Chapter 083　What Did Amazon Redefine?

Amazon began as a mail-order bookseller. Most of the operating income it earns today does not come from retail. It comes from a computing platform. A capability built to run the company’s own retail operation was opened to outsiders, and it became an industry. Among the cases of Enterprise Redefinition, that single fact is what makes this company matter. The question of this chapter is why an internal capability could become an external industry.

1 The question — what makes this company worth

asking about Start from the disclosed figures. Everything below comes from the company’s own disclosures, as confirmable as of August 2026. Consolidated revenue for the year ended December 2025 was $716.9 billion, and operating income was $79.9 billion. By segment, revenue was $426.3 billion in North America, $161.8 billion in International, and $128.7 billion at AWS. Operating income was $29.6 billion in North America, $4.7 billion in International, and $45.6 billion at AWS. The structure is visible in those numbers. AWS accounts for less than a fifth of revenue. It accounts for more than half of operating income. Retail carries the scale, and the computing platform carries the profit. Two businesses of entirely different character live inside one enterprise. Look further into the same period. Online stores produced $282.8 billion. Third-party seller services produced $171.1 billion. Advertising services produced $68.6 billion. Subscription services produced $49.6 billion. Physical stores produced $22.2 billion. Taken together, the businesses that support other people’s commerce are larger than the business of buying and selling goods on the company’s own account. Amazon therefore runs on at least three different logics. First, retail, in which the company buys and sells. Second, a marketplace, in which it rents space and logistics to others. Third, a set of businesses — computing capacity and advertising — in which internal capability was opened to the outside. What deserves a question is the relationship between the three layers. Many companies build internal systems to run their core business. Those systems are booked as cost, and they never leave the building. At Amazon, one of them became a product. Why could it become a product? We cannot assert what was decided inside the company. Public information yields only the outward shape, the timing, and the result. Even the outward shape, however, carries a decisive lesson for management. This chapter confines itself to that.

2 Conventional answers and their limits

Three explanations of Amazon circulate widely. Each is partly right. None is sufficient. The first answer: “scale and an obsession with logistics won it” Low prices, selection, and fast delivery. Keep that wheel turning and you win, the argument runs. The company has in fact invested heavily in its logistics network over a long period. Convenience for the customer surpassed that of competitors. So far the account fits the facts. The account cannot explain AWS. Why would excellence in warehouses and delivery lead to selling servers to third parties? There is no continuity between the two as businesses. The scale explanation closes inside retail and stays there. The second answer: “the founder’s long-termism won it” The 1997 letter to shareholders was titled “It’s All About the Long Term.” It states explicitly that long-term cash flow takes priority over the appearance of short-term profit. The document exists, and the company’s subsequent conduct is consistent with it. Long-termism, however, contains no direction. Deciding to wait a long time and deciding what to wait for are different decisions. Companies that proclaimed long-termism and produced nothing are beyond counting. Long-termism is a necessary condition, not a sufficient one. The third answer: “it found the cloud market first” This is the first-mover explanation. The company entered in 2006 and built scale before rivals arrived. As chronology, it is not wrong. As causation, it reverses the order. AWS did not find an existing market and enter it. At the time, the market did not exist. By AWS’s own account of its origins, the starting point was an internal problem. Procuring and building infrastructure was difficult and expensive, and it consumed the time of the company’s best engineers. The idea was to carve that inconvenience out as a component, in a form anyone could use. The company did not find a market. It made one. The phrase “first-mover advantage” presupposes that the market exists first. Here that premise does not hold. The three conventional answers share one weakness. Each describes how well the company played inside a given market. The point about this company sits one level above that. It carried its own internal capability out into a place that was nobody’s market yet.

3 What was redefined — read across the five dimensions

We read the case along the five dimensions of Enterprise Redefinition, in the fixed order: Purpose, Business, Organization, Capital, and Leadership.

3.1 Purpose — the core held, and its reach widened

What the company has long stated is that it intends to be the most customer-centric company on Earth. As far as public information shows, that core has not substantially changed. What changed is the range of who counts as a customer. At the start, the customer was a consumer buying books. Then sellers were added. Then developers and corporate informationsystems departments. Then advertisers. The words stayed the same while the object widened four times. The five dimensions do not change at the same frequency. Core Purpose can hold steady while its expression and its realization evolve. Amazon looks like the textbook instance. The core was not rewritten. Its reach was extended.

