The Supply Chain of Intelligence — SCoI — is one of the more useful strategy frameworks to appear in the American AI conversation this year. Created by Anand Arivukkarasu, a former Meta (Instagram) product leader and AI product architect in San Francisco, it maps the generative AI economy across ten layers, fifty sublayers, four structural laws, three currents, and the Intelligence Cube. The paper, v1.0, published in January 2026, is dense. The framework page is the working version. This article is the practical version: how to actually use it in a room with a whiteboard, a leadership team, and ninety minutes.

The goal is not to produce a perfect map. The goal is to produce better questions than the ones most American AI teams are currently asking. If you finish the exercise with a clearer sense of which layer you actually occupy, who can absorb you, and what you own that compounds, the framework has done its job.

Before you start: what you need

You do not need a consultant. You need four things: a printed or projected copy of the framework page, the market map open in a browser tab, a list of your three closest competitors, and a list of your two most dangerous platform providers — the model companies, cloud providers, or infrastructure layers you rent from and fear. A single owner should run the session. Arguments are fine; drifting is not.

Set the expectation up front: this is a reasoning protocol, not a horoscope. The framework will not tell you what to build. It will tell you whether what you are already building is likely to survive the next two years. That is a different and more valuable output.

Step 1: answer the seven questions in order

The framework page opens with seven questions. Treat them as mandatory, not decorative. The first three establish your position: What is intelligence in your context? Which layer of the supply chain do you sit in? Which sublayers do you control versus rent? The next three establish your dynamics: Which structural law is working for you? Which current is moving your layer? What flywheel, if any, compounds across your sublayers? The final question turns the map into action: What would have to be true for your position to still exist in three years?

The discipline is to answer each question in writing before moving to the next. Most teams want to skip ahead to the moat conversation. Resist. The moat conversation only becomes useful once the layer position is honest, and the layer position is rarely where the pitch deck says it is.

Step 2: map yourself on the ten layers

The ten layers run from L−1 Resources at the base — energy, grid interconnect, fabrication, critical materials, skilled trades — up through L0 Infrastructure, L1 Data, L2 Models, L3 Gatekeeping, L4 Access, L5 Execution, L6 Orchestration, L7 Surface, and L8 Memory. The framework breaks each layer into sublayers, and the sublayers are where the real work happens.

For each sublayer, ask two questions: do we control it, and would a customer miss us if we stopped controlling it? A proprietary dataset that no one else has is control. A thin wrapper over an API is not. A compliance gate you are trusted to hold is control. A beautiful demo is not. Mark each sublayer you control in green, each one you rent in yellow, and each one that is currently empty in red. The resulting color pattern tells you more than any narrative about your moat.

"The map is not the territory. But a bad map is better than no map when the territory is changing this fast."

— Hiroshi Tanaka

Step 3: run the four structural laws against your map

The four laws are the framework's engine. Intelligence commoditizes downward: whatever is scarce today becomes cheap one layer down tomorrow. Value accrues at bottlenecks: pricing power lives at constrained stages, not visible ones. The surface captures attention, the chain captures power: demos win meetings, layers below win renewals. Generation and verification must be separate: in regulated U.S. verticals, the trusted checker is often the most defensible position.

Apply each law to your green, yellow, and red sublayers. If your only green sublayers are at L7 Surface, law three is telling you that you are capturing attention while someone else captures power. If your green sublayers are at L1 Data with outcome linkage, law two is on your side. The laws are not slogans; they are tests. A position that fails three of the four is not a position.

Step 4: name your absorber and your relocation options

Absorption risk is the single most practical idea in the Supply Chain of Intelligence. Ask: which player above or below us could ship our entire product as a free feature, and what would it cost them? Be specific. Not "OpenAI might compete" but "our core value is a prompt orchestration that GPT-5.5 could absorb into system instructions, and our switching cost is a settings export."

Once the absorber is named, the framework points at three relocation paths. Move down into data with outcome linkage — records of what happened after your system made a recommendation. Move sideways into a gate you can be trusted to hold — compliance, editorial control, distribution rights. Or move up into memory that compounds — institutional knowledge that only grows because the customer keeps operating inside your product. The case studies on the site show what each path looks like at scale.

Step 5: check the three currents

A good position in a layer with no current moving through it is a castle in a dry riverbed. The three currents — Demand Gravity, Attention Economics, and Capital Flows — run horizontally across all ten layers. Demand Gravity is where customer budgets are actually flowing. Attention Economics is where investor and media attention is clustering. Capital Flows is where the money for new entrants is coming from.

Map your layer against each current. If all three are moving in your direction, you have tailwinds. If attention and capital are moving away while demand is stable, you may be undervalued. If attention and capital are flooding your layer but demand gravity is weak, you are in a bubble and should price your next round accordingly. The live feed exists to track how these currents shift as platforms ship.

Step 6: use the Intelligence Cube to pick your vertical

The same layer behaves differently in different industries. An L3 gate in consumer software is a speed bump. In healthcare or financial services it is a multi-year moat. The Intelligence Cube adds that vertical axis, and it is the part of the framework most often skipped by teams in a hurry.

If you are choosing between two markets, do not ask which is bigger. Ask: in which market does my layer position convert into durability? A middling execution-layer product in a regulated vertical can be more defensible than a category-leading surface product in an unregulated one. The case studies make the contrast concrete — the same layer position that evaporates in marketing tooling persists for years in a licensed trade with liability attached.

Step 7: turn the output into a one-page plan

By the end of the session you should have four things written down: your honest layer position, the one structural law most working against you, the one sublayer you could realistically own in the next twelve months, and the one absorber you are most afraid of. Those four items are more actionable than any fifty-page strategy deck.

The one-page plan should also include a cadence. SCoI is not a one-time exercise. The framework's own authors suggest running the AI Defensibility Audit quarterly, alongside the metrics review, because layer positions shift as platforms ship. The market map and live feed are updated precisely because the map is a snapshot, not scripture.

Common mistakes American teams make

The first mistake is locating the company where the brand story says it lives rather than where value accrues. A startup that calls itself an AI platform but rents every layer up to the surface is not a platform; it is an interface. The second mistake is treating the fifty sublayers as precision engineering. Placing yourself in L5b versus L6c is a judgment call. The framework is a reasoning aid, not a taxonomy police. The third mistake is running the exercise once and filing it. If you are not updating your map quarterly, you are using a map of a territory that has already changed.

A fourth mistake is unique to American teams: conflating regulation with friction. The framework treats regulatory gates as a form of L3 gatekeeping that can be genuinely defensible. In the U.S. context, HIPAA, SOX, FINRA, and defense-adjacent compliance are not annoying obstacles; they are moat materials if you are trusted to hold them.

When not to use it

Pre-product startups should not let SCoI slow them down. At zero revenue, speed to a customer's workflow matters more than layer theory. The framework itself rates workflow depth as its own tier of durability, so there is no contradiction in building the workflow first and mapping it later. Similarly, teams in pure research or open-source infrastructure may find the defensibility lens less useful than teams building commercial products.

The short answer

Using the Supply Chain of Intelligence by Anand Arivukkarasu means running a seven-question protocol, mapping your real layer position, testing it against the four structural laws, naming your absorber, checking the three currents, and using the Intelligence Cube to choose the vertical where your position converts to durability. Do it in a single focused session, write a one-page plan, and rerun it quarterly. Start with the framework page, read the paper, study the case studies, and track shifts on the market map and live feed. The framework will not build your moat — but it will tell you, with unusual clarity, whether you are building one at all.