Chapter 071　How Is the Enterprise Value of an AI Company Determined?

How is the Enterprise Value of an AI company determined? The question is asked constantly and answered vaguely. The reason is simple. The phrase “AI company” bundles together firms whose cost structures and barriers to entry have nothing in common. This chapter does not value individual companies. That work belongs to Vols. IX–X. What we do here is cut the industry into layers, and show, as structure, how the determination of value differs from one layer to the next.

1 The question — why it arises now

“AI company” became ordinary language within a few years. It is used in executive meetings, in conversations with investors, and in meetings with banks. What it refers to differs with every speaker. A firm that designs semiconductors is called an AI company. So is a firm that runs data centers, and a firm that develops foundation models. So is a firm that builds services on top of them, and an incumbent that has rebuilt its operations around AI. These firms differ by orders of magnitude in the capital they need. They differ in the years it takes to reach profit. They differ in the conditions under which competition ends. Use one word for different things and the argument will never close. “Are AI share prices too high?” never settles, because the answer differs by layer. In Vol. VII, Ch. 062 we argued that the source of Enterprise Value in the Age of AI moves from accumulated assets to the capability to redefine. That was an argument about the source. This chapter is about industry structure. The same theory of source produces a different appearance of value depending on which layer of the industry a firm stands in. Why is this sorting needed now? Two reasons. First, capital is concentrating in particular layers. Concentration raises the Enterprise Value of that layer and raises the costs of the others at the same time. The prosperity of one layer is built out of the burden of another. Second, many executives measure their own company with another layer’s ruler, without noticing which layer they occupy. Apply the criteria that fit a foundation-model firm to a manufacturer that uses AI, and nothing comes into view. The question is therefore restated. Not how the Enterprise Value of an AI company is determined, but which factors determine the value of a firm standing in which layer of the industry.

2 Conventional answers and their limits

Three answers circulate today. Each is partly right. None of them divides the industry into layers. The first answer: “AI is a growth industry, so AI companies will carry high Enterprise Value” This is the answer heard most often. The market expands, so the value of the firms inside it expands. Intuitively it is correct. But the growth rate of an industry and the distribution of value inside it are separate problems. History repeatedly shows industries that grew while the firms inside them did not make money. Air transport saw demand rise for decades; little profit stayed with the carriers. Solar panels tell the same story. When demand rises and the suppliers are homogeneous, price falls to cost. Value stays only where the supply side carries a structural constraint. Constraint means the cumulative nature of a technology, the scale of the plant required, or a reason customers cannot leave. Growth, then, is only a necessary condition. Who receives the value is decided by structure, not by growth. The point is especially easy to miss in AI. Wider use of AI does raise the revenue of the industry as a whole. Where that added revenue stops is an entirely separate matter. If most of what an upper layer pays flows straight down to the layer below, little stays above. The second answer: “The value of an AI company is determined by model performance” The second answer argues from performance. The firm with the smarter model wins. But performance decays. This is the decisive difference from earlier technology industries. Factory equipment lasts a decade. Patents are protected in units of years. A model’s performance lead has been closed on a far shorter cycle. To argue from performance, compare two speeds. The speed at which followers close the gap, and the size of the effort a customer must spend to switch. Where the first is fast and the second is small, a performance lead is not converted into Enterprise Value. Performance also changes meaning as it crosses layers. To a lower layer, performance is the product. To an upper layer, performance is a purchasing term. The same improvement is a source of revenue on one side and downward pressure on price on the other. The third answer: “In the end, the firm with the data wins” The third answer argues from data barriers. Data cannot be copied, so whoever holds it stands ahead. This too is partly right. But not all data carries the same value. Published text and images are not scarce. What is scarce is data that arises only inside the running of a business. Failure histories of equipment. The outcomes of credit decisions. The course of a treatment. The tacit judgments made on the floor. None of these can be bought on a market. Data decays as well. Customer behavior changes. Equipment is replaced. Three-year-old data does not necessarily support today’s judgment. Data becomes a barrier only when the firm holds the mechanism that keeps generating it. What all three answers miss All three treat “AI companies” as a single industry. In fact the AI industry is built from several layers stacked vertically. Change the layer and the cost structure changes, the payback period changes, the kind of barrier changes, and the conditions under which value stays change. Before analysis can begin, the bundle has to be undone.

