Microsoft CEO Satya Nadella has criticized the business practices of leading artificial intelligence developers, arguing that frontier AI companies are applying a double standard by relying on publicly available data to train their models while restricting others from using distillation techniques to build competing systems.
In a post on X on Sunday, Nadella questioned what he described as an imbalance in the AI ecosystem, where model developers benefit from broad access to public information but seek to prevent others from learning from their models.
“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, and to reserve the right to learn from customer usage and interaction data,” Nadella wrote.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
He added that if “learning only flows in one direction,” the companies controlling AI infrastructure would capture most of the economic value while the creators of the underlying knowledge receive little in return.
Although Nadella did not name any company, his remarks appeared to target Anthropic, which has been among the most vocal critics of model distillation. Distillation is a technique that enables developers to train smaller or cheaper AI models using the outputs of more advanced systems, significantly reducing the time and computing resources required to build competitive models.
The comments come amid escalating tensions between major AI developers over intellectual property, data usage and competition, as governments in the United States and China tighten scrutiny of frontier AI technologies.
Anthropic has repeatedly argued that unrestricted distillation threatens innovation by allowing competitors to replicate years of research at a fraction of the cost. In February, the company said distillation enables rivals to acquire “powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.”
The debate intensified last month when Anthropic accused Alibaba of carrying out what it described as “the largest known distillation attack” against the company to date in a letter sent to U.S. lawmakers. Anthropic alleged that Chinese firms had attempted to extract capabilities from its Claude models, though Alibaba did not publicly respond to the accusations.
Nadella’s criticism also touches on a broader legal and ethical dispute surrounding how frontier AI models are developed. Companies including Anthropic, OpenAI and Google DeepMind train their systems using vast amounts of publicly available text, images and other online content. That practice has triggered numerous copyright lawsuits from publishers, authors, artists, and media organizations, who argue that their work has been used without permission or compensation.
Microsoft itself has largely avoided positioning Azure as a developer of proprietary frontier models, instead emphasizing its role as an infrastructure provider. Since OpenAI ended Microsoft’s exclusive cloud hosting arrangement earlier this year, the company has increasingly adopted a more model-agnostic strategy, supporting a growing portfolio of AI models through Azure AI Foundry, including offerings from OpenAI, Meta, Mistral, xAI, DeepSeek and other developers.
Nadella’s latest comments reinforce that strategy by encouraging enterprises to reduce dependence on any single AI model provider.
He argued that businesses should own their AI infrastructure, retain control over their proprietary knowledge and establish independent evaluation and learning systems instead of relying entirely on external foundation models.
According to Nadella, organizations should build their own “learning loop,” allowing AI systems to continuously improve using enterprise-specific knowledge while maintaining strict control over sensitive data.
“That is why enterprises need a real trust boundary for their human capital and token capital to compound,” he wrote. “And it is a hard boundary across which nothing crosses, not even the intelligence exhaust, without consent.”
His reference to “intelligence exhaust” reflects growing concerns among enterprise customers that prompts, usage patterns, and model interactions could become valuable training data for AI providers. While major AI companies maintain enterprise privacy commitments, businesses continue to seek stronger guarantees that proprietary information will not be used to improve third-party models.
The issue has become more important as corporations deploy generative AI across software development, legal services, finance and healthcare, where sensitive commercial data represents a key competitive asset.
Nadella’s comments also echo criticism from other industry leaders. Palantir CEO Alex Karp recently criticized the industry’s “tokenmaxxing” business model, arguing that enterprises should control their own compute infrastructure, models and data rather than depend on external AI providers. Elon Musk has similarly accused Anthropic of using copyrighted material to train its models while opposing the use of distillation by competitors.
The dispute highlights a growing divide within the AI industry over who ultimately owns the value generated by artificial intelligence. Frontier model developers believe that restricting distillation is necessary to protect billions of dollars invested in research and computing infrastructure. Infrastructure providers and enterprise customers contend that organizations deploying AI should retain ownership of the data, workflows, and institutional knowledge they generate while using these systems.