become a true utility (like electricity)?
- How do model providers balance quality improvements (which increase value) with cost deflation (which decreases price)?