3 Redefinition — the unit of analysis is not the AI

company but the layer From the standpoint of Future Value Theory, we state the central proposition of this chapter. “AI company” does not hold as a unit of analysis for Enterprise Value. The unit that holds is the layer of the industry. The AI industry divides into at least five layers. We take them from the bottom.

3.1 Layer 1 — the compute foundation

Semiconductor design and manufacture, the equipment that makes them, and the materials. The layer that makes AI computation possible at all. Its character comes from the cumulative nature of the technology and the scale of the plant. One generation of process technology can only be built on the one before. Plants cost enormous sums and take years to reach operation. The number of firms in the world able to build them is structurally limited. Enterprise Value in this layer is determined less by the size of demand than by how many firms can supply. Where there are few, pricing power stays on the supply side. But the layer cannot make its own demand. If the layers above stop investing, its demand disappears. Strong, but not independent.

3.2 Layer 2 — infrastructure

Cloud, data centers, and the power and cooling behind them. The layer that turns computation into something usable and delivers it. Its character comes from heavy up-front investment and a long payback period. Built capacity depreciates. When the technology generation turns, capacity can lose competitiveness before it is written off. Equipment here is an asset with some of the properties of a liability. What determines Enterprise Value in this layer is not the equipment. It is the customer relationship sitting on top of it. A customer who has placed data, connected operations, and embedded permission design cannot leave easily. Utilization is a result, not a cause.

3.3 Layer 3 — foundation models

The layer that trains and supplies large models. It behaves unlike any other. Training costs enormous sums, but a finished model does not survive long as an asset. When the next generation arrives, the economic value of the previous one falls sharply. Factories remain. Models do not. Enterprise Value here is therefore not determined by the performance of the model a firm holds now. It is determined by the capability to keep shipping the next model. Capital, compute, people, and the speed of learning. The moment any of these breaks, a firm in this layer loses its source of value. This is where Future Value Theory applies most sharply. Value is not the past result. It is the capability to keep creating. First Principle 2 states it. Future Value Precedes Enterprise Value. Future Value comes before Enterprise Value (the market’s valuation).

3.4 Layer 4 — applications

The layer that uses foundation models to build services aimed at particular operations and particular customers. Its character is ease of entry. Initial investment is small. Most of the technology can be procured from the layers below. The number of firms is therefore large, and turnover is fast. Enterprise Value here is determined not by technology but by how deeply the product is embedded in the customer’s work. A merely convenient tool is replaced. A tool that has become the procedure itself is not. This layer holds a double relationship with the layers below it. Price cuts below become profit here. But when a lower layer extends its functions upward, the reason for this layer to exist is cut away. It carries a structural instability: a supplier can become a competitor.

3.5 Layer 5 — incumbents that use AI

The last layer is made of firms that do not sell AI. Manufacturing, retail, healthcare, finance, construction, and the small and midsize firms of the regions. AI is not the product. It is an input. This layer is usually set aside in discussions of the AI industry. Yet by number of firms and by employment, the overwhelming majority sits here. So do most readers of this chapter. How Enterprise Value is determined in this layer differs fundamentally from the other four. Section 5 takes it up in detail.

3.6 How the layers connect

Listing five layers is half the work. The relations between them have to be read. The first relation is the flow of money. The revenue of a lower layer is made of the spending of an upper one. What Layer 5 pays for AI becomes Layer 4’s revenue, and what Layer 4 pays becomes Layer 3’s revenue. In the end the chain depends on whether the final user actually feels the value. If no value is created at the end of the chain, upstream revenue is temporary. The second relation is the flow of cost. Profit in an upper layer is bound by prices in a lower one. Cuts below raise profit above; increases below cut it. Layers 4 and 5 are highly sensitive here. The third relation is functional encroachment. Every layer has an incentive to extend upward. Compute foundations try to supply tools. Foundation models reach into applications. The Enterprise Value of a layer therefore cannot be explained by competition inside that layer alone. Movement in an adjacent layer shakes the reason the layer exists.

