This paper is written for the leaders who decide where a firm spends its best energy. The CEO setting strategy, the board pricing the bet, the head of a practice choosing between shipping faster and building something only this firm can build. It answers one question. When the capability everyone is racing to adopt can be bought by the hour, where does lasting advantage actually come from.
The Depth Ladder, the Wrapper Ceiling, the Commoditization Floor, the Depth Moat, and the Moat Depth Index are original constructs of this work, published under the license on the back cover. They rest on decades of strategy research into why some advantages last and most do not [4][5], applied here to the one shift that changed the question. Every quantitative claim is cited to a named source or labeled directional, without exception.
For two years the winning move in enterprise AI was speed. Wrap a model in a clean interface, point it at a workflow, ship before anyone else. It worked, and it worked because the capability underneath was scarce and expensive. That scarcity is gone. The same capability that cost a fortune to reach in 2022 now costs cents, and the gap between the best open model and the best closed one has nearly closed.
When the scarce thing becomes abundant, the advantage moves. It always does. It moves to whatever is still hard to copy, and in this era the hard thing is depth. Proprietary data a rival cannot buy. A mechanism you built and only you understand. Learning that compounds every quarter you run it. None of this is downloadable, and that is exactly why it lasts.
This paper is my answer to the question I now hear in every strategy room. Our competitors have the same models we do, so what is ours. The framework in these pages is a ladder out of that question, five rungs from reselling access to originating research, with the economics of each rung made explicit. The constructs are original work. The numbers are cited or labeled directional, without exception. I hold this to the standard of the rooms it will be read in.
The capability that enterprise AI strategy was built around has been commoditized in public. What everyone can buy by the hour cannot be the thing that sets a firm apart. The advantage has moved to depth, and depth has a structure. Five findings follow.
Access to capability is no longer scarce. The price to reach a fixed capability level fell more than 280 times in roughly eighteen months [1], and the gap between the best open and best closed model narrowed to under two points [3]. A resource that cheap and that widely held cannot be a moat.
A thin layer has a low ceiling and a falling floor. A pure integration layer owns no mechanism, so the price it can charge trends toward the cost of rebuilding it. We name the low upper bound the Wrapper Ceiling and the erosion that pushes down on it the Commoditization Floor. Together they squeeze the margin of anything easily copied.
Durable advantage is accumulated, not purchased. Strategy research is settled on this point. Advantage that lasts comes from asset stocks a rival cannot buy in a market and cannot rebuild quickly [4][5]. In AI that means proprietary data, an owned mechanism, and learning that compounds. We call the result the Depth Moat.
Depth compounds, rent does not. A firm that rents capability tracks the vendor frontier and plateaus there. A firm that builds on its own data pulls away a little more every quarter, because the gains feed the next gain. Exhibit 6 shows the two paths separating over time.
The move is a ladder, not a leap. The Depth Ladder gives five rungs from reselling access to originating research. A firm does not jump to the top. It climbs, and the first owned mechanism is reachable in a single quarter with a small team.
For two years, having good AI capability was itself a source of advantage, because reaching it was hard and expensive. That is no longer true, and the change is measured, not felt. The price to reach a fixed capability level fell from about twenty dollars per million tokens to about seven cents between late 2022 and late 2024, a reduction of more than 280 times in roughly eighteen months [1]. Adoption followed the price. By 2024, 78 percent of organizations reported using AI, up from 55 percent a year earlier [2].
Capability also stopped being a place to hide. The gap between the best open model and the best closed one narrowed to under two points on a public leaderboard, down from eight [3]. When most firms adapt from the same small set of base models, the base is shared by design, and the differences that matter must come from somewhere else [6].
When the scarce input becomes abundant, advantage moves to whatever is still hard to copy. In this era that is depth, and depth is the subject of this paper.
The structural problem is in Exhibit 2. Firms pour their best effort into the layers that are easiest to copy, while the defensible value sits in the layers almost no one is building.
Depth is not a single choice. It is a ladder. The Depth Ladder places any AI effort on one of five rungs by how hard it is for a rival to copy. Value captured and defensibility both rise with the rung. Most firms sit on rungs two and three and wonder why the advantage will not hold.
Wrapper Ceiling. The low upper bound on the value and margin a thin layer can ever capture, because it owns no mechanism of its own. Anyone can rebuild the layer, so the price it can charge is held down near the cost of rebuilding it.
Commoditization Floor. The downward pressure that pushes the value of anything easily copied toward the cost of copying it. It is formally argued in Section 04 and is the force closing on the Wrapper Ceiling from below.
A firm can stand on more than one rung at once, and most should. The point of the ladder is not to abandon the lower rungs, which keep the lights on and the clients served. The point is to know that nothing below rung four will hold, and to make sure the firm is climbing, not just standing.
