Over the past year, the epochal innovation of artificial intelligence has sparked dramatic swings in market sentiment. Initially, there were fears of overinvestment and an infrastructure bubble. More recently, there have been fears that AI will render entire industries obsolete. Neither narrative can be true simultaneously.
We cannot be massively overspending on AI capacity with insufficient demand, while also standing on the brink of widespread corporate extinction driven by the same technology. This is not a bubble bursting but rather a market struggling to adapt to a structural transformation.
Last year, the Chinese DeepSeek AI model sparked a sharp selloff in AI-related stocks, including Nvidia, the world’s leading semiconductor company. The advances that it made were interpreted as a threat to compute demand.
Yet this reaction misunderstood a common pattern in technological progress, in which efficiency gains often expand markets rather than shrink them. Compute demand did not collapse. Instead, it surged, and these AI-related stocks rebounded and flourished.
We see a similar dynamic today, where advances in Claude, Anthropic’s generative AI tool, have led to broad declines in share prices across software and other sectors. The selling has been largely indiscriminate. The fear is now not underutilised capacity, but total obsolescence.
Hundreds of billions of dollars’ worth of market capitalisation has been wiped out on the assumption that AI can instantly replicate many established industries. Volatility like this is uncomfortable, but it can also create opportunities for long-term investors.
We do not think that hyperscalers are allocating capital recklessly. We acknowledge that the scale of investment is enormous and execution will inevitably be imperfect.
Delays, cost overruns or unforeseen engineering challenges are entirely likely. We have seen similar reactions before. Nvidia experienced well-publicised design and thermal issues with early Blackwell deployments. These are normal challenges in cutting-edge chip design.
Yet commentary from sources that seemed unfamiliar with how semiconductor development works lit up our screens and amplified these issues beyond what the facts warranted.
Building advanced data centres on time and on budget is a complex undertaking. Delays, cost overruns, or unforeseen engineering challenges are entirely likely.
Crucially, hyperscalers base capacity expansion on visible demand and backlog. These companies have disciplined capital allocation and make demand-led, not speculative, investments. This contrasts with building capacity ahead of demand, hoping customers will follow, as happened during the late 1990s tech bubble.
The second narrative currently dominating the market, which we think is equally overstated, is that AI will swiftly render large parts of industries obsolete. AI disruption will not be uniform. Some business models will weaken, while others remain resilient or strengthen.
Markets often shift too quickly between bear narratives. A similar overreaction occurred with Amazon and e-commerce, which many expected to devastate retail. While some retailers struggled, others, such as Costco and Walmart, adapted and thrived.
More recently, the rise of GLP-1 weight-loss drugs triggered a sharp sell-off across food and beverage stocks amid fears of reduced consumption. Yet we see companies like McDonald’s continuing to perform well, with shares near all-time highs.
Not all software firms will be impacted by AI as heavily as others. Those that offer deeply integrated systems of record, such as SAP’s ERP platform or Salesforce’s CRM, act as the official source of enterprise data. They are deeply embedded in operations, closely linked to compliance and regulation, and extremely costly and risky to replace.
Enterprise migrations from on-premises to cloud environments have taken years and many are still incomplete. The underlying data is sacrosanct. We believe that firms will not allow AI to interact with it freely without strong permissions, security and governance.
In contrast, workflow-oriented software and business services that rely on common data rather than proprietary information are more susceptible to AI competition, as AI can more easily replicate or enhance these functions. The trend is already seen in industries such as credit bureaus, insurance brokers and wealth management.
We have been positive about AI’s long-term prospects for more than three years and continue to believe it represents the next industrial paradigm. But technological transitions are rarely linear. They create winners, losers and mispricing as they progress.
We think it is these periods of narrative-driven selling, volatility and fear that active fundamental investors can add value. When markets treat entire sectors as either doomed or invincible, careful analysis and stock selection matter most. This is not a time for retreat but for targeted investment in well-researched opportunities.
Christopher Rossbach is CIO at J. Stern & Co. The views expressed above should not be taken as investment advice.