“The real stuff here is all physical”, says Thomas Thygesen, Head of SEB’s equity strategy team. “In our view proper productivity gains come when AI is so commoditised that you can put the ghost in the machine and then the machine starts doing stuff we couldn't imagine before.”
Investment by large technology companies in data centres has surged over recent years and consensus expectations point to another doubling over the next two years. That spending is supporting demand well beyond semiconductors, extending to areas including electricity grids and turbines.
However, financing is becoming more challenging. After previously funding investment from their own cash flows, hyperscalers are increasingly turning to debt as free cash flow comes under pressure. At the same time, markets have begun questioning whether the companies at the end of the AI investment chain can ultimately generate enough revenue to support the spending.
Costs are falling
One challenge is the rapidly falling cost of competing AI models. US frontier models remain among the strongest, according to SEB’s analysis, but some alternatives can perform similar tasks at substantially lower cost. Users are consequently becoming more selective about when they need the most advanced models.
“People are actually starting to shift towards cheaper versions”, says Mads Bossen, equity strategist at SEB. “You don't need the most intelligent models to do the job that you want done.”
SEB nevertheless sees little evidence that the investment cycle is about to stop immediately. The strategists expect capital expenditure to continue for now, while the next few quarters should provide clearer evidence about the development of demand and profitability.
Longer term, they see the AI investment theme becoming increasingly physical. As AI becomes cheaper and moves into machinery and industrial applications, sectors such as industrials, materials and transportation could benefit. That could favour European and Nordic markets, where these sectors have a larger presence than in the US.