@dotey: Microsoft CEO Satya Nadella published a lengthy article proposing a new concept: Token capital. His core argument is that in the AI era, every company needs to simultaneously manage two types of capital. One is traditional human capital—employees' knowledge, judgment, relationships; the other is Token capital—the company's own...

X AI KOLs Following News
satya-nadella token-capital ai-economy enterprise-ai human-capital learning-loop

Summary

Microsoft CEO Satya Nadella proposed the concept of 'Token capital', arguing that in the AI era, enterprises need to simultaneously manage human capital and Token capital, and build a learning loop to accumulate proprietary AI capabilities, to prevent a few models from monopolizing value.

Microsoft CEO Satya Nadella published a lengthy article, proposing a new concept: Token capital. His core argument is that in the AI era, every company needs to simultaneously manage two types of capital. One is traditional human capital—the knowledge, judgment, and networks of employees; the other is Token capital—the AI capabilities that the company itself builds and owns. The two are not in a zero-sum relationship; the stronger human judgment is, the faster Token capital grows. Without human direction, computing power just spins idly. This sounds abstract, but Nadella provides a concrete litmus test: Can you swap out the underlying generalist model at any time without losing the company's accumulated proprietary experience? If yes, it means you truly own your AI capabilities; if not, it means you are just renting someone else's intelligence. He advises companies to transform workflows, industry knowledge, and decision-making experience into AI systems that can continuously improve, establishing private evaluation systems to measure model performance in actual business contexts, rather than just looking at public benchmarks. Once this learning flywheel starts turning, it's like compound interest—each improved workflow generates better training signals, further accelerating knowledge accumulation. Nadella also issued a politically charged warning. He draws an analogy with globalization: during the first wave of globalization, GDP numbers looked good, but entire industries were hollowed out by outsourcing, and the consequences are still visible today. If the AI era repeats this script, with a few models devouring all the knowledge and value across industries, "the political and economic system will not tolerate this outcome." --- Original Text Translation --- Frontier technology without ecosystem support is doomed to be unsustainable. Satya Nadella Lately, I've been pondering: In the AI-driven economic wave, where exactly does the future of enterprise lie? This transformation is completely different from any previous platform shift. In the past, we only used digital systems to enhance human productivity. But this time, we have, for the first time, established a true cognitive loop between humans and digital systems. This is a truly paradigm-shifting concept, because it fundamentally changes our definition of the nature of "work" within an enterprise. When AI models can continuously absorb human and organizational expertise and turn it into a commoditized cheap commodity (that is, turning originally scarce professional skills into universally accessible general capabilities, thereby weakening companies' core moats), the real crisis emerges. The key challenge we face is no longer just how to use a certain digital tool or system, but how enterprises can continuously learn, accumulate intellectual property (IP), maintain uniqueness, and thrive in this new world. Every company must build two types of capital: one is what we know as "human capital," and the other I call "Token capital." Human capital encompasses employees' knowledge, judgment, networks, creativity, and ability to recognize patterns; Token capital refers to the AI capabilities that the enterprise itself builds and controls (here, the term "Token capital" is quite vivid, because the fundamental unit of information processing for large language models (LLMs) is the token). It must be emphasized that as Token capital grows, human capital does not depreciate. On the contrary, it becomes more valuable than ever! I firmly believe that human agency will be the core engine driving the growth of Token capital. Humans are responsible for setting ambitious goals, connecting dots across domains, building networks, and discerning the most critical patterns. Without humans guiding the direction, all that powerful computing power is just spinning its wheels. This means that the real opportunity is not about picking the "best" model on the market, but about building a "learning loop" on top of the model that can compound human capital and Token capital. You can outsource a specific task or even an entire job role, but you absolutely cannot outsource the "learning ability." The future core competitiveness of an enterprise lies in its ability to continuously accumulate and amplify this learning capability between humans and AI. This requires a new architectural mindset: every enterprise must be able to build AI agentic systems that can self-iterate over time, while maintaining tight control over its intellectual property. A company should be able to swap out the underlying "generalist model" at any time without losing the rich professional experience that has been accumulated in the system, like seasoned company veterans. In the future, this will be the key "litmus test" for whether an enterprise has data control and technological sovereignty. Enterprises need to transform their workflows, domain knowledge, and years of accumulated judgment into AI systems that evolve with every use. Companies should establish private evals (i.e., internal model capability test standards tailored to their actual business scenarios) to check whether the model is truly making progress on outcomes that matter to the enterprise, rather than blindly relying on public benchmarks for self-congratulation! A dedicated reinforcement learning environment should allow the model to become more powerful by absorbing real business data and work traces from the organization. Such a proprietary knowledge base makes the company's organizational memory instantly retrievable, while also greatly improving the efficiency of token usage. This loop will become a new form of intellectual property for the enterprise. I envision it as a hill climbing machine. And unlike most assets, it has powerful compound effects. Each optimized workflow generates better training signals, thereby accelerating the accumulation of the enterprise's unique tacit knowledge. Companies that invest early in building such a loop will gain a moat that is difficult to replicate, and no matter what capability-exploding new models appear on the market in the future, they will not be easily dislodged. The last thing we want to see is companies in every industry ceding value to a few ravenous giant models. If all economic value is monopolized by just a few models, the political and economic system will absolutely not tolerate it. Society will also never allow an AI future that hollows out entire industries. Recall what happened in the early days of globalization: massive business outsourcing hollowed out many industrial economies. On the surface, GDP data still looked good, but the stark reality was the displacement of countless industrial workers, and the severe consequences are still with us today. We must not let this tragedy repeat in the AI era—we must not allow a few AI systems to capture all the economic rewards while entire industries watch helplessly as their hard-earned expertise is mercilessly commoditized. In my view, our pressing priority is not just to build frontier models, but also to foster a thriving "frontier ecosystem." Only then can value flow like living water, broadly reaching every company, every industry, and every country. In this ecosystem, every organization can have its own learning loop, embed organizational wisdom within it, and achieve snowballing growth of both human capital and Token capital. This is also the core philosophy that has accompanied me throughout my career: a true platform enables the value built on top of it to far exceed the value it captures itself. In such an ecosystem, every company can continuously innovate and build its own genuine value. When this is realized, enterprises can create huge dividends not only for themselves but also for the entire surrounding economy. Employees will see their professional skills amplified infinitely, and their personal judgment will be embedded into the system, becoming replicable and scalable. And the benefits of all this will ultimately flow back to the enterprise and the broader communities they serve. This is the right way for enterprises to create value for themselves and the macroeconomy. This is also the stable and enduring ecological balance we should build together.
Original Article
View Cached Full Text

