A fast-moving shift is under way: money is moving out of flashy AI apps and into chipmakers, data-center suppliers and inference-focused hardware. That trade has winners — and a few traps.
Illustration by IMF Alpha editorial · Reviewed by Pedro Marini
For the last three years most of the noise around AI has lived in apps and cloud-native services. Venture and retail money chased glossy platforms promising to replace white-collar workflows. Lately, though, attention — and capital — has been quietly shifting to the physical plumbing: chips, servers, memory and the data-center infrastructure that actually runs large language models.
Think of it like a retail boom giving way to an electricity grid build-out: storefronts get the headlines, but the utilities collect steady checks.
Retail investors tend to chase the next breakout software story. Betting on infrastructure is less glamorous. But that unglamorousness buys something: recurring economics driven by long lead times and supply discipline. It’s not for everyone — you need patience and a tolerance for capital-cycle risk.
The AI trade is maturing. Momentum will still lift obvious winners, but some of the smarter, lower-profile opportunities sit with the companies that provide compute, memory and systems. For investors comfortable with operational nuance and cyclical swings, the infrastructure side can offer a steadier payoff than chasing the next chat app fad — though it will require a steadier nerve and a longer time horizon.
From data marketplaces to GPU demand, a quiet supply shock in training data is shifting winners in the AI race — and not always in predictable ways.
From neural engines in phones to new edge silicon, on-device AI is reshaping hardware economics. Here’s who benefits, who doesn’t, and how investors should think about it.
Smartphones are running LLMs and fraud detection locally. That changes privacy, cost structures, and who controls financial data — fast, but messy.