On this page
On June 14, 2026, Microsoft CEO Satya Nadella published an essay arguing that every company now needs two kinds of capital. One is human capital. The other he called token capital: the proprietary AI capability a firm builds and owns, distinct from the tokens it merely rents. The essay drew more than 28 million views, and within days it triggered a wave of commentary explaining what he meant.
Almost none of that coverage answers the harder question every leader is actually asking: is this real, and if so, what does it take to build? This is a guide for that reader. What the term means, why the economics back it up, and what building it actually requires.
What is token capital?#
Token capital is the proprietary AI capability a company owns: its structured data, documented processes, and the history of decisions it has made, organized so that AI can put them to work. As Nadella put it, "Human capital does not become less valuable as token capital grows. It only becomes more valuable!" The two compound together, not against each other.
The distinction that matters is token capital versus token spend. Token spend is a line item: money paid to an AI provider for each request, gone the moment the answer comes back. Token capital is an asset: the trusted data and decision history a company keeps, reuses, and improves over time. A company can spend enormous sums on AI and build no capital at all, if none of that usage gets captured into something the business owns and can draw on again next quarter.
Why the data backs this up#
It's tempting to file "token capital" under CEO-essay hype and move on. Two things argue otherwise.
First, data itself has quietly become a formally recognized capital asset. Intangible assets such as data, software, brand, and institutional know-how now make up roughly 92% of S&P 500 market value, up from about 17% in 1975. In March 2025, the UN Statistical Commission adopted the 2025 System of National Accounts, which creates a new formal asset category for "data and databases," alongside software and R&D. Accounting standards rarely move fast. When they do, they're catching up to something that already happened.
Second, not all data clears the bar to count as capital. It has to be controlled, put to a use that generates real economic benefit, and clearly bounded. A pile of ungoverned exports sitting in a shared drive isn't a capital asset. It's a liability with a login.
The data behind token-based competition#
Boston Consulting Group's 2026 analysis, The Era of Token-Based Competition Is Here, studied 107 public technology companies. Those in the highest token-use quintile grew revenue by a median 16.5% year-over-year. Those in the lowest-usage group grew only 5.1%, less than a third as fast.
BCG's explanation is close to Nadella's. When frontier intelligence is a commodity every competitor can rent, the advantage shifts from having intelligence to applying it productively inside a specific business. To measure that, BCG proposes tracking Return on Intelligence (ROInt): the value of AI-powered output divided by the combined cost of labor and tokens. Companies that track only labor savings use the traditional "AI ROI" math. That math misses the new revenue and faster decisions a well-built system also produces, so those companies under-invest in the exact thing that's compounding for their competitors.
The part most companies are missing: the "why" behind decisions#
If token capital is more than a big pile of data, what is it actually made of? The most valuable part isn't the raw data at all. It's the record of why decisions got made. Not merely that a 15% discount was approved, but under which policy, with which exception, and based on what precedent.
Most business systems, such as CRMs, ERPs, and data warehouses, capture the what: the final state of a record after a decision lands. They rarely capture the why. That reasoning usually lives in Slack threads, email chains, and the memory of whoever made the call. So when a company points AI at a new, unfamiliar situation, it has almost nothing real to reason from. Feed it years of accumulated decisions and their reasoning, though, and it can apply genuine precedent instead of guessing.
Foundation Capital's research makes the structural point: the systems companies already own were built to record outcomes after the fact, not to capture the reasoning in the moment a decision is made. A company that captures that context as decisions happen builds an advantage that is genuinely hard for a competitor to reconstruct later. And the gap between a company with five years of that history and one starting from zero only widens.
What it actually takes to build#
There's no shortcut that skips the plumbing, but the essentials are simpler to state than the vendor landscape suggests. Building token capital comes down to three things:
- Get your data into one place you can trust, rather than scattered across systems that each tell a slightly different story.
- Agree on what it means. The single most valuable and most-skipped step is a shared, governed definition of your core business terms. What counts as an "active customer," a "qualified deal," "revenue," so that every dashboard, every team, and every AI system uses the same one. Without that agreement, AI guesses at your business from raw field names, and guesses differently each time.
- Make it usable by AI, and let it learn. Give AI a governed way not just to read that data but to act on it, so that every decision it helps make becomes context the next decision can draw on.
The order matters, and so does treating these as one coherent system rather than three tools bought separately and wired together afterward. That's the difference between capital that compounds and a bigger pile of data.
The takeaway for leaders#
The practical takeaway isn't a technology roadmap. It's a shift in how you think about AI spend. Every dollar you put toward AI either evaporates as token spend or accrues as token capital, and the difference is entirely whether your organization is capturing what it learns. The companies pulling ahead aren't the ones with the biggest AI budgets. They're the ones turning that budget into an asset they own.
The encouraging part is that you don't have to solve it all at once. Start with the step that's cheapest to get wrong and most expensive to skip: agreeing on what your core data actually means, over the data you already have. Everything else compounds from there.
This is the approach we take with RevOS, which builds these layers as one system rather than something assembled after the fact. If you'd like to talk through what building token capital would look like against your own data, schedule a call with our team.
Read more about revenue operations, growth strategies, and metrics in our blog and follow us on LinkedIn and Youtube.
All articles