The next phase of the AI cycle will be defined by the “three Ds”: deployment, debt and demand, according to Schroders investment director of global equities Ben Arnold.
Arnold says the past year has seen enormous sums being poured into technology as major tech players race to build the infrastructure needed to support artificial intelligence, and investors are questioning whether that will translate into real revenue and justify the scale of spending.
“The first question is about deployment of capital. If you’ve owned stock in any AI business over the past two or three years, it’s likely you’ve done very well. But the level of capex upgrades we have seen during reporting season has been huge,” Arnold says.
“Technology giants have committed hundreds of billions of dollars to data centres, chips and computing power to build the backbone of the AI economy. But as spending rises, so too does the pressure to generate returns.
“Now, the market is starting to take a different view on how sustainable these projections really are, and what we saw toward the back end of 2025 was different companies being rewarded or punished in very different ways.
“We’ve seen a sell off across several major tech companies in recent months. As a result, investors are being punished, software companies are being punished, and the hyperscalers are being looked at very differently to how they were only twelve months ago.”
Arnold says debt is the second pressure point emerging in the sector.
“The combination of increased levels of debt and an inability to catch up on revenue expectations can become a real problem,” Arnold says.
“Share prices went up when the story around the growth potential of AI was positive. Now, as valuations have soared, we’re starting to see that optimism temper.”
The third and perhaps most important factor impacting investment is demand.
“As innovation occurs at a rapid rate, demand is becoming harder to track, but it remains the most influential factor. We spend a lot of time going through the AI tech stack asking whether demand is actually justifying the capex we see today,” Arnold says.
“Companies building AI infrastructure argue the spending is necessary because they are seeing the demand. Markets, however, are looking at these numbers and becoming more sceptical.”
Arnold says this has been most clear in software companies. The recent sell-off reflects a growing concern that some software companies may struggle to defend their business models in a world dominated by large language models.
However, Arnold believes some areas of the market will prove more resilient than others. Businesses that operate in environments where accuracy is critical, where switching systems is difficult, or where regulation creates barriers to entry are likely to maintain stronger competitive positions.
“If a company’s value is based purely on a publicly available data set, then a large language model is going to commoditise that,” Arnold says.
“Areas where there isn’t a tolerance for errors will likely be more protected.
“Similarly, companies that control proprietary data or have high switching costs embedded in their products may be better placed to defend their margins.”
As the AI investment cycle continues to accelerate, Arnold says investors will increasingly focus on these fundamentals rather than broad narratives about technological disruption.
“The market is moving beyond the excitement of the disruption theme, and for stock pickers it means there are opportunities to outperform. But it also comes with a higher level of risk,” Arnold says.
“What if AI turns out to be less ‘bubble versus boom’ and more a stress test for who can actually turn US$660 billion of capital expenditure into revenue?
“What matters now is whether the spending we’re seeing today can actually turn into sustainable revenue tomorrow.”