Ray Dalio writes about the Dutch East India Company the way you’d write about a species discovering fire. “The Dutch created the world’s first mega-corporation,” he notes in Principles for Dealing With the Changing World Order, one that “accounted for about one-half of all world trade.” What made that possible wasn’t the ships, and it wasn’t the trade routes, which other powers also had. It was the new financial instruments invented to fund the voyages. Capitalism itself, assembled on the fly to serve an ambition that had outgrown the tools available to it. The instruments came first. The entity that would use them at civilization scale came second, and arrived larger than anyone had planned for.
That sequence has been sitting differently this week, because the instruments are being invented again in public and the entities that will use them are still doing their paperwork. SpaceX and OpenAI are reportedly preparing public offerings that could break records for size, and the commentary arriving alongside that news is exactly what you’d expect. Analysts told CNBC that a flurry of mega-IPOs could signal a market top, with liquidity seeking an exit and insiders cashing out before the tide turns. It is a reasonable frame with real history behind it. It is also the most surface-level read available, and surface reads have a habit of missing what’s being built in the infrastructure underneath them.
What a Market Top Doesn’t Explain
Consider what else landed in the same week those IPO headlines did. Amazon introduced AgentCore Payments for autonomous transactions, a system designed to let AI agents buy things on their own without stopping to ask a human to approve each purchase. MoonPay launched an institutional trading platform, which is a deeper push into institutional crypto execution and, read a step further, rails built so that serious money and eventually machine money can move through these markets with less drag. Circle kept expanding the settlement layers USDC reaches, which sounds like housekeeping until you ask what a stablecoin is actually for when the counterparty isn’t a person. None of these are separate stories. They are different camera angles on one thing being assembled.
The mega-IPO argument is about valuation timing. It asks whether we are near the peak of what the market will pay for these assets, which is a real question and a legitimate one to ask. But it is a different question from whether these companies are building infrastructure that persists and compounds regardless of where equity prices go over the next eighteen months. Conflating the two is how people end up right about the drawdown and wrong about the decade. $NVDA didn’t stop being important when it pulled back in 2022, and the architecture it was supplying to the rest of the industry didn’t pause while the multiple compressed. The same logic applies now. Whether $AMZN‘s agent payments layer is priced correctly into the stock today is a separate matter from whether autonomous agent commerce becomes a real economic layer. One question is about timing and the other is about direction, and only one of them has an answer that stays true for more than a quarter.
The systems thinker’s real job is to watch for the variable that doesn’t appear in the headline metric, the place where profits drop even while output stays consistent because something structural moved that the obvious dashboard was never built to measure. Right now the obvious dashboards are measuring IPO size and valuation multiples, which is why every story this week reads as a story about prices. The structural shift is somewhere else entirely: in who controls the compute, who owns the payment rails, and who gets to set the settlement standards for transactions that no human ever initiates.
The Chip Layer Is Where This Gets Decided
It would be easy to look past the chip story next to a potential record-breaking float, and that would be a mistake. Anthropic is in talks with Microsoft over using the MAIA 200 in-house AI chip, a conversation that follows a $5 billion investment and points toward a much deeper infrastructure relationship than a normal customer arrangement would produce. Whoever controls the chip layer controls the cost structure of inference. The cost of inference controls the economics of every product built on top of it, which means it controls which products are viable at all, which means it quietly controls the shape of the industry three layers up.
Microsoft designing its own silicon and putting it under a leading lab isn’t only a cost play, then. It’s vertical integration of the kind that rearranges leverage across the entire stack, because the party that sets your marginal cost has a standing claim on your margin forever. The Dutch East India Company eventually needed its own army, not out of ambition but because the system it had built became too complex and too valuable to operate through external parties who had their own incentives. The leading AI labs are arriving at a version of that same inflection, and the direction of travel is toward owning compute rather than renting it. That is a durable structural story. It runs underneath the IPO timing question and stays true whether the float happens this year or next, whether the market is at a top or somewhere in the middle of a cycle nobody will name correctly until it’s over.
What the Crypto Rails Are Building For
MoonPay building institutional infrastructure and Circle extending USDC settlement look, on the surface, like crypto firms finally chasing mainstream legitimacy. That reading is correct and incomplete, which is the most dangerous kind of correct. The longer arc is that programmable money, meaning money that can be held, moved, and deployed by software with no human in the loop, is prerequisite infrastructure for exactly the agent economy Amazon is trying to stand up with AgentCore. The two announcements are not competing for the same narrative slot. One needs the other to function.
Human payment rails carry friction by design, and the friction is not a defect. Approvals, compliance checks, and human signatories exist because the system assumed a human actor who could be slow, could be asked, and could be held responsible. That assumption is load-bearing. When the actor is an agent executing a workflow at machine speed, every one of those checkpoints becomes a bottleneck, and the bottleneck doesn’t just slow things down. It caps what the agent can be trusted to do at all, which caps the size of the problem anyone bothers pointing an agent at. Crypto settlement layers, stablecoin infrastructure, and programmable wallets are not merely fintech products competing with card networks. They are the payment primitive the agent economy requires, and they are being built now, before the demand that justifies them is visible on anyone’s revenue line.
Instruments First, Then the Entities
The Dutch invented the instruments before anyone fully understood what was being built, and the company that used them came later and far larger than anyone anticipated. Nobody sat in Amsterdam in the early 1600s and said the joint-stock structure would eventually move half the world’s trade. The instruments were a solution to a financing problem, and the mega-corporation was what grew in the space the solution opened up. That ordering matters because it inverts how we usually narrate technology. We tell the story as though the great entity arrives and then builds its tools. It generally runs the other way.
So the payment rails, the chips, and the settlement layers are the part of this week worth writing down. The next generation of large-scale economic actors will not be companies in the traditional sense, and they need agent-capable payment infrastructure, compute at the chip level, and money that software can move before they can operate at any meaningful scale. Each of those three arrived in the same news cycle as the IPO headlines, and each got a fraction of the attention.
The IPO isn’t the event. It’s the moment a public market starts tracking something that has already been running for a while, at a price that reflects how late everyone is to noticing.
Leave a Reply