We’ve all seen this debate play out before. Amazon was seen as a threat to Walmart, Salesforce would replace Oracle, and PayPal would overtake Visa. In each case, the challengers grew into meaningful businesses, yet the incumbents adapted and survived.
Today, artificial intelligence (AI) is often portrayed as a force capable of replacing a wide swathe of knowledge-based jobs and companies.
Sceptics, however, point to record high valuations – including $750bn for OpenAI – alongside an estimated $650bn in hyperscaler capital expenditure in 2026. They also cite persistent shortcomings, such as model hallucinations, as fuel for speculation that we may be approaching bubble territory.
The more plausible outcome may lie between these extremes. In our view, this ‘middle ground’ looks like a steadier roll-out of AI as a commercially useful technology where winners are defined by execution, profit margin and demand – rather than market speculation.
As AI adoption broadens, the pace of change is likely to accelerate as costs fall. History shows that some markets have consolidated around companies able to shift technical advantage into sustainable returns. The US auto industry provides a compelling illustration, shrinking from 88 manufacturers in 1921 to just 44 by 1927.
The contradiction at the heart of the boom
There is little doubt that we are living through transformative times. Earlier this year, a man without a medical background leveraged AI tools to develop a cancer vaccine for his dog, illustrating how these technologies are already enabling us to do more with less.
Yet the prevailing AI boom rests on a contradiction. It assumes hyperscalers will generate sustained growth by selling infrastructure and tools to large established companies, while also assuming that those same customers will face pressures on revenue, pricing power and headcount as AI-enabled efficiencies take hold.
We believe that over time both assumptions cannot hold true. If AI erodes profitability in sectors such as banking, the capacity of those companies to spend more on AI services will likely weaken.
You don’t need a Ferrari to deliver a pizza
Historical precedent suggests that technological adoption rarely converges on a single dominant platform. In markets such as cloud computing, clients have typically adopted a mix of providers.
A similar pattern could emerge with AI. Major companies are likely to deploy a range of off-the-shelf private models, open-source alternatives and customised programmes built on proprietary data, depending on the task at hand.
Just as a Ferrari is unnecessary for local deliveries, a ‘good enough’ AI model will often prove sufficient for many tasks. This optionality opens the door for a range of providers to embed AI into existing products and workflows, rather than ceding the field to a single dominant platform.
The prevailing narrative around consolidation may therefore be overstated. In practice, users tend to favour flexibility – switching between different tools for specific tasks and adopting or abandoning them as their needs evolve.
A similar dynamic is evident in market share. Rather than quickly coalescing around a handful of winners, the industry remains fragmented with scope for multiple players to scale up and capture share over time.
Adoption is likely to be constrained by inertia, as integrating new systems and processes is typically slow and costly. Enterprise resource planning software offers a useful precedent, where a long tail of smaller providers has ensured that even the largest incumbents – Oracle and SAP – account for only about 10% market share each.
Practical barriers to the boom
The boom argument also assumes that the regulatory response will be limited. Yet in economies such as the US, where growth is driven by consumer spending, large-scale disruption carries social and political consequences. If AI weakens employment or household incomes, regulatory reforms will likely follow.
Constraints extend beyond regulations. The energy demands of large-scale data centres, rising memory costs and a shortage of specialist talent all pose challenges.
Data privacy concerns, cashflow limitations of AI providers and the deployment of these tools across entire organisations could also slow adoption.
None of these roadblocks are insurmountable in isolation, but taken together, they could point to a more measured pace of adoption than the boom narrative suggests.
The middle ground
Looking beyond boom-or-doom scenarios, our AI approach maintains a bottom-up focus on companies that meet our quality-growth criteria. By contrast, semiconductor companies are more cyclical, with an uncertain outlook for supply and demand.
Sceptics tend to think of companies as standing still while technology evolves. In our experience, the best companies tend to adapt by integrating innovations, passing productivity gains back to their clients and in effect disrupting themselves.
Their advantages lie in scale, data, established customer relationships, regulatory approvals and financial resources, all of which support durable, mutually beneficial solutions.
For established businesses with diversified revenue streams – in e-commerce and online advertising – AI exposure comes with a natural cushion visibility over future spend and the flexibility to absorb or redirect investment as the technology matures. These are not companies waiting to be disrupted. They are the ones doing the disrupting.
The middle ground may lack the drama of boom or bust narratives. However, if history is any guide, it is where the most resilient businesses – and the most sustainable returns – are ultimately found.
Justin Streeter is a US equities analyst and portfolio manager at Comgest. The views expressed above should not be taken as investment advice.