AI infrastructure benefits first because every model and application needs compute, memory, networking, power and data centres. Demand is tangible and spending is already enormous.

But infrastructure is capital-intensive and eventually risks overcapacity and commoditisation. Software has the opposite challenge: competition is intense today, yet successful applications can scale with much lower marginal costs and become deeply embedded in business workflows.

My preference would be infrastructure while AI capex remains strong, then gradually shift attention towards software companies that demonstrate genuine pricing power, recurring revenue and measurable productivity gains.

The next big winners may not build AI. They may be the companies that figure out how to monetise it.