Last night, Microsoft CEO Satya Nadella posted a lengthy thread on X titled "A frontier without an ecosystem is not stable".
This thread attracted massive attention as soon as it went live, and its view count has now surpassed 28 million.
Original post: https://x.com/satyanadella/status/2066182223213293753
Nadella profoundly explores the future form of enterprises, digital sovereignty, and the socio-economic impact of AI development in an AI-driven economic era.
He introduces two concepts: "Human Capital" and "Token Capital".
In his view, AI is changing the underlying logic of corporate competition. In the past, companies bought or built digital tools to enhance human efficiency: tools amplified people, and people created value. But today, AI models themselves possess the ability to absorb human expertise and "commoditize" it. A model, after learning enough enterprise data, can turn the unique skills originally belonging to a company into a standard service accessible to everyone.
In this way, AI seems to have the ability to devour other software services, or as Nadella puts it, "every company in every industry is ceding value to a handful of voracious, all-consuming models."
This also worries him greatly. In the text, he calls for "We cannot allow a handful of AI systems to capture all the economic returns."
Interestingly, world's richest man Elon Musk, who often mocks Microsoft, commented with a sarcastic tone that the article was interesting:
After all, this blog post seems to validate a comment Musk made in August last year: "OpenAI is going to eat Microsoft alive."
At that time, Nadella had just announced that GPT-5 would be fully integrated into Microsoft's Copilot, GitHub, and Azure product lines. His response was then light and confident: "People have wanted to eat us for 50 years, and that's what makes it fun!"
The implication of Musk's words is clear: Microsoft's proud software empire is handing over its own moat to a partner. In the field of AI, relying on someone else's model will eventually be replaced by that model.
Today's post indicates that Nadella no longer possesses that same confidence.
The following is a full translation of Nadella's post:
I've been thinking about the future of the enterprise in an AI-driven economy.
This shift is unlike any previous platform transition. In the past, we used digital systems to augment human capital. For the first time, we can establish a genuine cognitive loop between humans and digital systems. This is deeply disruptive because it changes how we conceive of work inside the enterprise.
The key issue transcends the use of certain digital tools or systems. What we need to focus on is how organizations can continue to learn, build intellectual property, stay differentiated, and thrive in a world where AI models can continuously absorb and commoditize the expertise of humans and organizations.
Every company must build what I call "Human Capital" and "Token Capital":
Human Capital encompasses employees' knowledge, judgment, relationships, creativity, and pattern recognition abilities.
Token Capital represents the AI capabilities that a company builds and owns.
Importantly, the value of Human Capital does not decrease as Token Capital grows. It only becomes more valuable! I believe human agency will be the driver of Token Capital growth. Humans will set audacious goals, connect dots across domains, build relationships, and identify the most critical patterns. Without human direction, compute just spins its wheels.
This means the real opportunity is not in picking the best model. We should build a learning loop on top of models where Human Capital and Token Capital compound. You can outsource a task, even a job, but you can never outsource your learning process. The future of a firm depends on its ability to accumulate this learning compound between humans and AI.
This requires a new kind of architectural approach, one where every enterprise can build agentic systems that evolve over time while maintaining control over their intellectual property. A company should be able to swap out a "general" model while retaining the "corporate veteran" level expertise accumulated in its learning system. This will be a crucial "test" of your control and sovereignty in the era to come.
Enterprises need to turn their workflows, domain knowledge, and accumulated judgment into AI systems that improve with every use. Private evaluation frameworks should capture whether models are truly getting better at the outcomes that matter most to the enterprise (external benchmarks aren't enough!). Private reinforcement learning environments should make models stronger on the execution trajectories real to the organization. Its knowledge base makes institutional memory queryable and token use more efficient.
This loop becomes the new intellectual property of the enterprise. I think of it as a hill-climbing machine. And unlike most assets, it compounds. Every improved workflow generates better training signals, accelerating the accumulation of tacit knowledge unique to the firm. Companies that build this early will have an advantage that's hard to replicate, no matter what new capabilities individual models come up with.
We cannot have a world where every company in every industry cedes value to a handful of voracious, all-consuming models. Assume all value gets captured by a few models, the political economy won't tolerate it. Society won't accept an AI future that hollows out entire industries.
Think back to what happened in the first phase of globalization when whole industrial economies were hollowed out by offshoring. Even though GDP numbers looked good on paper, the job losses were real and the consequences are still with us today. We cannot bring that dynamic into the AI era, with a few AI systems capturing all the economic returns while whole industries watch their knowledge get commoditized without even realizing it.
In my view, our imperative must be to build a frontier ecosystem. Merely building a frontier model isn't enough. Only then will value flow broadly across every company, every industry, and every country. In this ecosystem, every organization can own the learning loops that encode its institutional knowledge and compound its Human and Token Capital.
This is the ethos I grew up with: platforms should enable more value creation on top than they capture inside, so every company can innovate and create its own value.
When that happens, enterprises will create value for themselves and the economies around them. Employees will see their expertise amplified, their judgment becoming part of a system that's replicable and scalable. And the gains will accrue to the businesses and communities around them.
This is how enterprises drive value for themselves and the broader economy. And it's the stable equilibrium we should build together.
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), author: Machine Heart
