Key Takeaways:
The future of the firm depends on building AI learning loops, not picking the best foundation model, Microsoft's CEO argues.
Satya Nadella wants every company to build its own AI model. In an interview published Friday, Microsoft's chief executive told Applied Compute cofounder Yash Patil that organizations should create models tailored to their unique business data and context, warning that relying on a handful of frontier AI providers amounts to outsourcing a company's ability to learn.
"My simple thing is there should be as many models in the world as firms in the world," Nadella said. "Because after all, what is a firm? A firm is a learning system."
The comments represent one of Nadella's clearest articulations yet of enterprise AI strategy. Microsoft has increasingly embraced a multi-model approach through Azure AI Foundry, which hosts models from OpenAI, DeepSeek, Cohere, and others rather than relying solely on its $13 billion OpenAI partnership. Amazon has pursued a similar path with Bedrock, while Google Cloud offers third-party and proprietary models alongside Gemini.
"You can always buy a tool, you can even outsource a task or even a job, but you can't outsource your learning," Nadella said. "If you outsource your learning, then why exist?"
The learning loop, not the model
In a separate essay published on X, Nadella argued that the real opportunity lies not in selecting the best model but in building what he called a "learning loop" where human capital and token capital compound over time. The durable asset, he said, is not the model itself but the system surrounding it — one that retains what he termed "company veteran" expertise even when the underlying model is swapped.
This marks a departure from the past two years of enterprise AI, where conversations have centered on model capability: which model reasons better, which writes better code, which tops the benchmark rankings. As frontier models from OpenAI, Anthropic, Google, and Meta continue to improve rapidly, the intelligence layer is becoming abundant. Nadella's argument shifts the question from "which model is smartest" to "how does intelligence get organized, deployed, and continuously improved inside the enterprise."
The concept mirrors previous platform transitions. Companies did not rebuild their ERP systems every time databases improved, nor did they redesign CRM strategies when processors became faster. The durable value lived above the infrastructure layer. Nadella is arguing the same principle applies to AI.
The economic case against concentration
Nadella also warned that a world where all value accrues to a small number of foundation models is neither economically nor politically sustainable.
"It can't be, 'Hey, look, I have two frontier models or three frontier models' or whatever, some finite set that have learned everything that is differentiated today in the economy because then it collapses," he said.
The concern is not purely philosophical. Big tech firms that nurtured frontier AI companies are now facing a dilemma. Microsoft, Amazon, and Google have poured tens of billions into data center infrastructure to support AI model training, yet the most advanced models are increasingly encroaching on their core businesses — from programming assistants to office software. According to Wall Street forecasts, free cash flow at the four largest US technology firms is expected to hit its lowest level since 2014 this year, squeezed by AI infrastructure spending.
Microsoft is already shifting strategy. The company has launched Copilot Cowork, an AI agent product for office workers, and is considering integrating China's DeepSeek — a low-cost, open-weight model — into its platform. The goal is to offer affordable token pricing that makes AI accessible to average users rather than competing solely on frontier model capability.
For investors, the implications cut both ways. Microsoft shares have underperformed this year as the market weighs the cost of AI infrastructure against uncertain revenue returns. A world where every company builds its own AI model would expand the total addressable market for cloud platforms like Azure, but it would also commoditize the model layer, potentially compressing margins for frontier AI providers. The companies that ultimately define the next phase of enterprise AI may not be the ones building the most powerful models — they may be the ones building the systems that allow every organization to convert intelligence into compounding institutional knowledge.
This article is for informational purposes only and does not constitute investment advice.
