Nadella Says Private Evals May Become Companies' Most Valuable IP
"I think about it, like what's that private eval that you can then use even a frontier model to hill climb on and not leak the traces, maybe one of the biggest drivers of IP. In fact, there may be— like human capital was never possible to go put on a balance sheet. Whereas now I think you can with the agents that have learned through the traces."
About this episode
Microsoft CEO Satya Nadella joined hosts Swyx Patel and Elad Gil for a crossover episode of No Priors and Latent Space following Microsoft Build 2025, delivering one of his most detailed explanations yet of how enterprises will capture value in the AI era. Nadella argued that private evaluation systems and institutional knowledge captured through agent traces will become the most valuable intellectual property companies possess—potentially important enough to appear on corporate balance sheets where human capital never could. He revealed Microsoft built more Azure infrastructure capacity in the last 15 months than in its first 15 years combined, an unprecedented buildout reflecting massive AI compute demands. Nadella positioned Microsoft's strategy around enabling every company to operate at the frontier with their own intelligence rather than depending on a single model provider, using multimodal harnesses, proprietary tools, and context layers. He shared how Microsoft's own Azure networking team transformed their work by building an agentic system called Miles, shifting from performing operations to creating meta-systems that automate work—and requesting token budgets instead of headcount. On business models, Nadella described a paradox where customers enthusiastically support outcome-based pricing until they achieve results, then resist sharing value with vendors. He warned the tech industry can no longer rely on promises of future benefits and must deliver tangible community-level outcomes including jobs, energy grid improvements, and measurable economic value, or risk losing social and political permission to continue. The conversation covered Microsoft's MAI model training strategy emphasizing clean lineage and hill-climbing scaffolds, the unbundling and rebundling of SaaS applications in an agentic world, and predictions for new types of engineering roles focused on agent management and infrastructure science.
Key takeaways
- Nadella revealed Microsoft built more Azure capacity in 15 months than its first 15 years combined to meet AI compute demands.
- Private evaluation systems for AI models may become the most valuable IP companies possess, potentially appearing on balance sheets according to Nadella.
- Microsoft's Azure networking team reconceptualized their work as building agentic systems rather than performing operations, requesting tokens instead of headcount.
- Nadella warned tech companies must deliver tangible community benefits from AI or lose social and political permission to continue infrastructure buildouts.
- Microsoft is positioning as an enablement platform where every company can operate at the frontier with proprietary intelligence using multimodal harnesses.
- Outcome-based pricing faces a paradox where customers resist sharing value with vendors once they achieve results, favoring subscription models instead.
- Traditional SaaS applications will unbundle into data models, business logic, and UI layers that get reassembled in agentic workflows with new business models.