The Token Is Not the Point

The Token Is Not the Point

7:23

Three companies went public in the span of two weeks. SpaceX raised $75 billion in the largest IPO on record. Anthropic filed confidentially, coming off a valuation that touched $965 billion. OpenAI followed days later, targeting somewhere north of $1.75 trillion.

None of this changed the models. None of it moved a data center. The infrastructure is where it was. The research is where it was. But something did shift — and it's worth being precise about what.

When a private company goes public, it crosses a line that has nothing to do with technology. It becomes legible to capital markets. It becomes something you can own a piece of without being an insider. It stops being a bet on the future and starts being a position in the present. That's what happened last week. Artificial intelligence — as an asset class, as an industry, as an infrastructure layer — became normalized. Not in the cultural sense. In the financial sense. It's on the tape now. And the moment that happened, two very different arguments went viral.

The first came from Satya Nadella. His piece, titled A Frontier Without an Ecosystem Is Not Stable, made the case that selling tokens isn't a business. It's a utility. The real game, he argues, is building the learning loops around those tokens — the organizational intelligence that compounds over time as humans and AI systems work together, correct each other, and accumulate proprietary context that nobody else has.

His framing: human capital defines direction, token capital fills the gaps. The two are intertwined. A company that outsources both to a frontier model has no moat. A company that builds the loop — where its own people, processes, and data continuously train the system on what actually matters to the firm — has something defensible. It's a smart argument. It's also a self-interested one. Microsoft sells the infrastructure that runs the ecosystem. Of course Satya wants the ecosystem to be the value layer.

But the underlying point holds regardless of the messenger. Tokens are already becoming a commodity. The cost curve on AI inference has been dropping at a rate that makes Moore's Law look leisurely. A capability that costs a dollar today costs ten cents in eighteen months. If your entire business model is charging for access to that capability, you're selling a product that reprices itself out of margin before you can build anything around it. The potato analogy isn't wrong. If the price of potatoes drops 90% year over year, you don't want to be in the potato business. You want to be in the business of whatever potatoes make possible.

The second argument was starker. Andrew Curran's piece said simply: the window has closed.

His thesis: if you wanted your nation to be at the AI frontier, you had three years to do it. February 2023 to February 2026. That was the window. It is now closed.

The reasoning is about compounding. Models like Mythos and Fable don't just outperform their predecessors — they accelerate the development of everything that comes after them. A lab that has Mythos-level capability can use that capability to build the next generation faster than any competitor starting from scratch. The gap between the frontier and everyone else isn't fixed. It widens with every cycle. For nations that missed the window — most of Europe, most of the world outside the US and a narrow band of competitors — the trajectory from here is falling further behind in each subsequent generation of models, not catching up.

It's a hard argument to dismiss. The compounding logic is real. RSI — the idea that AI systems capable enough can accelerate their own improvement — isn't hypothetical anymore. Anthropic published internal data showing Claude going from a 3x research optimization speedup to a 52x speedup in under a year, with 80% of their own production code now being written by Claude. If that curve continues, the advantage of being at the frontier in 2026 compounds into something that can't be closed by policy, by investment, or by political will alone.

Here's what both arguments are circling without quite landing on.

Tokens are infrastructure. Ecosystems are leverage. But neither of those is actually the scarce resource. The scarce resource is the compounding loop itself — the organizational and national capacity to build systems that get smarter from their own outputs. Satya is describing it at the firm level. Curran is describing it at the national level. They're pointing at the same thing from different altitudes.

What makes this moment genuinely different from previous technology transitions — the internet, mobile, cloud — is the speed at which the gap becomes structural. When broadband rolled out, a country that was five years behind could close that gap with infrastructure investment. The underlying technology didn't compound against them. It just waited.

This doesn't wait. Every month that a frontier lab operates at the Mythos or Fable level, the distance between that lab and the field grows. Not because the field is standing still. Because the frontier is accelerating.

The IPOs didn't cause any of this. But they made it legible. Capital markets now have a stake in the outcome. Pension funds, retail investors, sovereign wealth funds — they're all going to be holding positions in these companies. That changes the politics. That changes the urgency. That changes who pays attention.

The window Curran is describing wasn't closed by a single announcement. It closed incrementally, in the daily compounding of model capability against the rest of the field, until one day someone did the math and wrote it down. The IPOs just made it official.

So what do you do with this?

If you're a firm: Satya's argument is correct. The token is not the moat. The loop is. Build organizational intelligence that compounds — proprietary data, proprietary workflows, feedback systems that make your AI more useful to your business specifically over time. That's the asset. The token is just the fuel.

If you're a nation outside the frontier: the honest answer is that the Curran thesis deserves a serious response, not a political one. The question isn't whether to launch a sovereign AI initiative. Those are everywhere now. The question is whether you're building something that compounds, or something that performs competence for domestic audiences while the real gap continues to widen.

If you're paying attention to the capital markets: understand that what just went public isn't three companies. It's the infrastructure layer of the next fifty years of economic value creation. The question is whether the companies trading on top of that infrastructure understand the difference between riding the token wave and actually building the loop.

Most don't. Yet.

---

Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital, and CIO of Data Power Supply.