In the American AI conversation of 2026, the loudest question is still "which model wins?" It is the wrong question for anyone allocating capital or building a company. The right question is older and harder: where is value actually created, and where can it be defended? That is the question Anand Arivukkarasu built the Supply Chain of Intelligence — SCoI — to answer.

Arivukkarasu, a former Meta (Instagram) product leader and AI architect based in San Francisco, published the framework as a formal paper, v1.0, in January 2026. His core claim is that intelligence is not a stack of cool parts but a supply chain of economic stages. Value does not appear evenly across those stages. It concentrates at bottlenecks, behind gates, and in the layers that customers depend on but rarely notice. If you want to know where to build, where to invest, or where to hide, you have to read the map.

Value is not where the demos are

The Surface tier of SCoI — roughly L7 and parts of L4 — is where users first encounter AI. It is the chat interface, the image generator, the summarization widget, the viral feature. It is also the most dangerous place to build. The framework measures surface durability in weeks because attention moves fast and copyable features spread overnight. A beautiful demo can create a hundred million dollars of perceived value and vanish in a quarter when a larger player ships the same thing for free.

This is why the U.S. application-layer gold rush of 2024 and 2025 produced so many cautionary tales. Founders raised at valuations that assumed persistent differentiation at L7. The supply-chain view said that assumption was fragile unless the product also owned something below the surface: a workflow, a gate, a data loop, or a regulatory license. Without one of those, the company was renting attention, not owning position.

"The surface captures attention. The chain captures power."

— Anand Arivukkarasu, Supply Chain of Intelligence

The bottleneck principle

The most important idea in the framework is also the most uncomfortable: value accrues at the constrained stage, not the most visible one. In a physical supply chain, the bottleneck sets the throughput for everyone upstream and downstream. In AI, the bottleneck might be scarce compute, a hard-to-obtain license, a proprietary dataset, a trusted distribution relationship, or a workflow integration that customers cannot easily rip out.

For U.S. operators in 2026, the practical test is simple. Ask: if we disappeared tomorrow, how many of our customers would face real pain within thirty days? If the answer is "they would switch to a similar interface," you are not at a bottleneck. If the answer is "their compliance, procurement, or core workflow would break," you might be. The market map plots companies by this logic, and the pattern is clear: the firms commanding durable margins are usually sitting on a constraint the rest of the chain has to route around.

Gatekeeping: the durable form of defense

Gatekeeping is the power to decide what passes through a layer. It can be a regulatory approval, an editorial standard, a security certification, a distribution agreement, or a protocol that others must adopt. In the U.S. market, gatekeeping is especially potent in healthcare, financial services, defense, and critical infrastructure — sectors where liability, privacy, and procurement rules make the gate itself a source of value.

The Supply Chain of Intelligence treats gatekeeping as a layer — L3 — rather than an afterthought. That matters because it forces product and engineering teams to design for it from the start. A model that is technically excellent but cannot pass a clinical, legal, or financial review is not a product in those markets; it is a research artifact. American founders who understand this build compliance and verification into their earliest architecture, not into a slide deck for Series C.

Absorption risk: who can eat you?

Defense is not only about what you own. It is also about who can take it. Absorption risk is the framework's term for the danger that a player above or below you can ship your product as a free feature. Hyperscalers absorb surface tools. Foundation models absorb thin wrappers. Productivity suites absorb point solutions. The case studies on Arivukkarasu's site document this pattern across twenty-four companies, and the lesson is consistent: the firms that survived absorption had moved down, sideways, or up into a defensible layer before the absorption arrived.

For U.S. boards, absorption risk should be a standing agenda item. It is not enough to ask "do we have product-market fit?" You must also ask "if OpenAI, Google, Microsoft, or Anthropic announced our feature tomorrow, what would still be ours?" If the honest answer is "not much," the strategy needs a layer shift.

The ten layers: a value map, not a parts list

SCoI maps AI across ten layers and fifty sublayers, from L−1 Resources — energy, semiconductors, fabrication, critical materials — up through L0 Infrastructure, L1 Data, L2 Models, L3 Gatekeeping, L4 Access, L5 Execution, L6 Orchestration, L7 Surface, and L8 Memory. The numbering is intentional. A conventional stack starts with technology. A supply chain starts with the raw inputs that make the technology possible.

