Nvidia-led gains pulled AI into the center of U.S. markets. Now smart money is thinning positions, buying ETFs, and chasing software plays. Here's a concise playbook.
Illustration by IMF Alpha editorial · Reviewed by Pedro Marini
The headline is simple: AI stocks have rewritten portfolio construction, but that change is less diversified than it looks.
If you sat through the cloud and crypto cycles, this feels familiar. A small group — chipmakers, the big cloud platforms, a few software names anointed by the market — soak up outsized valuations while a long tail scrambles to stay relevant. The result is a particular risk profile: big upside if growth and margins keep up, and sharply amplified downside when they don’t.
Where we stand
Data-center buildouts and specialized AI chips have been the engines behind the recent rally. Passive flows and thematic ETFs have then concentrated retail and institutional cash into a narrow list of names. A second wave is forming too — inference software, enterprise AI services, and vertical models for healthcare and finance are starting to draw selective active bets.
I’ve seen this pattern before. The late 1990s focused attention on internet darlings; the 2010s did the same for cloud. What’s different with AI is the tight coupling of hardware and software. A chip breakthrough can lift a swath of companies at once. But that same cycle — plus foundry constraints — can widen the downside just as fast.
Three real signals to watch — no clickbait
Where smart money is rotating
Some of it is obvious: broader AI ETFs as a hedge against single-stock risk. More telling is the tilt toward firms that monetize models repeatedly rather than through one-time chip shipments. There are also selective hardware plays where supply dynamics and foundry relationships create a more durable edge.
A quick counterpoint: the biggest winners can stay dominant for years. Network effects, model ecosystems, and scale in datacenters have produced long runs before. That just means timing matters. Buying the story too late risks paying for optimism that’s already reflected in the price.
A practical checklist for investors
The practical discipline
Treat AI as a set of linked but distinct sub-themes — chips, cloud infrastructure, model infra, enterprise applications — not a monolith you can buy through a single poster child. That kind of allocation looks conservative during rallies, yes. It also preserves returns when narratives reset.
Pedro Marini
Privacy-preserving datasets, data clean rooms, and marketplaces are reshaping how companies feed models. The winners will be those who pair quality with governance.
Tiny LLMs, aggressive quantization and faster mobile NPUs are shifting intelligence from the cloud to your pocket. What that means for privacy, latency and the next wave of fintech apps.
Synthetic audio and automated social engineering are turning phone calls into the new frontline. Here's what companies and investors must do next.