Recent experiments in fine-tuning cheaper AI models for professional domains suggest specialised models could challenge Big Tech's dominance. Legal AI firm Harvey AI partnered with Fireworks AI to fine-tune Chinese model Kimi 2.6, achieving a 40 per cent performance improvement at an eleventh of the cost, briefly matching frontier models before Anthropic's new Opus surpassed it.

A more durable advance came from Bridgewater Associates, which worked with Mira Murati's Thinking Machines Lab to fine-tune Qwen3-235B using proprietary investment workflows. The bespoke model reached 85 per cent accuracy — a 30 per cent error reduction over frontier models — at a fourteenth of the cost. Gains stem from tacit expertise that general models cannot access.

Satya Nadella warned against a world where "every company... is ceding value to a few models." While firm-specific fine-tuning reduces dependency, it raises geopolitical risks from Chinese base models and employee concerns about knowledge extraction. The balance of trust and power in each workplace will determine adoption.