Value investing in AI narrative-driven markets

Artificial intelligence is now best understood as a general purpose enabling layer – one that augments decision-making, automation and information processing across the economy. Its applications span public services, global enterprises, science, healthcare, defense and consumer platforms, embedding themselves into workflows rather than standing alone.

This is not controversial. What is far less settled is how this technological shift translates into economic value, competitive advantage and long-term returns. As AI moves rapidly from promise to deployment, we believe the central challenge for investors is in determining where it will sustainably create value, and where market narratives are moving faster than fundamentals.

Few technologies have attracted capital at the scale and speed of AI. Hyperscale cloud providers – namely Microsoft, Alphabet, Amazon, Meta and Oracle – have collectively committed over USD 1 trillion in data centers, networking and semiconductors over the past three years.¹ Annual capital expenditure at the largest players is running at levels previously associated only with oil majors or national infrastructure programs.

Critically, AI exhibits non-linear improvement characteristics. Model performance may improve through scale, data feedback loops and self-learning architectures, while demand may expand as costs fall and capabilities improve. We are seeing rapid and compounding gains in functionality. AI token consumption and overall compute demand are growing exponentially, while the unit cost of computation is declining. Progress is no longer governed by Moore’s Law’s predictable, hardware-driven gains, but by scale, data and software effects that allow capability to compound far more rapidly.²

While AI adoption curves will vary by industry, we expect this technology to become universally embedded in digital workflows, in much the same way the internet moved from a standalone product to invisible infrastructure within two decades.

AI will not impact all companies equally. The business and financial impact will be uneven, creating clear winners, structural losers and a wide middle ground.

The primary beneficiaries fall into two distinct groups. The first are infrastructure and picks-and-shovels providers: companies that design, own or control critical AI intellectual property and infrastructure, spanning leading-edge semiconductor manufacturers, cloud platforms, foundational model developers, specialist networking firms and data center infrastructure providers.

Technological relevance alone, however, does not guarantee attractive returns – capital intensity, competitive dynamics and pricing power generally determine economic outcomes. In many technology markets, returns can be highly concentrated on a winner-takes-most basis, but the identity of the winner is not always obvious. This makes balance-sheet strength, reinvestment capacity and ecosystem lock-in critical when underwriting long-duration cashflows.

The second group are pragmatic AI adopters: businesses that deploy AI to improve returns on capital rather than to create headline products. Applications range from accelerating revenue growth and improving pricing to personalizing offerings, optimizing supply chains and structurally reducing costs – in each case, AI functions as a margin and productivity lever rather than a product in itself.

Many of the most successful users may never be branded as ’AI companies’. We believe the most attractive adopters are often those with proprietary data, high-frequency decision loops and the ability to embed AI into workflows that customers are reluctant to rip and replace.

The downside also polarizes into two groups. The first are under-adopters – companies that underinvest in AI and organizational change due to complacency or weak execution – who risk the slow erosion of competitiveness as peers adapt more quickly.

The second, and more dangerous category for value investors, are structurally disrupted franchises: businesses whose underlying economics are undermined by third-party, AI-enabled competitors.

These are particularly treacherous because headline revenues or earnings may appear stable even as returns on invested capital deteriorate.

Rapid innovation compresses the timeframe in which disruption shows up in the financial statements, meaning value traps can form faster than in most areas, and seemingly cheap multiples can be a mirage rather than a margin of safety.

Value investors traditionally place greater weight on the present and past, favouring tangible evidence over more speculative views of the future.

AI challenges this instinct. Past success is no longer a reliable guide to future durability, and value investors must explicitly accept that business model disruption risk has increased. It also challenges a second core assumption: mean reversion. In many technology markets, supernormal returns can persist longer than expected because software economics and network effects can deepen moats over time. Meanwhile, losers can fade into irrelevance rather than reverting back to ’fair’ returns.

Avoiding losers is the primary discipline. Crucially, losers are rarely obvious ex-ante. Many disrupted companies are screening as cheap, stable and cash-generative. This reinforces the central role of value trap analysis within our investment process. By systematically assessing reinvestment rates, management’s ability to adapt, pricing power, customer switching costs and balance-sheet flexibility, we seek to distinguish temporary mispricing from structural decline. History shows that assets with eroding strategic relevance tend not to recover – remaining cheap or becoming cheaper still.

