微软CEO：在AI时代，如何定义一家公司的护城河？

Original title: A frontier without an ecosystem is not stable

Original author: Satya Nadella, Microsoft CEO

Original translation: Peggy

Editor's note: Microsoft CEO Satya Nadella believes that the true competitiveness of enterprises in the AI era lies not in betting on the strongest model, but in their ability to distill their own workflows, domain knowledge, organizational judgment, and employee experience into a continuously evolving learning system. In other words, companies cannot simply purchase AI capabilities; they must possess their own 'learning loop' (a system where human experience, business processes, and model capabilities continuously reinforce each other).

Within this framework, future companies will simultaneously accumulate two types of capital: human capital, which includes the knowledge, judgment, relationship networks, creativity, and pattern recognition abilities of employees; and Token Capital (the AI capabilities that a company builds and owns itself). Nadella emphasizes that AI will not devalue human capital; instead, it will make human goal-setting, cross-domain connection, and critical pattern recognition even more important. Without human direction, computing power just spins its wheels; without an organization's own accumulated knowledge, even the strongest model is merely an external tool.

The core judgment of this article is: a frontier without an ecosystem is not a stable future. The value of AI should not be absorbed by a few general-purpose models. Instead, it should form a frontier ecosystem where every company, every industry, and every nation can possess its own learning loop. Companies need to build private evaluations, private reinforcement learning environments, and queryable knowledge bases, transforming tacit experience into reusable, scalable, and iterable systemic capabilities. The true moat may not be a specific model itself, but the accumulated 'company veteran' experience that a business retains, even if the underlying general-purpose model is swapped out.

This is also the key to enterprise sovereignty in the AI era: whoever can turn organizational knowledge into a system of continuous compounding interest will retain their IP, amplify employee capabilities, and keep the economic value generated by AI within their own business, industry, and community, even as models iterate rapidly.

The following is the original text:

I've been thinking a lot lately about what the future of the enterprise looks like in an AI-driven economy.

This transformation is unlike any previous platform shift. In the past, we used digital systems to augment human capital; this time, for the first time, we can create a true cognitive loop between humans and digital systems. This is a very disruptive concept because it changes how we understand 'work' itself within an organization.

The truly critical question isn't how a particular digital tool or system is used, but rather, how an organization continues to learn, accumulate intellectual property, differentiate itself, and thrive in a world where AI models can continuously absorb and commoditize human and organizational expertise.

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

Importantly, as token capital grows, human capital does not become less important. Quite the opposite, it becomes even more critical. I believe human agency will be the core driver of token capital growth. Humans set ambitious goals, connect dots across domains, build relationships, and identify truly significant patterns. Without human direction, computing power just spins its wheels.

This means the real opportunity isn't about choosing the best model, but building a learning loop on top of the model where human capital and token capital compound each other. You can outsource a task, you can even outsource a job, but you can never outsource your own learning. The future of a company lies in whether it can make this learning compound continuously between humans and AI.

This requires a new architectural approach: every company should be able to build intelligent agent systems that improve over time, while still retaining control over their own intellectual property. A company should be able to swap out a 'generalist' model without losing the 'company veteran' expertise accumulated in its learning system. This will be the key test of a company's control and sovereignty in the future era.

Companies need to turn their workflows, domain knowledge, and long-accumulated judgment into AI systems that improve with every use. Private evaluations should measure whether a model is genuinely getting better at the business outcomes the company cares about, not just external benchmarks. Private reinforcement learning environments should allow models to become stronger based on the organization's internal real-world trajectories. The enterprise knowledge base will make institutional memory queryable and improve the efficiency of token usage.

This loop will become the company's new intellectual property. I see it as a 'climbing machine.' And unlike most assets, it compounds. Every improvement in the workflow generates better training signals, which in turn accelerates the accumulation of the company's unique tacit knowledge. Companies that build this system earlier will gain an advantage that is hard to replicate, regardless of future breakthroughs in individual model capabilities.

The last thing we want is a world where every company across every industry cedes its value to a handful of models that consume everything they see. If all value ends up being captured by a few models, the political and economic structure simply won't tolerate that outcome. An AI future that hollows out entire industries cannot obtain social license.

Think about what happened in the first phase of globalization: entire industrial economies were hollowed out by outsourcing. On the surface, GDP numbers looked okay, but the real industrial transfer and job displacement were real, and their consequences are still being felt. We must not replicate this dynamic in the AI era – letting a few AI systems capture all the economic returns while the knowledge of entire industries is commoditized and hollowed out beneath them.

In my view, our priority must be to build a frontier ecosystem, not just a frontier model. Only then can value flow broadly to every company, every industry, and every nation. In such an ecosystem, every organization can possess its own learning loop, encode its institutional knowledge, and let human capital and token capital compound together.

This is also the platform spirit I've always believed in: the value created on top of the platform should be greater than the value captured by the platform itself; every company should be able to continuously innovate and create its own value.

When this is achieved, companies will create value for themselves and for the economic environment they operate in. The expertise of employees will be amplified, their judgment will become part of the system, replicable and scalable, and these gains will flow back to the company and its surrounding community.

This is how companies create value for themselves and the broader economy. This is the stable equilibrium we should collectively build.

Original link