In a blog post published on Monday, Satya Nadella, chief executive of Microsoft, warned that organisations using AI are effectively paying twice. The first payment is the obvious monetary cost for tokens or AI services; the second, less visible cost is the proprietary knowledge companies relinquish when they feed models with data, prompts and corrections.
“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 wrote.
‘Exhaust’ and how enterprise know‑how leaks to models
Nadella argues the most dangerous aspect is that models learn the nuances of companies’ businesses from the usage ‘exhaust’: the prompts people write, the tools agents use, and especially the corrections made when models are wrong. Each correction, he says, is distilled into institutional know‑how — information a competitor could never buy, yet companies hand over in the normal course of using AI.
Distillation and the double standard
Nadella contends that if model providers can freely train on public internet data, it is fair for enterprises to study or “distill” those models in return. Distillation uses a model’s outputs to understand its behaviour and to train a new, often cheaper model. This issue has surfaced before: in February, Anthropic accused Chinese open‑source models of sending millions of prompts to Claude to improve their own systems and urged U.S. officials to tighten export controls.
His argument is that model makers cannot have it both ways — benefiting from permissive use of public data while then imposing restrictive terms on others who attempt to extract knowledge from their models. Nadella is especially concerned when providers reserve the right to learn from customer usage and interaction data.
Nadella’s recommendations: data ownership and orchestration
The remedies Nadella proposes align with what the CEO of a major cloud provider would recommend. He urges companies to retain ownership of their data — including prompts and feedback — and to build their own proprietary learning environments in the cloud where their data likely already resides. He also recommends building ‘orchestration layers’ to enable easy switching between AI models from different vendors, avoiding lock‑in. Tools such as AI gateways that route requests across models have grown in popularity for this reason.
Although Nadella does not explicitly call for open source, that approach is an implicit subtext: keeping control over created intelligence often means running open models on premises.
A shift toward on‑prem and open models
Many large companies that still maintain some of their own data centres alongside cloud usage are moving to on‑prem deployments of open‑source models. Idit Levine, founder and CEO of Solo.io, which makes networking and security software for enterprise AI management, says her customers follow this pattern: after experimenting with proprietary providers they ask, “Can I take an open‑source model and run it on‑prem? It will do almost 90% of what the big provider does and cost far less.” Solo.io’s technology was chosen last year for the Linux Foundation’s Agentgateway project, and the company counts T‑Mobile, ADP and SAP among its customers.
Other vendors see the same trend: Vercel (known for website hosting and development, which added AI model‑switching tools) and OpenRouter (which helps developers route requests across different models) report surging interest in open models. In fact, 29% of traffic routed through Vercel’s gateway last month went to open models.
Implications
With the CEO of Microsoft — a company that has invested in both OpenAI and Anthropic — publicly warning enterprises about the risks of proprietary models, the industry may see continued momentum toward data ownership, model portability and on‑prem or open solutions. As Nadella put it: “In consuming intelligence, you are creating intelligence. And what you create should belong to you.”
