You pay for an AI model subscription. You feed it your company's internal data. You get back answers that help your business run better. That sounds like a fair trade, right?
Not exactly, says Microsoft CEO Satya Nadella.
In a recent blog post, Nadella laid out a problem that a lot of companies might not see coming. When you use proprietary AI models - the kind made by labs like OpenAI and Anthropic - you are not just paying with cash. You are also handing over something harder to replace: your institutional know-how.
He put it bluntly. "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," Nadella wrote.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
The issue is the data you give the model. Every prompt you type, every correction you make when the model gets something wrong, every tool your AI agent uses - that all gets fed back to the model maker. Nadella called that data "exhaust," and he said the labs learn from it. "Every correction is distilled into institutional know-how," he added.
That know-how, he warned, is "the kind of knowledge a competitor could never buy." But with proprietary models, a competitor might not have to buy it. They could get it through the model itself.
Nadella is not the only one raising this flag. Venture capitalist Jason Calacanis and Palantir CEO Alex Karp have said similar things. But Nadella's warning carries extra weight because Microsoft is a major investor in both OpenAI and Anthropic.
So what is the alternative? Open-source AI models that companies can run on their own servers.
Idit Levine, founder of Solo.io, a company that makes networking and security software for managing AI systems, said her customers are already moving that way. "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," she said.
The numbers back up the trend. Vercel, a platform that added AI model-switching tools, reported that 29% of all traffic routed through its gateway came from open-source models last month. OpenRouter, a service that enables developers to direct queries to various AI models, also notes a rise in open-source usage.
Nadella's solution is threefold: retain ownership of your data, build proprietary learning environments on the cloud, and use "orchestration layers" that let you swap between different AI models rather than being locked into one. The Linux Foundation's Agentgateway project, which uses Solo.io's technology, is one example of that approach.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
