One concern is increasingly shaping the debate around artificial intelligence: proprietary AI models may be functioning less like neutral tools and more like strategic Trojan horses.

As startups and large enterprises rely on models from companies such as OpenAI and Anthropic, critics argue that model providers gain access to valuable institutional knowledge that could eventually become a competitive advantage against the very companies using their systems.

Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram

The Data Paradox At The Heart Of Enterprise AI

Warnings about this dynamic have come from investors and executives, including Jason Calacanis and Palantir CEO Alex Karp. Now Microsoft CEO Satya Nadella has entered the debate with a blog post published on Sunday, arguing that enterprise customers are effectively paying twice for AI.

First, they pay for token usage. Then, more quietly, they pay with the proprietary knowledge required to make the model genuinely useful.

“You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful. The better you want the model to perform, the more of that knowledge you have to feed it!”

Nadella argues that enterprises are teaching AI models how their businesses operate through prompts, workflows and corrections.

“Models learn from ‘exhaust,’ the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know-how.”

Fair Use, Distillation, And The Battle Over Model Access

Nadella also challenges the industry’s own logic. If AI companies are allowed to train their models on publicly available content, he argues, enterprises should also be free to learn from those models.

Distillation, the practice of using one model’s outputs to train another, has become one of AI’s most contentious issues. Earlier this year, Anthropic accused Chinese developers of sending millions of prompts to Claude to improve competing models and called for tighter U.S. export controls.

Nadella argues that the industry cannot champion openness when it benefits model developers while restricting imitation when it benefits customers.

“While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation.”

Ownership, Control, And The Push Toward Open Systems

Another of Nadella’s concerns is that some AI providers reserve the right to learn from customer prompts and interaction data, creating what he sees as a structural conflict between vendors and enterprise customers.

His proposed solution is for organisations to retain ownership of their data, including prompts and feedback, while building proprietary learning environments in the cloud. He also encourages companies to adopt orchestration layers that make it easier to switch between AI models instead of becoming dependent on a single provider.

That approach is already gaining traction. AI gateways that route requests across multiple models are becoming increasingly popular as businesses seek greater flexibility, stronger governance and tighter cost control.

Although Nadella does not explicitly frame his argument as a case for open source, it aligns closely with a broader enterprise shift toward models that organisations can run and manage themselves.

Why Open Source Is Winning Share In The Enterprise

Large organisations with their own data centres are increasingly deploying open-source models on premises, allowing them to keep sensitive data within their own infrastructure while reducing costs.

Idit Levine, founder and CEO of Solo.io, says many customers are moving in that direction after experimenting with proprietary vendors.

“Can I take an open source model and run it on-prem? It will do almost 90% of what the big one’s doing. It will cost way less. They understand that, and they can control it.”

The trend extends beyond infrastructure providers. Companies including Vercel and OpenRouter have reported growing adoption of open-source models. According to Vercel, open models accounted for 29% of traffic routed through its AI gateway last month.

The Strategic Signal For Enterprise Leaders

Microsoft’s position reflects a broader shift in enterprise AI, where ownership, portability and control are becoming almost as important as model performance.

As Nadella concluded:

“In consuming intelligence, you are creating intelligence. And what you create should belong to you.”

For enterprise leaders, that is increasingly becoming not just a philosophical principle, but a procurement strategy.