Nadella warned enterprises using proprietary models pay in fees and again in surrendered knowledge.
The argument lands as open-weight models from Murati’s lab and Chinese labs pull the same direction.
His company has invested billions in both OpenAI and Anthropic, which makes the warning strange.
When the CEO whose company bankrolled the closed-model era warns you about closed models, that is worth five minutes.
The two-payment problem
In a blog post published Sunday, Microsoft CEO Satya Nadella warned that enterprises using proprietary AI models effectively pay twice: once in subscription costs and again by handing over business knowledge embedded in their prompts and corrections, which can be absorbed into future model versions.
Break that down, because it is not abstract. Every time your team writes a prompt that captures how your business actually works, every time someone corrects a model’s output with real domain expertise, that knowledge leaves your building.
You paid to generate it. You paid again in licence fees to use it. And it can end up baked into a product your competitor rents next quarter.
The unusual part is who said it. Microsoft has invested billions in both OpenAI and Anthropic. Hugging Face CEO Clem Delangue made a similar prediction to TechCrunch last week, and Palantir CEO Alex Karp went viral on CNBC saying frontier AI tools from closed model providers are too expensive and do not offer enough clarity on IP protections.
Why the timing is not a coincidence
This landed days before Thinking Machines Lab released Inkling, its open-weight 975 billion parameter model, under Apache 2.0.
The lab’s own argument, published a week earlier, was that AI trained centrally by one company and then set in stone underperforms AI that organisations shape themselves, because so much expertise is specific to the people who hold it.
Cheap open-weight models, many of them from China, pull in the same direction. Organisations can fine-tune open-weight models on their own proprietary data and deploy them on infrastructure they control, giving more flexibility over hosting and costs than most closed models.
Do not read this as “closed models are finished.” They are not. The frontier is still closed, and for most teams a closed API is faster, safer, and cheaper than running a 975B model yourself.
Read it as a question to put on your next renewal. What exactly does our vendor keep from what we send them? Is it written into the contract, or into a policy page they can edit? If the answer is fuzzy, Nadella just told you what you are actually paying.
Quick Links:
Comments
Be the first to leave a comment.