Satya Nadella runs Microsoft. And on Monday, he published a blog post that caught a lot of people off guard. 

He warned businesses about the hidden cost of using AI tools. It’s not about money but something a lot more valuable.

Hidden Price Tag

When companies use AI tools like ChatGPT or Claude, they pay for it. But Nadella says there’s a second, invisible cost. 

Every time a business uses an AI model, it feeds that model information. That sums up business secrets and trade knowledge; basically, everything that makes a company special.

And that information doesn’t just disappear after you use it. The AI company can learn from it.

Nadella put it simply. Companies pay for smart AI twice. Once with cash and once with their own knowledge.

He wrote that the better you want the AI to work, the more secrets you have to share with it. That’s a tough trade-off.

People type prompts, give feedback, and correct the mistakes the AI makes. Nadella says all of that gets soaked up by the model. 

Every fix, tweak, and little correction becomes a lesson for the AI. Over time, this adds up to deep knowledge about how a business really works. 

And no competitor could ever buy that kind of insider knowledge, no matter how much money they had.

Yet businesses are handing it over for free, often without even thinking about it.

Fair Play

AI companies built their models by scraping tons of information from the internet. They call this fair use.

So Nadella asks a fair question. If AI labs get to freely learn from everyone else’s data, shouldn’t companies get to learn from the AI models too?

This process is called distillation. It means studying how a model behaves, then using that knowledge to build your own, often cheaper, version.

Right now, many AI companies don’t allow this. They put rules in place blocking customers from doing it. Nadella thinks that’s a bit hypocritical.

He’s especially bothered by AI companies that reserve the right to learn from how customers use their tools, while blocking those same customers from learning back.

Nadella’s Pitch

Nadella wants businesses to own their own data and build what he calls a proprietary learning environment. 

He recommends a private space in the cloud where a company’s AI data stays under its own control.

He also wants businesses to use what he calls orchestration layers. These let a company switch between different AI models easily, instead of getting stuck using just one. 

Switching Models

Idit Levine leads a company called Solo.io. Her business helps big companies manage their AI systems safely. She’s watching changes happen with her own clients.

Many companies started out using big proprietary AI models. But now they’re asking a new question. Can they use an open source model instead, and just run it on their own servers?

According to Levine, an open source model can often do about 90 percent of what a big, expensive model can do. And it costs way less. 

Companies also get to control it fully, without handing their information over to anyone else. Solo.io works with major names like T-Mobile, ADP, and SAP. 

The Numbers

Vercel, a platform known for hosting websites, has added tools that let developers switch between AI models easily. 

And they’re seeing a lot more traffic heading toward open source options. In fact, open source models made up 29 percent of all traffic through Vercel’s AI gateway last month.

OpenRouter, another company that helps developers route their AI requests, is seeing something similar.