Chamath Palihapitiya's firm observed that gains from frontier models in production systems for regulated enterprises are shrinking while costs are doubling, particularly for complex work beyond basic tasks. The "last 5%" of difficult technical challenges in getting complex systems into production is becoming extremely expensive, making the fully loaded cost of the model plus engineer uneconomical.
Enterprise AI projects are hitting a cost wall for advanced use cases, shifting focus from model capabilities to token cost efficiency and engineer productivity.