AI is best analysed as a value chain rather than a single asset class. Few companies are purely AI-native, while many are embedding AI into existing products, services and operations. Edison maps AI exposure across seven layers: equipment manufacturers, semiconductors and components, data centre infrastructure, cloud platforms, foundation models, application software and end-users adopting AI. As of June 2026, much of the value created by AI has accrued to companies exposed to the infrastructure build-out, particularly in the lower layers of the stack. Meanwhile, many application software companies have already been rerated on expectations of future AI-driven growth. However, the longer term distribution of value across the AI ecosystem is likely to prove far more nuanced, with winners and losers emerging at every layer of the value chain.

Both can be true at once: AI is creating genuine economic value while parts of the market exhibit bubble-like characteristics. Every major technology cycle has been accompanied by periods of excessive optimism and speculative behaviour. AI may prove to be the most significant technology cycle yet, given the speed, scale and breadth of the transformation it is driving across industries and business models.

History suggests that such cycles create substantial winners but also many losers, as early expectations are tested against commercial reality. Edison noted in October 2025 that some valuation measures were approaching levels last seen during the dot-com era, with the cyclically adjusted price-to-earnings (CAPE) ratio near 39.6x, compared with a peak of around 44x in 2000 and a long-run average of roughly 17x. However, the comparison is imperfect. Today's leading AI companies generate substantial revenues, cash flows and profits, whereas many of the most highly valued technology companies in 2000 had yet to establish sustainable business models.

At the same time, we are beginning to see increasingly broad-based valuation expansion across companies with varying degrees of AI exposure. Such indiscriminate rerating has often been a feature of the later stages of previous technology booms and can be a warning sign that expectations are running ahead of fundamentals. That does not necessarily imply that AI itself is a bubble; rather, it suggests that parts of the market may be vulnerable to correction if growth, adoption or returns on investment fail to meet expectations. The key indicators to watch are the pace of AI-related revenue growth, evidence of commercial adoption and the ability of companies to convert AI investment into sustainable earnings growth.

The answer depends heavily on where a company sits within the AI value chain. The question of defensibility is most relevant at the application software layer and above, where access to foundation models is increasingly commoditised. Here, we look for differentiated and proprietary datasets, specialist domain expertise and deep integration into customer workflows. In many cases, the competitive advantage comes not from the AI itself but from the ability to apply it within a specific industry context that is difficult to replicate. Companies that span an entire industry or workflow often enjoy a further advantage, as they can aggregate data, insights and best practice across a broader customer base than any single organisation can achieve independently.

Distribution, customer relationships, regulatory expertise and switching costs can be equally important sources of competitive advantage. Indeed, as foundation models become more widely available, ownership of the customer relationship may prove more valuable than ownership of the model itself. The companies most likely to capture long-term value are those that control the point of engagement with the customer, understand their workflows and can embed AI seamlessly into existing processes.

Ultimately, the immediate risk for many SaaS companies is not that they will be replaced by AI, but that AI reduces the value of adjacent features and limits their ability to capture incremental growth opportunities. The most defensible businesses are therefore those that use AI to strengthen existing moats, creating products that improve with usage and become increasingly embedded within customer workflows over time.

AI infrastructure companies should be valued less on their exposure to AI and more on their ability to convert AI-driven demand into sustainable returns on capital. Unlike application software businesses, their economics are determined by physical capacity, utilisation rates, pricing power, contract quality, power availability and capital intensity. The most attractive opportunities tend to be companies that occupy bottleneck positions within the value chain, whether through access to scarce technologies, critical components, power infrastructure or specialised data-centre capacity. In these cases, the key question is not how quickly AI demand grows, but how much of that demand can be captured as durable cash flow and attractive returns on invested capital.

Investors should also distinguish between temporary beneficiaries of the current AI infrastructure build-out and businesses with long-term competitive advantages. Many companies are currently benefiting from supply shortages, strong order books and elevated margins, but these conditions may not persist as capacity expands. As a result, factors such as the duration of customer contracts, exposure to recurring inference demand rather than one-off training expenditure, customer concentration and barriers to entry become increasingly important. Ultimately, the best AI infrastructure investments are likely to be those that can continue deploying capital at attractive returns long after the initial AI investment cycle has matured.