3.2 Business — what the company sells was rewritten four times

This is the dimension that moved most. Sell goods. Rent the venue. Sell computing capacity. Sell attention. The four businesses differ in how revenue is made, in the nature of their costs, and in whom they compete against. The third is decisive. The computing platform built behind the retail operation was rented out by the hour to outside companies. With that step the company entered territory where its customers and its competitors overlap. A retail rival can also be an AWS customer. In the terms of Vol. VI, Ch. 059, this is a combination of the Layer Shift Pattern and the Capability Redeployment Pattern. The Layer Shift Pattern moves a firm upstream or downstream in the value chain. The Capability Redeployment Pattern takes a capability that survives a change of market and puts it to work in another market. In Amazon’s case the movement was upstream, not downstream. The company moved from the layer at which goods are sold to the layer of computation on which every other business rests. And the capability it redeployed had been defined in functional terms rather than by product name. Not the power to sell books. The power to build, and to keep operating, an infrastructure that handles enormous transaction volume cheaply and reliably.

3.3 Organization — internal boundaries set the shape of the

external product Redeploying a capability has a precondition. The capability has to exist in a detachable form. The company opened its product-information systems to outside developers early; that work began in 2002. Internal functions were designed as components with defined entry points. That habit became the foundation on which the platform itself could later be turned into a product. What can be read from public information is that the shape of the company’s internal interfaces became the shape of its external products. One widely known description of the company’s organization is the “two-pizza” team: keep a team small enough to be fed by two pizzas. The founder has discussed the principle publicly, and it is confirmable in public sources. The point is not the headcount. The point is that if you cut an organization into many small teams, you need a boundary for each cut. Small teams cannot coordinate by sitting in the same room and knowing each other. They have to get their work done through defined entry points, without knowing the other side’s internal circumstances. Dividing an organization into small units is the same act as forcing interfaces into existence inside it. Here is the structural link to selling capability outside. When you sell something to an external company, the buyer does not know your internals. Unless the thing is usable without that knowledge, it is not a product. Another team inside the same building is the first example of a user who does not know your internals. Put differently, the company did not design for external sale. It designed to keep teams small, and what came out of that had a shape that could be sold. The order is what we should notice. This is where the most important lesson for an incumbent sits. When internal systems are fused to the departments that use them, no capability can be extracted. A capability that cannot be extracted cannot be sold, however excellent it is. Organizational design quietly sets the range of business opportunity available to a company. When units are large, internal coordination can be handled by meetings and personal relationships. As long as that suffices, the interface never gets written down. A capability that is never written down cannot be handed to anyone outside.

3.4 Capital — allocation kept going to capability rather than to

profit The Capital dimension moved substantially as well. For a long period the company has preferred reinvestment in the business over the maximization of accounting profit. In recent years the scale of that preference has been conspicuous. According to the second-quarter 2026 results materials, capital expenditure over the trailing twelve months was $173 billion, up 64 percent year on year. On the same quarter’s call, management put the full-year 2026 capital expenditure outlook at roughly $220 billion. That was a raise from the prior outlook, and the reason given was higher memory prices. Future Capital, in the canon, is not financial capital alone. Future Capital = Financial × Human × Learning × Trust × AI × Knowledge × Ecosystem × Purpose This is multiplication. If any single term is zero, the whole product is zero. Investment in data centers belongs to the Financial and AI terms. By itself it produces no value. It works only alongside operational knowledge, the trust surrounding population of businesses. of developers, and the

3.5 Leadership — the form of the conversation with investors

was changed The last dimension is management itself. What the company did here was change the form of its explanation. It does not ask to be assessed on quarterly profit. It speaks instead in terms of cash flow and the payback period on investment. On the second-quarter 2026 call, management described the useful life of a data center as more than 30 years, and said that several generations of server economics follow from it. Management also said that the following year’s capacity was already close to fully reserved. This is not an act of demanding patience from investors. It is the act of handing investors, in advance, the material with which to assess the long term. Miss that distinction and long-termism becomes a refusal to explain. The Leadership dimension yields one more reading. Cutting an organization into small teams is also a decision about where decisions are made. Keep the units small while continuing to decide at the center, and the center clogs. Once you have divided the work, you have to divide the places where things are decided. Who decides what is not a series of individual judgments. It is an object of design. And a designed arrangement outlives the executive who designed it.