A Depth Moat is made of three parts that reinforce each other. Data only you hold. A mechanism only you run, meaning a model, a pipeline, or a method you built and understand. And learning that compounds, because every month of running it makes the next month better. The parts form a loop, and the loop is the moat.
Depth Moat. The defensible position a firm holds when its advantage rests on proprietary data, an owned mechanism, and learning that compounds. Unlike a thin layer, it cannot be bought in a market or rebuilt on a weekend, which is exactly why it lasts.
| Rung | What you own | Time to replicate | Defensibility |
|---|---|---|---|
| 1. Resell access | Nothing of your own | Instant | None |
| 2. Configure | Settings and prompts | Days | Very low |
| 3. Integrate | Plumbing into your systems | Months | Low |
| 4. Own a mechanism | A model, pipeline, or method | Quarters to years | High |
| 5. Originate research | New method and data others cite | Years, if ever | Very high |
The economics are not a matter of taste. Strategy research settled the underlying result decades ago. An advantage lasts only when the resource behind it cannot be bought in a market or accumulated quickly by a rival [4][5]. Anything a competitor can reproduce on a short timeline stops being an advantage the moment it does.
Apply that to a thin AI layer and a clean result falls out. We state it as the Commoditization Floor. In a market where a capability is broadly accessible, the price a firm that does not own it can charge for providing it trends toward the cost of reproducing it. The capability a wrapper resells now costs cents [1], so the floor it is pressed toward is very low.
Cost of standing still. A firm that only rents is identical, to a buyer, to every other firm renting the same models. When nothing separates two options but price, price is the only lever left. That is the Commoditization Floor arriving at the door.
Cost of depth. The first owned mechanism is a single-quarter build for a small team, detailed in Section 05. It is a known, one-time cost set against an open-ended return.
Payback logic. Depth pays back the way an asset does, not the way a feature does. A feature is copied and gone. An asset compounds, and the gap it opens is the return.
You climb the ladder one rung at a time, and each gate is a measurable exit, not a status meeting. A firm that cannot pass Gate 1 on a single bet has no business starting five.
Do you hold data a rival cannot buy or quickly gather? No owned data, no moat. This is the first gate, and it is binary.
Is this task central to how the firm wins, or a convenience at the edge? Depth is for the center. The edge should be rented.
How long would a capable rival need to copy it? If days, it is not a moat. If quarters or years, it is worth building.
Does running it make the next run better? Compounding is what separates an asset you own from a feature you shipped.
The closing argument. The capability is commoditized, and that is the opportunity, not the threat. When every firm holds the same models, the one that builds something only it can build is the only one with a reason to be chosen. The ladder is in these pages. The first rung above the ceiling is one quarter away.
Answer honestly for your most important AI effort. Count the YES answers to find your Moat Depth Index and the rung it places you on. Designed to be completed in pen.
| No. | Probes | Question | Yes | No |
|---|---|---|---|---|
| 1 | DATA | Do you hold data on this problem a competitor cannot buy or easily gather? | ||
| 2 | DATA | Is that data growing as a by-product of running the product? | ||
| 3 | MECH | Do you run a model, pipeline, or method you built, not only a vendor interface? | ||
| 4 | MECH | Could you explain, in house, exactly why your mechanism works? | ||
| 5 | LOOP | Does each month of operation make the next month measurably better? | ||
| 6 | CORE | Is your deepest AI bet aimed at where the firm actually wins, not a side task? | ||
| 7 | COPY | Would a capable rival need quarters, not days, to copy your core AI work? | ||
| 8 | EVAL | Do you measure your mechanism against a rented baseline on a named metric? | ||
| 9 | IP | Have you produced a method or result others could cite or would want to? | ||
| 10 | FUND | Do this quarter's gains fund next quarter's climb up the ladder? |
This paper is a conceptual contribution. The Depth Ladder, the Wrapper Ceiling, the Commoditization Floor, the Depth Moat, and the Moat Depth Index are original to this work. They apply a settled result from strategy research, that durable advantage rests on resources a rival cannot buy or quickly accumulate [4][5], to the specific moment when AI capability became cheap and widely held [1][6].
Quantitative claims follow one discipline throughout, made visible in every exhibit. Solid fills carry cited data. Hatched fills and curves labeled directional exist to make the structure of an argument legible, not to report measurements. The Commoditization Floor is labeled a formally argued result, not an observed series. No statistic in this paper is attributed to a source that cannot be independently verified.
Arjun Jaggi writes and advises at the intersection of enterprise AI strategy, research, and the economics of advantage. His published body of work, spanning original frameworks, concept papers, and books, is read by executive teams navigating AI adoption and is available in full, without a paywall, at arjunjaggi.com.