If history is any guide, the artificial intelligence boom will not be those who build the technology, but those who deploy it. That is the view of Kevin Khang, head of global economic research at Vanguard. He argues that today’s market narrative risks mistaking the early winners for the ultimate ones.
After studying 150 years of technological cycles, Vanguard’s economists have found a striking pattern: they tend to unfold in two phases. “In the beginning, it’s the producers of the technology that benefit most. In the second half, it’s the users,” said Khang. He points to past breakthroughs such as the steam engine, electricity and the internet, where early gains were concentrated in hardware and infrastructure before spreading across the wider economy.
Vanguard believes the current AI boom is still in that first phase. Investment continues to flow into semiconductors, data centres and power infrastructure, underpinning earnings growth in a narrow group of companies. The tech-heavy Nasdaq Composite index reached new all-time highs last week, suggesting markets remain firmly in the first phase for now.
A more decisive turning point would come when productivity gains become visible across the broader economy. Sectors such as healthcare and business services could offer early evidence, particularly if companies begin to report measurable improvements in efficiency.
“If you start to see doctors and nurses able to handle more patients because of better workflows enabled by AI, that’s when you know the second phase is really taking hold,” Khang told Investment Officer. “Between three and five years, it’s quite possible that half of the top companies in the world are not tech.”
So far, however, that transition remains incomplete. While adoption is accelerating, the impact on productivity has yet to show up convincingly in aggregate data. Recent Federal Reserve research highlights the uneven nature of AI uptake. Around 18 per cent of US companies had adopted AI by the end of 2025, although roughly 41 per cent of workers report using generative AI in their jobs. Adoption is concentrated in higher-value sectors such as professional and financial services, suggesting the technology is primarily augmenting cognitive tasks rather than transforming the wider economy.
For investors, the timing of AI’s boost to economic growth is critical. Over the past decade, strong and consistent earnings growth from a handful of US technology companies has justified elevated valuations and driven market returns
Earnings forecasts for 2026 still show technology firms delivering the fastest growth. But recent revisions suggest the momentum is beginning to shift. Upgrades have been strong in sectors such as energy and materials, pointing to a broader set of beneficiaries from the AI investment cycle.
If earnings growth broadens further, the valuation premium enjoyed by large technology groups may come under pressure, particularly as they ramp up spending to stay competitive in AI. UBS estimates that hyperscalers’ capital expenditure this year will absorb nearly all of their operating cash flow, compared with a 10-year average of about 40 per cent. “That’s going to start putting a dent on their cash flows and earnings profile,” Khang said.
The market, he said, seems to be “really moving away from about a 10-year long environment, where all the earnings growth was being driven by several really big businesses.”
That could have implications for portfolio construction. Heavy exposure to US mega-cap technology stocks has left many investors concentrated in a single theme.
A more diversified earnings backdrop would favour a broader allocation across sectors and regions. Vanguard points to non-US markets as a potential hedge, given their lower exposure to technology and greater sensitivity to industrial and commodity cycles tied to AI investment.
Khang cautioned that the long-term outlook for technology remains strong. The largest companies are investing heavily to defend their positions, and some will continue to lead.
But the dominance of a small group of firms may not persist indefinitely. “There’s still a lot of uncertainty around how this plays out,” he said. “But if history is any guide, the biggest gains won’t stay concentrated where they are today.”
Ultimately, the trajectory will depend on how quickly AI is integrated into business processes. A gradual rollout could support steady productivity gains, while a faster transition could prove more disruptive, particularly for labour markets.
“The technology is already proving more powerful than many expected,” Khang said. “The question now is how quickly the rest of the economy can put it to work.”