Satya Nadella warns companies are paying twice for AI

The concern centers on the 'exhaust' generated by AI interaction: prompts, agent tools, and human corrections that essentially teach labs the nuances of a client's business. Nadella describes this as a double payment, where enterprises trade proprietary data for intelligence, effectively building a competitive advantage for the very companies they hire. He views the current industry trend of restricting 'distillation'—the process of learning from model outputs—as hypocritical, given that labs rely on scraping the public internet to train their own systems.

To reclaim ownership, Nadella advocates for proprietary learning environments and orchestration layers that allow companies to switch between providers, preventing vendor lock-in. This shift toward autonomy is gaining momentum as enterprises increasingly turn to open-source models hosted on-premise. Idit Levine, CEO of Solo.io, notes that companies are discovering that local, open-source models can achieve 90% of the performance of proprietary alternatives at a fraction of the cost and risk. As traffic to open-source gateways surges, the industry appears to be moving away from total dependence on centralized labs, prioritizing data sovereignty over the convenience of plug-and-play AI.