3.7 The layers are not fixed

Three notes. First, a firm can straddle several layers. Straddling produces the benefits of integration, and at the same time makes it ambiguous which layer’s ruler will be used to measure the firm. Second, the boundaries move. When one layer extends its functions up or down, the reason the adjacent layer exists contracts. The map of layers has to be redrawn every few years. Third, whichever layer a firm stands in, the order of the Future Value Chain does not change. Purpose → Learning → Redefinition → Creation → Enterprise Value What changes is the object and the cycle of Redefinition. The lower the layer, the longer the cycle. The higher the layer, the shorter.

4 Structure — five factors for reading a layer

We read the differences between layers with five factors. The five apply to every layer. They are a common ruler. First, the scale of capital investment and the payback period. The longer the years before invested capital returns, the more Enterprise Value depends on the outlook for future demand. When the outlook moves, value moves. Layers 1 and 2 are highly sensitive here. Second, switching costs. The size of what a customer loses in moving to a competitor. Data, work procedures, training, integration. Where this is small, a performance lead drains into price competition. Third, the decay speed of models. The time a technology asset takes to lose its economic value. The faster it is, the less accumulation counts, and the more the capability to reinvest determines value. It is fastest in Layer 3. Fourth, the data barrier. As set out above, only continuously generated data becomes a barrier. Layer 5 is often underrated on this point. Fifth, contact with the user. Who is directly connected to the final user. A layer that holds the contact learns of changes in demand first and sets its price last. Lose the contact, and the firm becomes a component supplier.

4.1 The equations give the frame

The central equation of Future Value Theory binds the five factors into one line. Value = Purpose × Trust × Capability × Time Do not miss that this is multiplication. Under addition, a weak term can be covered by the others. Under multiplication, the moment one term is zero the whole is zero. Because the relationship is multiplicative, value without purpose has no direction, without trust cannot spread through society, without capability cannot be realized, and without time cannot endure. Capability may be high, but if Trust is zero no value appears. The decision to hand over data is made on trust, not on performance. With Purpose at zero, capital has no direction. Purpose, Trust, Capability, and Time here are the four terms of the Value Equation. They are not the five elements of Future Value, and they are not the seven terms of the FVCC Formula. And Time means something different in each layer. Time in Layer 1 is the number of years until plant is paid back. Time in Layer 3 is the number of months until the next generation. The same equation works on a different clock in each layer. One more equation deals with time itself. Future Value = Future Time × Future Capability Future Value is the product of time directed at the future and the capability to create the future. It is not one or the other. A Layer 3 firm that raises capital without a matching speed of learning creates no value. A Layer 5 firm that gains time through AI and does not point that time at the future creates none either.

4.2 Read the composition of capital

Differences in strength between layers also show up in the composition of capital. Future Capital = Financial × Human × Learning × Trust × AI × Knowledge × Ecosystem × Purpose These eight are the eight forms of Future Capital — a different set from the five elements of Future Value and from the seven terms of the FVCC Formula. This equation is multiplicative as well. Layer 1 is thick in Knowledge Capital and Financial Capital. Layer 2 is thick in Trust Capital and Ecosystem Capital. Layer 3 is thick in Learning Capital and AI Capital. Layer 4 depends on Ecosystem Capital and Trust Capital. Layer 5 holds capital lying dormant in Human Capital and Knowledge Capital. Every layer has a term close to zero. The difference in Enterprise Value is decided not by the height of the thick terms but by how far the thin ones have been raised. An abundance of financial capital cannot compensate for absent purpose, and advanced AI cannot compensate for absent trust.

