source: techcrunch ai: satya nadella has issued a shocking warning to companies using ai
level: business
microsoft ceo satya nadella published a blog post warning companies about hidden costs of using proprietary ai models. he said buyers pay twice: once with money for token usage and again by revealing proprietary knowledge. the better a company wants a model to perform, the more sensitive data it must feed in. nadella argued that models learn from prompts, tool use, and corrections, distilling institutional know-how that competitors could never buy.
nadella criticized model makers for scraping public data to train models while restricting others from distilling their models. he called it hypocritical to impose restrictive terms on distillation after benefiting from fair use of public data. his solution involves companies retaining ownership of their data, including prompts and feedback. he urged building proprietary learning environments on the cloud and using orchestration layers to switch between models without lock-in.
the post aligns with a growing shift toward open source models run on-premises. companies like solo.io, vercel, and openrouter report increased traffic to open models, which can handle most tasks at lower cost with more control. nadella's warning, coming from a major investor in openai and anthropic, may accelerate this trend. he concluded that what companies create through ai should belong to them.
why it matters: enterprises risk losing competitive advantage by exposing proprietary data to ai model providers, making data ownership and model portability critical for ai strategy.
source: techcrunch ai: satya nadella has issued a shocking warning to companies using ai