Clearly though, opportunities do exist. AI-driven uncertainty has created cases where bad news is over-discounted and optionality ignored. Applying a large margin of safety remains critical. One area we find interesting is software companies that support industry verticals. Here, SaaS (software as a service) applications are industry specific, bespoke and deeply embedded. This software is routinely used in regulated or operationally complex industries where switching costs are high due to proprietary datasets and deeply integrated customer relationships. Where these businesses can incorporate AI as an enhancement rather than a replacement, we see scope for durable moats alongside valuation underwriting that is currently overlooked by narrative-driven markets.

Additionally, value investors can uncover hidden value in businesses with undiscovered or latent AI exposure. This logic also extends to parts of the ’picks and shovels’ layer of AI. While much investor attention has gravitated toward highly visible infrastructure winners, crowded positioning means these stocks are trading at very high multiples in many cases. By contrast, we have focused on less obvious enabling segments – for example, specialized semiconductors tied to mobile and Internet of Things devices.

Overall, valuation nuance matters: a company on a headline ’high’ multiple can still be value if it has sustainably high returns on capital, strong cash conversion and reinvestment runway – whereas a low multiple can be expensive if durability is deteriorating.

Equity markets are actively trying to discount AI’s economic impact, but with very limited hard data. Investor behaviour has swung between enthusiasm and fear; at times defaulting to a ’shoot first, ask questions later’ mindset: SaaS models are pronounced dead, and entire layers of middleware and platform businesses are treated as structurally compromized.

Our view is deliberately contrarian. This is rarely how technological transitions resolve. In our view, AI does not eliminate software demand; it reshapes value pools, pricing models and delivery mechanisms. History suggests that fear driven extrapolation and fear of missing out (FOMO)-led trend following may generate momentum, but they seldom produce durable long-term value.

Disciplined investors need to look through the narrative noise to identify where perceived disruption may be mispriced as permanent decline. In software specifically, disruption risk is real – but it is also uneven. Businesses with proprietary datasets not freely available to general models, and ’systems of record’ embedded in customer workflows, may be harder to displace than markets assume. In some cases, AI features (e.g., copilots and automation layers) may create incremental revenue streams or improve retention, rather than purely commoditizing the core product.

Conversely, parts of semiconductor manufacturing and enabling infrastructure may benefit from AI in structurally different ways. These areas are demonstrating improving oligopoly-like characteristics, where tighter supply discipline and durable demand can change the long-term earnings power relative to prior cycles. However, this is only investable where industry structure, capital intensity and pricing behaviour support it.

Taken together, these dynamics emphasize the importance of fundamental investing: differentiating between narrative disruption and structural change requires granular analysis of competitive position, industry economics and cash flow durability, not thematic exposure alone.

While we are cautious about drawing direct comparisons with the internet, history can offer useful lessons.

Profound technological advances with broad applications tend to become universal, but this universality is often underestimated early on. At the same time, early leaders do not always prevail – AOL dominated early internet access but then failed to adapt as economics shifted. Some of the companies that are first movers at present may ultimately lag if they are unable to commercialize and to finance growth.

Finally, technology exhibits a persistent bias towards concentration. Search consolidated around Google; social networking around Meta; e-commerce around Amazon. We think AI is unlikely to be different. The mechanism of that concentration often includes aggressive reinvestment, product suite expansion and acquisitions of nascent competitors – which can entrench leaders but also creates a moving target for ’who’ the long-term winners are.

For value investors, the challenge is not to predict the narrative, but to price durability, adaptability and cash generation correctly as that narrative evolves.

¹ Big Tech to Spend USD 650 Billion This Year as AI Race Intensifies, Bloomberg, Feb 6, 2026 and Charted: The USD 448B AI Spending Surge by Big Tech, Visual Capitalist, Apr 20, 2026.
² Moore's Law is the empirical observation that the number of transistors on a microchip doubles roughly every two years, while the cost of computers is halved. Coined by Intel co-founder Gordon Moore in 1965, this principle has served as the guiding roadmap for the semiconductor industry for over half a century.