5 What it looks like in practice — the difficulty of

valuation, and where incumbents stand

5.1 Why AI companies are hard to value

Four reasons. First, technology moves fast. Valuation is the work of folding in the future, and the assumptions being folded in change while the work is under way. Second, enormous investment precedes monetization. Costs are booked today. Revenue appears years later. Through that period the financial statements do not show the state of the firm. The size of a loss can be evidence of failure or evidence of the scale of the investment. Third, nobody knows whether a barrier will hold. Whether a barrier that looks strong today is still effective in three years depends on the kind of barrier. Equipment barriers tend to hold. Performance barriers tend not to. Value a firm without separating the two and the answer will be wrong. Fourth, the number of winners cannot be read. Some industries end with one firm taking everything. Others settle with several side by side. Which one it becomes changes the value per firm by an order of magnitude. As of August 2026, which shape each layer of the AI industry settles into is not yet determined. None of the four is a difficulty of the “not enough information” kind. Each is a difficulty of the kind where the structure itself is undetermined. Refining the analysis does not dissolve it. This carries a practical implication for executives. Faced with an undetermined structure, do not manufacture determined numbers. A precise business plan carries meaning only where the assumptions are settled. What is needed here is not a single outlook but an explicit statement of conditions. Which conditions, if they hold, keep our assumptions intact? Which, if they break, require us to revise? Write that down first.

5.2 The state in which “being an AI company” lifts the valuation

As of August 2026, there are settings in which the mere fact of being involved with AI lifts a firm’s valuation. How should we read that state? It is not necessarily irrational. When structure is undetermined, markets price a bundle of probabilities. Even where the chance of becoming a winner is small, the value in the winning case can be extreme enough to raise the expected value. A high valuation can be a statement of variance rather than a statement of conviction. The danger sits elsewhere. What is dangerous is not the height of the price but the loss of the distinction between layers. When firms are valued together under the heading “AI company,” structurally strong layers and weak ones are measured with the same ruler. In that state, capital does not travel to where value is most likely to stay. We set out the conditions for this state to continue, with a reservation. What follows is not a forecast. It is an ordering of conditions. Three conditions for continuation. First, the appetite for investment in the upper layers is sustained, since the revenue of a lower layer is made of the spending of an upper one. Second, gains in performance keep translating into wider use. Third, the expected timing of payback does not slip far backward. Three conditions for a break. First, supply clearly exceeds demand somewhere in the stack. Second, performance gaps narrow and a layer becomes internally homogeneous. Third, it becomes widely understood that monetization will take longer than first assumed. We do not assert when, or in which direction, this moves. As of August 2026, one thing can be said with confidence. When it breaks, it breaks differently by layer. A bundled valuation does not come apart as a bundle. Layers adjust at different times and by different amounts. What an executive should prepare for is therefore not the outlook for the market as a whole but the conditions of their own layer.

5.3 The Enterprise Value of incumbents that use AI

Then there is Layer 5. This is the most pressing topic for most readers of this chapter. One thing has to be confirmed first. Adopting AI barely moves Enterprise Value in this layer. Adoption becomes standard across every firm within a few years. What has become standard is not a difference. Having an accounting system is not a competitive advantage either. So what determines value? Three things. First, what the slack AI created was allocated to. AI returns time and hands. The returned slack has only two destinations. Collect it as cost reduction, or invest it in creating new value. The first is a one-time effect. The second accumulates. The gap between two firms with identical deployments comes not from the skill of the deployment but from this allocation. Second, whether data that arises only from the firm’s own business is being turned into an asset. Layer 5 firms generate, every day, data the lower layers can never obtain. Judgments made on the floor, conversations with customers, the behavior of equipment. In most cases it disappears unrecorded. Only firms that record it, structure it, and connect it to judgment hold a barrier specific to this layer. Third, whether contact with the customer is retained. Procuring AI from outside is not the problem. The problem is handing the customer relationship itself to an external mechanism. A firm that loses the contact becomes the last to learn that demand has changed. None of the three belongs to the success or failure of a deployment project. They are questions of capital allocation and design, not questions of technology. First Principle 3 states it. Capital Exists to Create Possibility. Capital exists to create possibility, not merely to maximize return. One reservation. Layer 5 firms are not pressed to redefine at the same speed as firms in the layers above. The Enterprise Redefinition Maturity Model (ERMM) states that reaching Level 5 as rapidly as possible is not the objective, because different industries may require different levels of organizational adaptability. Copy the cycle a foundation-model firm needs into a regulated business, and the organization exhausts itself for nothing. What is needed is a cycle matched to your own layer. Two further cautions travel with the model wherever it is used. Progression through the levels is not linear. Organizations frequently display characteristics from several levels at once, and a firm may hold Level 4 AI capability while remaining Level 2 in leadership. The model evaluates organizational coherence rather than isolated excellence. Maturity is also assessed across all five dimensions in balance. Exceptional technological capability with weak redesign of leadership cannot reach higher maturity, and strong purpose without adaptive organizational systems remains insufficient. One more point. Layer 5 firms hold a strength the other layers do not. They already have the customer’s problem. Firms in the upper layers hold technology and are looking for problems. Layer 5 firms hold problems and are looking for technology. Which position is the better one cannot be settled in the abstract. The strength does not become value on its own. A firm that merely has a problem goes on merely having it. The problem turns into Future Value only when it is put into words as Purpose and set as the object of learning and redefinition. First Principle 7 states it. Social Challenges Are Future Opportunities. Social challenges are the origins of future markets, industries, and capital.

