An article written by Microsoft CEO Satya Nadella,

An article written by Microsoft CEO Satya Nadella.

This kind of article is worth reading. Don’t go read what most AI-generated content from Chinese-language and overseas Chinese writers put out.

Satya:

I’ve been thinking a lot about where companies should be headed in an AI-driven economic environment.

This transformation is radically different from any platform shift we’ve seen before. In the past, we used digital systems to enhance human capital. Now, for the first time, we can establish a true cognitive feedback loop between people and digital systems. That’s refreshing, because it completely changes how we understand work inside organizations.

The key isn’t about certain digital tools or systems—or how we use them. The key is how organizations continue to learn, build intellectual property, differentiate themselves, and thrive in a world where AI models can continuously absorb professional knowledge from humans and organizations and commoditize it.

Every company must build what I call human capital and token capital. Human capital includes employees’ knowledge, judgment, relationships, creativity, and pattern recognition capabilities, while token capital is the AI capability that a company builds and owns.

What matters is that as token capital grows, the value of human capital won’t decline—it will increase! I believe human agency will be the driver of token capital growth. Humans will set ambitious goals, connect information across different domains, build relationships, and identify the most important patterns. Without human guidance, computers will just go around in circles.

This means the real opportunity isn’t choosing the best model—it’s building a model-based learning loop so that human capital and token capital can grow through compounding. You can outsource a task, even an entire job, but you can never outsource learning. The future of enterprises lies in whether learning outcomes can compound as people and AI grow together.

This requires a completely new architectural approach, one that enables every business to build intelligent systems that improve over time while still maintaining control over its intellectual property. Companies should be able to replace existing “generic” models without losing the “company veterans” expertise embedded in their learning systems. This will be the key “test” of corporate control and autonomy in the era ahead.

Enterprises need to translate their workflows, domain knowledge, and accumulated judgment into AI systems—and ensure those systems keep improving with every use. Private evaluations should be able to capture whether models truly improved the results that matter critically to the business (not just external benchmarks!). A private reinforcement learning environment should allow models to grow continuously based on real data from within the organization. Its knowledge base makes institutional memory queryable and improves the efficiency of token usage.

This loop will become the company’s new intellectual property. I compare it to a machine that climbs a mountain. Unlike most assets, it has a compounding effect. Every improvement to the workflow generates better training signals, accelerating the accumulation of a company’s unique, tacit knowledge. Companies that build this loop early will have advantages that are hard to replicate, regardless of what new single-model capabilities they may have.

What we least want to see is that all industries and all companies hand over value to a small number of model players that take everything. If all value concentrates in the hands of a few models, political and economic systems simply cannot tolerate it. Society will never allow the future of AI to hollow out an entire industry.

Think about what happened in the first phase of globalization: outsourcing hollowed out entire industrial economies. On the surface, GDP data looked fine, but industrial reallocation was real, and its consequences are still showing up to this day. We must not let this pattern repeat in the AI era—where a small number of AI systems capture all economic benefits while the industry watches helplessly as its knowledge is commoditized and ultimately destroyed.

I believe our first priority must be to build a frontier ecosystem, not just a frontier model—so that value can flow broadly across every company, every industry, and every country. In this ecosystem, every organization can own a learning loop that encodes its institutional knowledge, allowing it to continuously accumulate human capital and token capital.

I’ve held this belief since I was young: platforms can create more additional value than what the platform itself can provide, and every company can keep innovating and creating its own value.

When that happens, enterprises don’t just create value for themselves—they also create value for the surrounding economy. Employees’ expertise will be enhanced; their judgment will be integrated into replicable and scalable systems, and both the company and the broader community around it will benefit.

That’s how companies create value for themselves and for the wider economy. And that’s the stable balance we should be building together.