The AI Rug Pull
5 hours ago
- #Future of AI Pricing
- #Training Data Economics
- #AI Subsidy Model
- Frontier AI assistants are currently priced below cost, with users effectively providing training data through interactions, making the user a training set rather than the primary customer.
- The subsidy, estimated at tens of billions annually, is justified by data collection for model improvement via human feedback, but synthetic data, agentic self-play, and scale are reducing the need for human input.
- When the apprenticeship ends, pricing will rise significantly, with the $20/month plan disappearing, capabilities gated behind enterprise contracts, and labs possibly integrating vertically to capture full economic value.
- This shift may feel like a 'rug pull' for users, similar to patterns in streaming, ride-sharing, and cloud services, where subsidized growth precedes price hikes and feature restrictions.
- Enterprise users and those using open-weight models on owned or rented hardware will adapt, while casual consumers and small operators relying on subsidies will be negatively impacted.
- Strategies for the interim include building for portability, focusing on capability over brand, avoiding reliance on free tiers for critical tasks, and self-hosting models to control costs.