Cached at: 06/15/26, 10:59 AM

Microsoft CEO Satya Nadella published a lengthy essay introducing a new concept: Token Capital.

His core argument is that in the AI era, every company needs to manage two kinds of capital simultaneously. The first is traditional human capital—employee knowledge, judgment, and networks. The second is token capital—the AI capabilities a company builds and owns for itself. These are not a zero-sum trade-off: the stronger human judgment becomes, the faster token capital grows. Without human direction, computing power just spins its wheels.

This may sound abstract, but Nadella offers a concrete litmus test: Can you swap out the underlying general-purpose large model at any time without losing the proprietary experience your company has accumulated? If yes, you truly own your AI capabilities. If not, you’re just renting someone else’s intelligence.

He recommends that companies transform their workflows, domain knowledge, and decision-making experience into continuously improvable AI systems. Companies should establish private evaluation systems to measure model performance in real business scenarios, rather than relying solely on public benchmark scores. Once this learning flywheel starts turning, it acts like compound interest: each improved workflow generates better training signals, which further accelerate knowledge accumulation.

Nadella also issued a politically charged warning. He draws a parallel with globalization: during the first wave, GDP numbers looked great, but entire industries were hollowed out by outsourcing, and the consequences are still unfolding. If AI repeats this script, with a handful of models consuming all the knowledge and value across industries, “the political and economic system will not tolerate that outcome.”

Translation of the original essay:

Frontier Technology Without Ecosystem Support Is Doomed to Falter

Satya Nadella

Lately, I’ve been thinking deeply: Where does the future of the enterprise lie in the wave of an AI-driven economy?

This transformation is fundamentally different from any previous platform shift. In the past, we used digital systems to augment human productivity. But this time, for the first time, we are establishing a true cognitive loop between humans and digital systems. This is a mind-bending concept because it completely changes how we define “work” inside an enterprise.