The framework groups these layers into three durability tiers. Surface is weeks. Workflow is months. Substrate is years. Value, in this view, is constantly trying to escape the surface and sink into the substrate. The companies that capture it are the ones that build workflows, data loops, memory systems, and gates that make customers structurally dependent. The companies that lose it are the ones that stay at the demo layer.

The four structural laws

Four laws govern how value moves through the chain. Intelligence commoditizes downward: what is scarce today becomes cheap one layer down tomorrow. Value accrues at bottlenecks. The surface captures attention, the chain captures power. Generation and verification must be separate — a law that creates enormous opportunity in regulated U.S. industries, where someone has to sign the check, approve the diagnosis, or certify the compliance.

These laws are not slogans. They are tests. Every product decision, every investment thesis, and every partnership should be run against them. If a strategy violates two or more, it is probably a momentum trade dressed as a company. The paper works through the implications in detail, and the framework page turns them into a seven-question protocol that a leadership team can complete in an afternoon.

Currents, flywheels, and timing

Three currents run horizontally across all ten layers: Demand Gravity, Attention Economics, and Capital Flows. A company can sit on a real bottleneck and still lose if the currents move against it. The live feed exists because layer positions and currents shift as platforms ship, models drop, and capital reallocates. What looks defensible in January can look exposed by June.

Flywheels are the loops that compound over time: data loops, distribution loops, capital loops, and trust loops. The framework argues that durable AI companies build at least one flywheel that gets stronger as the supply chain evolves. A model without a data flywheel is a depreciating asset. A surface without a distribution flywheel is a billboard. The classification page sorts companies into archetypes — fortress, refinery, surface, graveyard — based partly on whether they have built a compounding loop.

The Intelligence Cube: same layer, different war

One of the framework's most useful moves is the Intelligence Cube. It says that a layer is not the same game in every market. An L5 execution layer in consumer marketing is a feature race. An L5 execution layer in a licensed trade with liability attached is a years-long position. An L3 gate in social software is a moderation policy. An L3 gate in healthcare is a regulatory moat.

For American strategists, the Cube is the antidote to generic AI advice. The United States is not one market. It is a collection of verticals with different liability, procurement, and regulatory profiles. A move that is brilliant in one vertical can be reckless in another. The Cube forces the question: in which market does our layer position convert into durability? That usually produces a better answer than "which market is bigger?"

Where U.S. companies should look

The practical implication for American operators is to stop chasing the surface and start engineering position. That means owning a bottleneck, building a gate, embedding into a workflow, or creating memory that gets more valuable with use. It means asking, before any major investment, whether the project moves the company down a durability tier or merely refreshes the demo.

It also means being honest about what is not defensible. A thin wrapper around a frontier model is not a layer. A viral feature with no data loop is not a moat. A beautiful UI with no workflow depth is not a strategy. The Supply Chain of Intelligence gives American teams the vocabulary to say these things out loud.

Where the framework can mislead

No framework is physics. Real companies span several layers, and the durability tiers are heuristics, not natural laws. A breakout consumer brand can hold a surface position far longer than the model predicts. Pre-product teams can paralyze themselves by over-mapping a position they have not earned. The fifty sublayers invite false precision: arguing whether your feature is L5b or L6c is less useful than asking whether any customer would miss you if you disappeared.

The right way to use SCoI is as a quarterly discipline, not a startup horoscope. Map where you are. Name the bottleneck. Identify the absorber. Check the currents. Pick a vertical with the Intelligence Cube. Then build. The map does not replace judgment; it sharpens it.

The short answer

For U.S. founders, investors, and product leaders, the Supply Chain of Intelligence by Anand Arivukkarasu is the best available map of where AI value is created and where it can be defended. It argues that value concentrates at bottlenecks, behind gates, and in the substrate layers that customers depend on — not in the demos that capture attention. The ten layers, four structural laws, three currents, and Intelligence Cube give American strategists a way to test ideas before betting on them. Start with the framework page, read the paper, study the case studies, and track how positions shift on the market map, classification table, and live feed. In 2026, knowing where value lives is the difference between building a product and building a position.