4 Structure — maturity, the value chain, and the

equations

4.1 Which level of the Enterprise Redefinition Maturity Model

The Enterprise Redefinition Maturity Model (ERMM) has five levels. Here we avoid a verdict and speak dimension by dimension. The model evaluates organizational coherence rather than isolated excellence. For the Business and Capital dimensions, what can be read from public information is the profile of a Continuous Redefinition Enterprise (Level 4). Business models are continuously replaced, and resources are allocated toward future capability rather than past success. There are several episodes that read as redesigning the enterprise before external disruption required it. Looked at through the single fact that AWS brought a new industry into being, another reading is available. It is close to the description of a Future Value Enterprise (Level 5), an organization that actively shapes future industries. That, however, is an observation about one business. It cannot be asserted as the level of the enterprise as a whole. In the Organization and Leadership dimensions, external criticism has continued. Warehouse working conditions have been raised in the European Parliament. In the United States, the Federal Trade Commission and several states sued in September 2023, alleging the maintenance of a monopoly. In 2024, the core claims were allowed to proceed. In September 2025, the Commission announced a settlement of $2.5 billion concerning Prime enrollment and cancellation procedures. The levels are therefore not aligned across dimensions. Progression is not linear: an organization may hold Level 4 capability in one dimension while remaining at Level 2 in another. Maturity is assessed across all five dimensions in balance, and exceptional capability in one dimension does not raise the whole. Nor is Level 5 a target to be reached as fast as possible; the appropriate level differs by industry and environment. What we should take from this case is that a high capacity to redefine the business does not automatically bring maturity in organization and governance.

4.2 Where in the Future Value Chain the value was created

Purpose → Learning → Redefinition → Creation → Enterprise Value Enterprise Value comes last. Read Amazon in this order and the point of origin becomes clear. Value was created between Learning and Redefinition. The company’s own engineers were losing time to the construction of infrastructure. Most companies process that as a problem to be solved, and stop there. This company treated it as learning: the same inconvenience exists outside as well. Then Redefinition occurs. The definition of what kind of company this is gets rewritten. From a company that sells books to a company that supplies the platform on which other people’s businesses run. Creation follows from that. A market that did not exist became real. Enterprise Value arrived after that again. Imagine a company that runs the order backward. Management that starts from enterprise value and works back cannot choose to sell its internal systems. Early revenue is small, the existing business’s profit is squeezed, and internal opposition is heavy.

4.3 Two equations to give the case a skeleton

The formula for the capability to create Future Value is as follows. FVCC = Purpose × Learning × Redefinition × AI Integration × Ecosystem × Capital Allocation × Trust What multiplication means becomes concrete here. The company has kept Learning, Redefinition, and Capital Allocation high. The Trust term, meanwhile, remains exposed to criticism over working conditions and competition policy. Because the relationship is multiplicative, damage to that term pulls the whole product down. Trust does not appear in the financial statements. It arrives late, through regulation, through hiring, and through what customers choose. Set one more equation alongside it — the equation of time. Future Value = Future Time × Future Capability Long-termism means taking the left-hand term large. If the righthand term is zero, no amount of time produces Future Value. The reverse also holds: capability without time granted to it yields zero as well. Long-termism worked here because capability was being built at the same time. That is the limit of the second conventional answer.

5 What it looks like in practice — the turning points, and

what was let go

5.1 Four turning points in sequence

The first turning point came immediately after the listing. The 1997 letter to shareholders declared that the long term would take priority over short-term profit. Making the declaration first is what later supplied the grounds for investment decisions. The second turning point was the opening of internal functions. In 2002 the company opened its systems to outside developers; in March 2006 it released a storage service, and in August 2006 a compute service. At that point an internal capability became a product. The third turning point was disclosure. In 2015, the company reported AWS results separately for the first time. A business that had been buried inside the retail numbers could now be valued independently by the market. Redefinition is completed by disclosure. A redefinition invisible from outside is not reflected in enterprise value. The fourth turning point is reallocation toward AI. In November 2025, AWS announced a multi-year strategic partnership with OpenAI, reported at $38 billion. Alongside it, the company has been investing in its own semiconductor designs. According to the second-quarter 2026 call, its AI-related and semiconductor-related businesses have each passed an annualized scale of $25 billion. AWS backlog in the same quarter was described as $496 billion. AWS revenue in that quarter was $42.2 billion, up 37 percent year on year. Company-wide revenue was $200.6 billion, and operating income was $27.5 billion. Note that net income of $62.6 billion in the quarter includes a non-operating valuation gain of $53.4 billion related to the investment in Anthropic. That item has to be read out of the figure.