When AI models can continuously absorb human and organizational expertise and turn it into a cheap, commoditized resource—essentially turning scarce specialist skills into universally accessible generic capabilities, thereby eroding a company’s core moat—the real crisis emerges. The key challenge is no longer simply how to use a digital tool or system. It’s about how an enterprise can continue to learn, accumulate intellectual property (IP), remain unique, and thrive in this new world.

Every company must build two kinds of capital: the familiar human capital, and what I call token capital. Human capital encompasses employees’ knowledge, judgment, networks, creativity, and pattern-recognition abilities. Token capital refers to the AI capabilities that a company builds and controls itself. (The term “token capital” is vivid because the fundamental unit of information processing in large language models is the token.)

It’s critical to stress that as token capital grows, human capital does not depreciate. On the contrary, it becomes more valuable than ever! I firmly believe that human agency will be the core engine driving token capital growth. Humans set ambitious goals, connect dots across domains, build relationships, and discern the most critical patterns. Without humans steering the direction, all that powerful computing just spins in place.

This means the real opportunity isn’t about picking the “best” model on the market. It’s about building a learning loop on top of models that compounds human capital and token capital. You can outsource a task or even an entire role, but you can never outsource the ability to learn. The future competitive advantage of an enterprise lies in its ability to continuously accumulate and amplify this learning capability between humans and AI.

This requires a new architectural mindset: every enterprise must be able to build agentic systems that evolve over time, while maintaining full control over its own IP. A company should be able to swap out the underlying “generalist model” at any time without losing the rich, veteran-level expertise embedded in its systems. In the future, this will be the definitive touchstone for whether a company has true data control and technological sovereignty.

Enterprises need to transform their workflows, domain knowledge, and years of accumulated judgment into AI systems that self-improve with every use. Companies should establish private evals—internally tailored model capability benchmarks for their own real business scenarios—to test whether a model is actually making progress on outcomes that matter to the business, rather than blindly cheering at public benchmarks! A dedicated reinforcement learning environment should allow models to grow stronger by absorbing real business data and work traces from within the organization. Such proprietary knowledge bases make organizational memory instantly retrievable while dramatically improving token efficiency.

This cycle becomes the company’s new intellectual property. I imagine it as a hill-climbing machine. And unlike most assets, it exhibits powerful compound returns. Every optimized workflow generates better training signals, accelerating the accumulation of the enterprise’s unique tacit knowledge. Companies that invest early in building this cycle will create a moat that is hard to replicate, regardless of what impressive new models appear on the market.

The last thing we want is for every company across every industry to cede value to a few rapacious, all-consuming megamodels. If all economic value is monopolized by a handful of models, the political and economic system will not tolerate it. Society will never accept an AI future that hollows out entire industries.

Think back to the early days of globalization: massive outsourcing of business operations hollowed out many industrial economies. On paper, GDP numbers looked fine, but the reality of millions of displaced industrial workers was brutal, and the severe consequences still linger. We must not let this tragedy repeat in the AI era—we cannot let a few AI systems capture all the economic returns while entire industries watch their hard-earned expertise get commoditized.

In my view, our priority is not just building frontier models. It’s about building a thriving frontier ecosystem. Only then can value flow like living water, broadly reaching every company, every industry, and every country. In this ecosystem, every organization can own its learning loop, embed organizational wisdom, and let human capital and token capital grow together like a snowball.

This is the core philosophy that has accompanied me throughout my career: a true platform enables far more value to be created on top of it than the platform itself captures. In such an ecosystem, every company can keep innovating and build its own true value.

When this happens, companies create enormous dividends not just for themselves but for the entire surrounding economy. Employees will see their skills amplified infinitely; their judgment will be embedded into systems, becoming replicable and scalable. And the benefits will ultimately flow back to the enterprise and the broader communities they serve.

This is the right way for companies to create value for themselves and for the macroeconomy. This is the stable, enduring equilibrium we should build together.

Similar Articles

A frontier without an ecosystem is not stable (5 minute read)

TLDR AI

Microsoft CEO Satya Nadella argues that in the AI-driven economy, firms must build both human capital and token capital (AI capabilities) in a compounding learning loop, emphasizing that human agency remains crucial and that companies must retain control over their IP to avoid value being captured by a few frontier models.