6 Questions for the executive

The argument, in one line. The Enterprise Value of an AI company is not determined by its being an AI company. It is determined by which layer of the industry the firm stands in, and by what counts, in that layer, as the condition for value to stay. The first thing an executive should answer is therefore not whether the share price is justified. It is which layer the company occupies. Three questions to close. Question 1 — Which layer do you stand in, and what are the conditions for value to stay there? Most firms cannot answer immediately. Firms straddling several layers find it hardest. But without an answer you cannot know which factors move your Enterprise Value. Of the five factors in the previous section, which are acting on you? Which are not? That sorting is the starting point for everything else. Question 2 — Is your barrier a barrier of equipment, of performance, or of relationship? The kind of barrier fixes its duration. Equipment barriers take time to imitate. Performance barriers disappear quickly. Relationship barriers hold as long as they stay buried in the customer’s work. Can you name which one your advantage rests on? An advantage that cannot be named usually does not exist. Question 3 — What did we use the time AI returned to us for? For a Layer 5 firm this is the heaviest question. Cost that was cut appears once, as next period’s profit. Time invested in the future appears years later, as a business. Put the budget table beside the allocation of hours, and which one the firm chose is visible at a glance. None of the three questions asks how to use AI. All three ask where the firm stands inside the structure AI changed, and what it will create there. The map of layers will keep being rewritten. Boundaries will move, and the number of winners will change. The map as of August 2026 will look dated in a few years. One thing does not change when the map does. Enterprise Value is formed as the result of a firm’s power to create the future. That is an order, not a coincidence. First Principle 6 states it. Enterprise Exists to Redefine Itself. Continuous self-redefinition is its essence. The layer is given. How a firm stands inside it is chosen. And a firm can choose to move layers. How is the Enterprise Value of an AI company determined? The answer differs by layer. In every layer, what determines it is not the accumulation of the past but the capability to create from here.

In brief

“AI company” does not hold as a unit. Enterprise Value is determined by which layer of the industry a firm stands in.
The conditions for value to stay differ by layer. Whether the barrier is one of equipment, of performance, or of relationship fixes how long it lasts.
The first thing an executive should answer is not whether the share price is justified but which layer the firm occupies.
Layer boundaries move. The map of layers has to be redrawn every few years.

Key concepts

Future Value / Enterprise Value / Future Value Chain / Enterprise Redefinition

The chain of ideas

Position in the layer → cycle of Redefinition → Capability → Future Value → 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.

Related chapters

Vol. VII, Ch. 062 “What Is Enterprise Value in the Age of AI?” — how the determination of Enterprise Value changes in the Age of AI
Vol. IV, Ch. 031 “How Does AI Change Future Value?” — how AI rearranges the content of Future Value
Vol. VIII, Ch. 072 “What Is the Enterprise Value of a Startup?” — the mechanism by which a firm with no track record gets a price
Vol. IX, Ch. 081 “What Did NVIDIA Redefine?” — readable as an instance of a firm standing in the compute foundation layer

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, #085 “Where Does the Money Gather in the Age of AI?”

Chapter 071 How Is the Enterprise Value of an AI Company Determined?