5.2 What was let go

Redefinition is not an act of addition. It is the act of deciding what to protect and what to release. Three things this company released. First, the appearance of short-term profit. Prioritizing longterm investment makes margins look lower than peers’ in any given period. The company took on the burden of explaining that, continuously. Second, exclusive control of its own shelf. Opening the marketplace meant placing competing products on its own shelves next to its own. In the short run that decision cuts its own gross margin. In time, the same structure became a central issue in competition policy. What you let go always creates a new liability. Third, continuity of headcount. Reductions of roughly 14,000 employees in October 2025 and roughly 16,000 in January 2026 have been reported. Management explained the moves in terms of organizational layers and decision speed, while reporting also raised the effect of automation. The cost of redefinition is frequently borne by people. A case study that looks away from this is of no use to management.

5.3 Why other companies did not do the same thing

One question remains. This company was not the only one operating computing infrastructure at scale internally. Banks, telecommunications carriers, and large retailers all held facilities and operating organizations of their own. The option of opening them was available to several companies. Almost none took it. Where was the difference? The canon divides Enterprise Redefinition Capability into six: Strategic Intelligence, Learning Capability, and Design Capability; and Capital Reallocation Capability, Leadership Capability, and AI Collaboration Capability. We read the third and the fourth as decisive here. Take Design Capability first. Design capability is not the power to think of something. It is the power to bring a conception down into a working structure. Opening internal infrastructure to outsiders requires decisions made in advance. The unit of billing. Responsibility when there is an outage. Identification of users. A service level that does not stop another company’s operations. Infrastructure built only for oneself has none of these. Most companies stopped there. Internal systems grow to fit the circumstances of each department, and extracting one means rebuilding it. The moment the estimate appears, the proposal dies. A conception dies not because will is weak. It dies because the distance to a working design is too great. The difference in Capital Reallocation Capability runs deeper. Capital reallocation capability is the power to shift resources from past success toward future capability. Selling internal infrastructure outside generates almost no early revenue and incurs cost first. That cost falls on budgets assembled for existing businesses. The head of an existing business has no reason to raise another unit’s new business out of their own budget. The 1997 letter to shareholders had already moved that premise. The difference, then, was not in the idea. It was in the implementation. Where the distance to a working design is long, and capital is pinned to existing businesses, the idea stops there. What halts redefinition is usually not a competitor. It is the company’s own design and its own budget.

5.4 We reserve judgment on the future

The above is an observation based on public information as of August 2026. About the future we make no assertion. Investment at this scale rests on the assumption that demand continues. If the assumption breaks, the same investment turns into excess capacity. Rival computing platforms are also growing quickly, and there is no guarantee that the advantage persists. The regulatory questions remain unresolved. What this case shows is not a reproducible procedure for success. It is the order in which a redefinition occurred.

6 What transfers, and questions for the executive

6.1 Four implications

First, take the inventory of internal capability in functional language, not in product names. What does the company actually possess? As long as that question is answered with product names, no destination for redeployment will appear. Not the power of quality control, but the power to hold variance within a fixed band. Not the power of logistics, but the power to absorb swings in demand. The moment the words change, the adjacent industry comes into view. Second, design internal systems in a form that can be opened outward. Redeploying capability is a matter of design before it is a matter of conviction. Nothing can be extracted from arrangements fused to individual departments. Does the system you are building now have a shape that could be sold in five years? That is a question for the executive, not for the IT function. Third, hand over the material for assessing the long term before asking for the long term. Many incumbents ask investors to understand long-term investment. Understanding does not arise without material. How much is going where, over what period, for what return? Under what conditions does the payback assumption break? Speak of long-termism without that disclosure and it will be received as an abdication of explanation. As 100 Questions on Management in the Age of AI, #070 argued, long-term investment means something only when time and capability grow together. Fourth, know that redefinition is completed by disclosure. While the numbers of a new business stay buried in the existing business, the market cannot perceive the redefinition. Separate disclosure of AWS did not change the substance of the business. It changed how the business could be seen. That alone changed what was being valued. The first implication deserves a procedure. An inventory begun on impulse stalls halfway. We recommend four stages. The first stage is to relist the systems in use inside the company by function rather than by department. Not “the accounting system” but “the arrangement that reconciles a large daily volume of transactions.” Replacing names with functions changes the character of the list. What this requires is not technical knowledge. It is the discipline of banning product names. The second stage is to write down who outside the company has the same inconvenience. If it cannot be written, the capability depends heavily on circumstances peculiar to the company. If it can, that party is the first candidate customer. The same inconvenience is often found in the industry next door. The third stage is to measure the distance to extraction. Could an external user use this as it stands? If not, what has to be rebuilt? Write the required period in months. Argue without measuring and the discussion ends in impressions. The fourth stage is pricing. Who outside would pay, and how much? Until a figure can be written, you have a list of capabilities and not a list of business candidates. Run the four stages and the candidates narrow to a handful. The narrowing is itself the result. The purpose is not to find many. It is to find one that can be started.

