Part One of this series considered what happens when the gap between demonstrated performance and future expectations becomes unusually wide. Artificial intelligence (“AI”) makes that tension particularly visible because many companies operating across the AI ecosystem are being valued on outcomes that have yet to be proven.
The principles of valuation do not change simply because the evidence becomes less certain. What changes is the work required to develop, challenge and support the assumptions on which value depends.
Historical performance remains relevant, but it may no longer provide enough information on its own. When markets are still developing, commercial models are evolving, and future returns depend on substantial investment today, the independent expert has to look beyond the track record without losing the discipline that historical evidence normally provides.
The challenge is therefore to form a view of value that distinguishes between what is supported by the evidence available today, what remains uncertain and which assumptions carry the greatest influence over the outcome.
When History Offers Fewer Answers
Historical financial information is valuable because it places boundaries around expectations. It shows what a business has achieved, how margins have behaved, how much growth capital has been required and whether management has historically delivered against its plans.
Where that record is limited or becoming less representative of the future, the evidence does not disappear. It becomes more fragmented.
For companies operating across the AI ecosystem, useful evidence may sit outside the income statement. Customer adoption, contracted revenues, unit economics, capital requirements, infrastructure capacity, industry economics, and the commercial progress of comparable businesses may all help build a clearer picture of what future performance could look like.
The challenge is deciding how much weight each piece of evidence deserves.
A large addressable market, for example, can support a growth thesis without demonstrating that a particular company will capture a meaningful share of it. Strong customer adoption can support revenue expectations without proving that those revenues will ultimately translate into attractive margins. Significant capital investment can demonstrate conviction without proving that the returns on that capital will justify the investment.
This distinction is particularly relevant in AI. The current infrastructure build-out is creating the capacity on which future adoption will depend, but investment in data centres, compute and supporting infrastructure is not itself evidence of the productivity or cash flows that may ultimately emerge from that capacity.
For the independent expert, the question is therefore not simply whether the AI opportunity is large. It is what evidence supports the assumption that the business being valued will participate in that opportunity, and on what economics.
Beyond the Forecast
The absence of a long track record often places greater emphasis on management forecasts. Those forecasts may be highly detailed, extending across customers, products, markets and cost categories. The detail is useful, but it is not the same as evidence.
A forecast can be internally coherent and still depend on assumptions that have limited external support. Revenue can reconcile perfectly to customer numbers, pricing and market penetration, while the market-share assumptions themselves remain highly uncertain. Margin expansion can be mathematically consistent while depending on scale benefits that have yet to be demonstrated.
The role of the independent expert is not to replace management’s view with another forecast. It is to understand what drives the forecast and identify which assumptions carry the greatest influence over value.
That distinction matters particularly in AI because many companies are making assumptions about markets that are still developing. The size of the future profit pool may itself be uncertain, and even where the overall opportunity proves substantial, a second question remains: who captures the economic value?
The answer will not necessarily be distributed evenly across the AI value chain. Infrastructure providers, semiconductor companies, model developers, cloud platforms and application businesses may all benefit from adoption, but that does not mean each layer will earn the level of returns implied by its own growth expectations.
A forecast therefore needs to do more than demonstrate that AI will create value. It needs to support why that value accrues to the specific business being valued.
Accounting for More Than One Future
Where several materially different outcomes remain plausible, relying on a single forecast can imply a level of certainty that the available evidence may not support.
Scenario analysis provides one way of addressing that problem. Rather than forcing every uncertainty into one central forecast, different commercial outcomes can be considered explicitly. A business may, for example, achieve rapid adoption and establish a durable market position, grow successfully but face sustained margin pressure, or fail to achieve the scale assumed in its base case.
The purpose is to recognise circumstances where different commercial paths could produce meaningfully different values.
Probability weighting may also be useful where those outcomes are sufficiently distinct, and there is a basis for assessing their relative likelihood. It does not eliminate judgement from the exercise. It simply makes the effect of that judgement more transparent.
Uncertainty can also affect different parts of a valuation in different ways. Some risks may influence forecast cash flows directly, while others may affect the return required for bearing those risks. The appropriate treatment will depend on the nature of the uncertainty, the evidence available and the methodology being applied. What matters is that its effect on value is considered deliberately rather than dealt with through a broad adjustment simply because the future is difficult to predict.
External constraints are equally important. A forecast may assume substantial growth in compute capacity, but that growth ultimately depends on physical infrastructure, capital and access to power. If those constraints affect how quickly a business can expand, they also affect the cash flows being valued.
Regulatory and political developments can have a similar effect. Their relevance lies not in the headline risk itself, but in how they may influence adoption, compliance costs, investment requirements or the range of commercial outcomes available to the business.
The Value of Staying Power
Part One introduced the durability of competitive advantage as one of the central challenges in valuing companies across the AI ecosystem.
A company may have a genuine competitive advantage today. The more difficult valuation question is how long that advantage can support superior economic performance.
In AI, that assessment can be particularly challenging because the commercial landscape is changing as fast as the technology. Model performance improves quickly, barriers to entry can shift, infrastructure economics continue to evolve, and capabilities that are scarce today may become more widely available.
The strength of a current market position therefore cannot automatically be extended throughout a long-term forecast.
Evidence of durability may come from several sources: proprietary data, embedded customer relationships, switching costs, distribution, intellectual property, infrastructure advantages or the capital required for competitors to replicate a position. None is decisive on its own. The question is whether those advantages are likely to remain economically meaningful for as long as the valuation assumes.
That distinction matters because the duration of superior performance can influence value as much as the level of performance itself. A company may plausibly earn attractive returns, but the period over which those returns can be sustained remains a separate judgement.
From Infrastructure to Economic Value
The scale of investment across the AI ecosystem provides visible evidence of how seriously businesses are positioning for future demand. Data centres are being built, compute capacity is expanding, and significant capital is being committed across infrastructure, models and applications.
The next question is what that investment ultimately produces.
Infrastructure creates capacity. It does not determine how much productivity will be generated from that capacity, how quickly adoption will occur or which participants will retain the economic benefits.
This distinction becomes particularly important when assessing forecasts that assume substantial growth across multiple parts of the AI value chain. It is entirely possible for AI to deliver significant productivity gains and economic value while the distribution of those benefits looks very different from what individual company forecasts currently assume.
For valuation purposes, the size of the opportunity therefore cannot be considered separately from the company’s position within it. Growth in the overall market may support a forecast, but it does not remove the need to understand the economics of the business being valued.
The Discipline Remains the Same
AI makes the valuation challenge particularly visible because so much of the value attributed to companies across the ecosystem depends on outcomes that have yet to be demonstrated. Infrastructure must still translate into productive use. Adoption must translate into revenue. Revenue must translate into sustainable cash flows. Competitive advantages must persist long enough for those cash flows to justify the investment required to generate them.
None of those questions is unique to AI.
New technologies, emerging industries and rapidly changing business models have always required independent experts to form views where the historical record provides only part of the answer. AI simply presents a particularly current example of that challenge.
One needs to be more explicit about what is known, more disciplined about what is assumed and more deliberate about how uncertainty is reflected in value.
That is ultimately where professional judgement matters most.