India barred a JPMorgan unit this week over alleged stock-market manipulation, and the story produced exactly the cycle you would expect: regulatory overreach, fairness in price discovery, the integrity of cross-border capital. All of those concerns are real and none of them is the point. India barring a major bank unit from its markets is the public face of something, and the public face is almost never where the leverage sits.
That is the claim worth carrying through everything else here: every system has a public face and a private structure. The public face is what gets traded on the evening news, argued about, and forgotten within a week. The private structure is where the actual leverage lives, and it changes far more slowly than the coverage implies. Once you start reading for structure instead of for headline, the same pattern turns up in places that look nothing like finance, including the part of the technology industry currently absorbing the most attention and the most money.
What the Heavy-Truck Market Shows About Structure
Richard Rumelt once dissected the United States heavy-truck business precisely to show what good strategy actually looks like when you strip away the language people use to describe it. The market is mature. Growth is low. There is nothing glamorous happening. Yet the top four players, Daimler, Paccar, Volvo, and Navistar, each hold a tightly coordinated position rather than a randomly distributed share. Daimler leads at 38 percent, and it got there by buying Ford’s troubled truck division in 1977 and integrating it deliberately over years. Paccar holds 25 percent through an entirely different design aimed at different customers. From a distance the numbers look like an ordinary competitive market resolving itself. The structure underneath tells a different story, one in which each position was built and defended on purpose.
Ludwig von Mises made a related observation about ownership that generalizes further than most people expect. Even when a government or a municipality owns part of the means of production, he argued, the defining characteristics of a market economy do not disappear. The surface changes, sometimes dramatically. The incentives, the pricing logic, and the coordination problem all persist underneath it, because those are properties of the structure rather than of the nameplate. Public ownership is a headline. Market structure is the operating system, and operating systems are not changed by press release.
The Model Is Becoming a Commodity
Which brings all of this uncomfortably close to the technology news most people scrolled past today. OpenAI paused training to rewrite its Preparedness Framework. Anthropic’s run rate reportedly hit $65 billion. Cursor shipped something that looks a great deal like a GitHub competitor. Stripe is said to be paying more than $7 billion for OpenRouter, the layer that routes an application to whichever model happens to be cheapest that minute. And the price of GPT-5.6 Luna dropped 80 percent overnight, to $0.20 per million input tokens.
The narrative wrapped around all of that is about models: who has the best one, who is leading the race, who paused for safety and what the pause signals. That is the headline layer, and it is genuinely interesting. Beneath it a different logic is forming, and the 80 percent overnight price cut is the clearest evidence of it. The model is becoming a commodity. What stays scarce is the power contract, the cooling loop, the shelf life of the silicon, and the distribution layer that captures the workflow rather than the keystroke.
Cursor appears to understand this. Its review product is not valuable because it replaces a tired human reviewer; it is valuable because it owns the decision point, and whoever owns the decision point owns everything that flows through it. Stripe understands it as well. A payments company buying a model router is not diversifying into a hot sector. It is securing the infrastructure that decides which model gets called, which is a position that keeps its value no matter which lab is ahead this quarter.
Clearing Rails, Routing Layers, and Who Owns the Decision
The same distinction runs straight back through the JPMorgan story. Alleged manipulation is a regulatory headline with a beginning, a middle, and an end. The structural issue is that cross-border capital flows still move through legacy clearing rails that were never designed for the speed or the volume they now carry, and the incentive to game those rails grows with every new participant added to them. You can bar one unit from one market. You cannot bar the architecture that made the bar necessary, and almost no one is discussing the architecture.
Von Mises would not be surprised by any of this, and Rumelt would point out that the real strategy sits in the coordination rather than in the market-share number that gets reported. The heavy-truck industry teaches the lesson plainly because it is boring enough to see clearly: four players, a mature market, low growth, and each one maintaining a distinct and deliberately designed position. The winners there are not winning by chance, and they are not winning by having the best engine. They are winning by the structure they built and then defended for decades.
The technology industry is moving through the same transition right now, whether or not it has the vocabulary for it. The race for better models is starting to resemble the race for bigger truck engines in a saturated market, where the engineering is real and the differentiation is gone. Differentiation shifts instead to who owns the financing layer, the routing layer, and the cooling layer. Ripple bought Hidden Road for $1.25 billion, which is not a speculative bet but ownership of the financing layer sitting between institutions and assets. Google is buying up to $12.2 billion in Marvell shares tied to TPU products through 2033, which is not an investment so much as a message to Broadcom that the silicon underneath all of this will not be single-sourced. Put those two beside each other and the shape is identical, despite one being a crypto-adjacent firm and the other a hyperscaler.
Read the Schematic, Not the Screen
None of this is an argument for ignoring regulation or for cheering infrastructure spending indiscriminately. It is a narrower reminder that the thing everyone is talking about is rarely the thing that matters, and that the two are frequently in the same story. The model is the headline; the moat is underneath it. The enforcement action is the noise; the clearing architecture is the signal. That split is visible in every domain once you stop watching the screen and start reading the schematic, and the schematic is usually public, just unappealing.
So the hard questions are almost never the ones being asked out loud. The hard question is not whether one bank’s unit broke a rule. It is whether the rules were written for the market that exists or for the market that existed twenty years ago, because the answer determines how many more of these stories are coming. The hard question in AI is not which model wins the benchmark this month. It is who owns the infrastructure that decides which model gets used at all. The hard question in strategy is not what a competitor shipped last week. It is what coordination they built that you cannot see from outside, and that they have had years to defend.
Headlines are designed to be consumed in seconds, which is why there are so many of them. Structure takes years to build and decades to break, which is why so few people bother to look at it and why the looking keeps paying. The people who understand this do not need to manipulate markets. They design them.
Leave a Reply