6.2 Three questions

Three questions to close. Each can be answered at your next executive meeting. Question 1 — Among the systems you built for yourselves, which would another company want? If no answer comes, it may not be that the capability is absent. It may be that the capability has never been defined in functional language. Build the list first. Then, for each item, write who outside would pay and how much. That two-column table is the starting point of capability redeployment. Question 2 — When that system goes outside, who inside the company loses? Selling capability outward always collides with the interests of an existing unit. The collision cannot be avoided. What can be avoided is an unplanned one. Identify in advance whose metrics fall, and change those metrics first. A redefinition executed in the wrong order stalls inside the company while remaining a correct strategy. Question 3 — What material have you given investors about your long-term investment? Before asking for understanding, have you handed over the material for judgment? The amount, the period, the assumptions, and the conditions under which the assumptions break. An investment for which those four cannot be written is quite likely not agreed inside the company either. None of the three questions asks about technology. All three ask where you draw the line between inside and outside. What Amazon redefined was not a business and not an industry. It was the line dividing the inside of an enterprise from the outside. It read an internal inconvenience as an external market. That rereading was not analysis. It was definition. AI Optimizes. Humans Define. AI optimizes; humans define value, purpose, and direction. Where the line between inside and outside is drawn is still human work. And how that line is drawn determines how large an enterprise can become.

In brief

What Amazon redefined was neither a business nor an industry, but the line dividing the inside of the enterprise from the outside.
On what public information shows, Business and Capital read as a Continuous Redefinition Enterprise, while Organization and Leadership are not at the same level.
Value was created between Learning and Redefinition, at the point where an internal inconvenience was reread as a market.
If the Trust term thins under pressure from working conditions and competition policy, the product is pulled down.

Key concepts

Enterprise Redefinition Capability / the Enterprise Redefinition Maturity Model / Future Value Chain / Future Capital / the Layer Shift Pattern / the Capability Redeployment Pattern (→ Vol. VI, Ch. 059)

The chain of ideas

Learning → Design Capability → Capital Reallocation Capability → Redefinition → Enterprise Value

Related first principles

Principle 2 — Future Value Precedes Enterprise Value. Principle 3 — Capital Exists to Create Possibility. Principle 6 — Enterprise Exists to Redefine Itself. Principle 8 — Trust Compounds Faster Than Capital.

Related chapters

Vol. VI, Ch. 059 “Cases of Enterprise Redefinition” — the definitions of the Layer Shift Pattern and the Capability Redeployment Pattern are there
Vol. V, Ch. 043 “What Is Enterprise Redefinition Capability?” — the substance of design capability and capital reallocation capability
Vol. V, Ch. 047 “What Does It Mean to Redefine the Organization?” — why internal boundaries set the range of business opportunity
Vol. VIII, Ch. 075 “What Is IR in the Age of AI?” — designing the advance handover of material for assessing the long term

Papers and companion volumes

Kadowaki, N. (2026a). Future Value Theory: A Management Framework for Enterprise, Capital, and Society in the Age of AI. VURA Working Paper Series. SSRN: https://ssrn.com/abstract=7120980 / Zenodo: https://doi.org/10.5281/zenodo. 21255662
Kadowaki, N. (2026b). Enterprise Redefinition: Toward an Enterprise Evolution Theory for the Age of AI. VURA Working Paper Series. (Published on Zenodo; under review at SSRN)
100 Questions on Management in the Age of AI, #070 “Why Does Long-Term Investment Matter in the Age of AI?”

Chapter 083 What Did Amazon Redefine?