# Post-Training Data Becomes the Bottleneck

As gains from raw scale flatten, high-quality post-training and reinforcement-learning data becomes the scarce input, and a specialized vendor market will form to supply it.

- Conviction: 40 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-08-01T00:00:00.000Z
- Last updated: 2026-08-01T05:51:13.867Z
- Canonical: https://polylog.news/ai/trends/post-training-data-becomes-bottleneck
- Publisher: Polylog
- Affected regions: Global

## Recent evidence

- [confirms] A GeneBench-Pro Co-Author Leaves OpenAI to Build a Startup Selling Reinforcement-Learning Data (2026-08-01): A GeneBench-Pro co-author left OpenAI to start a venture selling specialized reinforcement-learning datasets for post-training, arguing data quality now limits model gains more than scale (per the edition report). A named researcher spinning out a dedicated RL-data vendor is direct evidence the specialized supply market the thesis predicts is forming.