The artificial intelligence frenzy is not just about chatbots and image generators—it's fueling a massive infrastructure buildout. Major cloud providers, including Google, Meta, Microsoft, and Amazon, are locked in a spending race that shows no signs of slowing, even as stock markets wobble and fears of an AI bubble persist. The scale is staggering. John Mowrey, chief investment officer at NFJ Investment Group, notes that hyperscalers could spend between $750 billion and $800 billion annually, with some forecasts reaching a trillion dollars—roughly 2.5% to 3% of U.S. GDP. A significant portion of that money will flow into data centers, offering investors a pick-and-shovel play on the AI wave.

Fortune has identified four distinct entry points into the data center economy: semiconductors, real estate, energy, and cooling. Each category comes with its own risks and rewards, and experts weighed in on standout stocks.

**Semiconductors: The Brain of the Operation**

No data center functions without chips, and demand for processors that run AI workloads has far outstripped manufacturing capacity. Craig Ellis, senior semiconductor analyst at B. Riley Securities, calls this the most acute bottleneck in the buildout. For investors, undersupply is a double-edged sword—it pressures prices but also signals sustained capital expenditure growth. Ellis advises shifting focus from chip giants like Nvidia, AMD, and TSMC to the companies that supply the equipment to make those chips. Applied Materials, the world's largest semiconductor equipment company, derives about 73% of its revenue from its Semiconductor Systems division, which sells tools for computing, logic, and memory chips. Its breadth means nearly every advanced chip passes through its machines, spreading exposure across the ecosystem. Lam Research, a narrower play on memory and storage chip equipment, is also well-positioned, with B. Riley Securities raising earnings estimates by 25% on expected capacity investment. Marvell Technology stands out in networking—the plumbing that links thousands of chips inside a data center. Though less famous than Nvidia, Marvell commands a market value of roughly $195 billion, with Amazon Web Services as its largest customer. Ellis sees that concentration as an opportunity: if Marvell diversifies its client base, profit margins could expand significantly.

**Real Estate: The Body That Houses the Brain**

Chips need physical space, and real estate investment trusts (REITs) provide the climate-controlled, high-connectivity environments required for round-the-clock operations. Patrick Wilson, a portfolio manager at CenterSquare Investment Management, highlights that the industry is shifting from training AI models to inference—running models for daily queries. Inference works best in urban data centers close to users, where established REITs already control scarce, carrier-dense land. Equinix, a major landlord for internet and cloud computing, hosts servers for thousands of companies, including AWS and Google Cloud. Digital Realty, with over two decades of experience, benefits from supply that still lags demand, allowing rent increases. Wilson notes that these REITs are tax-efficient, as they must pay out 90% of taxable income as dividends to avoid corporate taxes. However, risks include rising interest rates, which raise borrowing costs, and potential competition from new entrants. Yet Wilson argues that REITs are safer than heavily indebted AI infrastructure firms that could struggle if short-term customers vanish.

**Energy: Powering the Boom**

AI data centers consume enormous amounts of electricity, reshaping the energy sector. Andrew Bischof, utilities analyst at Morningstar, explains that U.S. electricity demand was stagnant for decades, growing only 0% to 0.5% annually. Now, AI has supercharged forecasts, turning utilities from low-growth yield stocks into growth vehicles. American Electric Power, a utility serving 5 million customers across 11 states, plans to invest $78 billion from 2026 to 2030 to meet data center demand. Morningstar expects 9% annual earnings growth from that spending. Bischof also recommends DTE Energy, Alliant Energy, and Evergy, which operate in regions with strong regulatory alignment. For a contrarian play, David Trainer of New Constructs suggests fossil fuel refiners like Valero and HF Sinclair. He argues that green energy cannot yet deliver the high-intensity power needed, leaving traditional energy stocks undervalued—the market treats them as if profits will permanently decline 30% to 40%.

**Cooling: Keeping the Factories from Melting**

Data centers generate intense heat, and standard air conditioning cannot handle it. Nick Lieb, industrials analyst at Morningstar, sees cooling and power equipment as a stable AI trade. Vertiv, a longtime provider of precision cooling, owns the original Liebert brand, which invented computer room air handling units in the 1960s. Over 80% of Vertiv's revenue comes from data centers, making it highly concentrated but also deeply embedded. For broader exposure, Lieb points to Eaton, where only a quarter of sales tie directly to data centers; the rest comes from commercial buildings and the electrical grid. Eaton produces transformers and switchgear needed for both new data centers and upgrades to an aging U.S. grid.

**Risks and the Long View**

Despite the optimism, the sheer scale of spending raises concerns. Lieb notes that hyperscalers' capital expenditures are starting to exceed the cash they generate, forcing them into debt and equity markets. Trainer calls the market a crowded trade, warning that many firms cannot sustain their spending rates. Yet Mowrey remains bullish: AI adoption is still in its early innings, with enterprise applications in healthcare, manufacturing, and finance taking years to unfold. The data center buildout, he argues, is laying the backbone for a technology shift that will touch every corner of the economy.