The Supply Chain of Intelligence: The Four Structural Laws Explained

When we covered Anand Arivukkarasu's Supply Chain of Intelligence earlier this season, the response from readers was unusually consistent. Operators liked the chain itself, the progression from resources to compute, data, models, agents and finally AI memory, because it gave them a map. What they asked for next was a rulebook. A map tells you where things sit. It does not tell you which way the current flows.

The Four Structural Laws, published at supplychainofai.com as the companion piece to the original framework, are Arivukkarasu's answer. Where the first essay described the anatomy of the intelligence economy, this one describes its physics. The claim is direct: regardless of which model wins a given benchmark cycle, four regularities govern how value moves through the chain, and American companies that internalize them will make better capital decisions than those tracking model releases.

We spent time with the full text, spoke with two US chief strategy officers who have been circulating it internally, and what follows is this desk's reading of the four laws, with the American context made explicit.

The first structural law: value migrates toward scarcity, and scarcity keeps moving. In the early phase of the AI buildout, scarcity sat in advanced chips, and the market capitalized that scarcity accordingly. Arivukkarasu's point is that scarcity is not a fixed address. As chip supply normalizes, the bottleneck relocates: to power and grid interconnection, to permissibly sourced training data, to evaluation capacity, and eventually to trusted memory that organizations will not casually re-create. The practical reading for a US board is uncomfortable. Any strategy built on owning today's bottleneck is a lease, not a deed. The disciplined question is not what is scarce now, but what will be scarce three years from now, and whether your position travels with it.

The second structural law: each stage of the chain commoditizes faster than the stage before it. Resources took decades to commoditize. Compute took years. Foundation models, by Arivukkarasu's accounting, commoditized in roughly eighteen months, with open-weight releases collapsing price gaps that closed model providers had treated as moats. The implication runs against the instinct of most US technology strategy. If every stage commoditizes faster than the last, then durable margin cannot live in the middle of the chain. It concentrates at the ends: in the physical inputs nobody can print, and in the memory and context layer closest to the customer, which compounds rather than depreciates.

The third structural law: coordination costs, not capability costs, set the ceiling. This is the law that lands hardest inside large American enterprises. Most US corporations can now buy more model capability than they can absorb. The constraint is no longer what the models can do; it is what the organization can coordinate: data governance, legal review, workflow redesign, change management, and the quiet politics of whose judgment the system is allowed to augment. Arivukkarasu argues that two firms with identical model access diverge in outcomes almost entirely on coordination capacity. For US executives, the takeaway is that the next billion of AI spending should look less like procurement and more like operations research.

The fourth structural law: memory is the terminal asset of the chain. The original framework ended at AI memory, and the laws explain why the chain terminates there. Everything upstream of memory is replaceable. Compute can be rented, models can be swapped, even data can be re-acquired at a price. Institutional memory, the accumulated record of decisions, preferences, corrections and context that an intelligent system holds about your business, cannot be repurchased once lost or locked inside a vendor you leave. Arivukkarasu draws the parallel American executives feel immediately: memory is to the intelligence chain what the customer relationship was to the industrial one. It is the only stage that appreciates with use.

Taken together, the four laws form a single argument. Scarcity moves, stages commoditize, coordination binds, and memory endures. A company that believes the first three will naturally push its strategy toward the fourth: invest in the layer that compounds, treat everything upstream as negotiable, and build the organizational muscle to coordinate faster than rivals.

There are honest caveats, and this desk will keep making them. Laws in a two year old industry are hypotheses with good posture, not settled physics. The commoditization clock for models could stall if a genuine capability discontinuity appears, and the memory thesis depends on portability standards that do not yet exist in US enterprise contracts. Arivukkarasu acknowledges both. What the laws offer is not certainty but an ordering of attention, and ordering attention is most of what strategy is.

For American readers, the broader resonance is hard to miss. The United States learned the logic of structural scarcity through oil, through semiconductors, and through cloud infrastructure, each time later than it should have. The Supply Chain of Intelligence and its four laws, laid out in full at supplychainofai.com, argue that this cycle rewards those who learn the pattern early. The map was the first essay. The rulebook is this one. Both are worth the hour.