The numbers come first. Hyperscalers — Google, Meta, Microsoft, and Amazon — are spending somewhere between $750 billion and $800 billion a year on AI infrastructure, according to John Mowrey, chief investment officer at NFJ Investment Group. Some forecasts push that figure to $1 trillion, which Mowrey notes would represent 2.5 to 3% of U.S. GDP. 'That is just extraordinary for a capital market,' he said.
Data centers sit at the center of that spending wave, and Fortune has identified four distinct entry points for investors: semiconductor chips, real estate, energy, and cooling. Analysts weighed in on which names stand out.
Chips: The Brain of the Buildout
Demand for specialized processors running AI workloads has outpaced the industry's ability to manufacture them, making semiconductors one of the most acute bottlenecks in the entire buildout, according to Craig Ellis, research director and senior semiconductor analyst at B. Riley Securities. 'We're at a point where undersupply is so severe that there needs to be a multi-year period of unusually strong capex growth in front of us,' Ellis said, adding that the growth 'has the potential to continue to lift expectations for revenues and earnings.'
Ellis's recommendation: look past Nvidia, AMD, and TSMC toward the companies that supply the equipment used to make chips in the first place.
Applied Materials (AMAT), the largest semiconductor equipment company in the world, is his lead example. Its Semiconductor Systems division accounts for roughly 73% of company revenue, per its most recent annual filing. Because nearly every advanced chip and display passes through Applied's tools at some point, its exposure is spread across the entire chipmaking ecosystem rather than concentrated in one chip type.
Lam Research (LRCX) is Ellis's second pick. The company focuses on equipment for memory and storage chips — a narrower product range than Applied Materials, but one Ellis sees as especially well positioned for a surge in new capacity investment over the next two years. B. Riley Securities has raised its earnings estimates for Lam by 25% to reflect that outlook.
Marvell Technology (MRVL) rounds out the semiconductor slate, with particular strength in networking — the infrastructure that lets thousands of chips inside a data center communicate fast enough to function as a single machine. Marvell carries roughly $195 billion in market value, according to FactSet, putting it in the same weight class as Nvidia and AMD. Nvidia has made a direct investment in Marvell as part of a partnership on next-generation networking technology. The bulk of Marvell's current business flows from Amazon Web Services, a concentration Ellis frames as opportunity: if Marvell can diversify its customer base and scale network products, he believes it could generate meaningfully more profit than it does today.
Risk Is Real, but Already Priced In
Chip stocks are volatile by nature and sensitive to geopolitical friction or shifts in hyperscaler capital allocation. The SOX index, which tracks AI-linked chip stocks, fell 26% on concerns that the spending wave would not translate into returns. Ellis argues that drop already reflected most of the downside risk, making current valuations a reasonable entry point for long-term investors.
CEO Times Take
This is what free-market capital formation looks like at scale: private companies, not government programs, are committing resources equivalent to a meaningful slice of GDP to build the infrastructure of the next economic era. The picks-and-shovels framing matters here — equipment makers like Applied Materials and Lam Research earn revenue whether Nvidia wins the chip wars or a challenger does. That is durable exposure, not a momentum bet.
The risk worth watching is not the technology. It is the regulatory and geopolitical environment surrounding semiconductor manufacturing. Capital rewards clear rules, and any Washington intervention that disrupts equipment supply chains or imposes new export controls could reprice these names faster than any earnings miss. The buildout is real. The question is whether policymakers will get out of its way.
