Tech-N-AI Talks logo Tech-N-AI Talks

AI Data Center Spending: Smart Investment Strategies for 2025

Learn how to invest in the $31.6 trillion AI infrastructure boom. We break down data center investment opportunities and risks with specific picks.

$31.6 Trillion Data Center Boom: How to Invest in AI Infrastructure Wisely — illustrative featured image
The parking lot behind the Santa Clara data center campus looks like a scene from a dystopian movie. Twelve diesel generators, each the size of a shipping container, roar in unison as a test crew runs them through their paces. The noise is deafening, the heat shimmer is visible from a hundred yards, and the whole operation is preparing for one thing: the next generation of AI models that will need more power than most small cities consume. That scene is playing out across the globe, and the numbers are staggering. Bloomberg Intelligence projects data center spending will hit $31.6 trillion by 2050, driven almost entirely by AI workloads. That is not a typo. Trillion with a T. For context, that is roughly the entire GDP of the United States, spent on concrete, copper, and cooling systems over the next quarter century. The question is not whether this boom is real. It is. The question is how you, as an investor, navigate a gold rush where the picks and shovels are measured in megawatts and teraflops. ## The Physics of the AI Buildout Here is what most people miss about AI infrastructure. The software gets the headlines, but the bottleneck is electricity and heat. Every time OpenAI or Google or Meta releases a frontier model, the training run requires tens of thousands of GPUs running at full tilt for months. Those GPUs generate heat like industrial ovens, and they draw power like small towns. The math is brutal. A single Nvidia H100 GPU can draw up to 700 watts under load. A modern data center with 100,000 of those GPUs needs roughly 70 megawatts just for the chips, before you account for cooling, networking, and lighting. Double that number for the total facility draw. That is why hyperscalers are now talking about gigawatt-scale campuses. A gigawatt is a nuclear reactor's worth of output. This creates a peculiar investment dynamic. The traditional technology sector rewards software margins and recurring revenue. Data centers are the opposite. They are capital-intensive, slow to build, and tied to long-term power contracts. But they also have something software rarely offers: contractual revenue visibility. When a hyperscaler signs a 15-year lease for 100 megawatts of capacity, that is a guaranteed cash flow stream that banks love to finance. ## Where the Money Actually Flows The $31.6 trillion figure is so large that it becomes abstract. Let us break it down into the four buckets where capital is actually deployed: | Investment Bucket | Share of Spending | Key Drivers | |---|---|---| | Physical construction | 35-40% | Shell buildings, raised floors, security, fire suppression | | Power and cooling | 25-30% | Transformers, switchgear, liquid cooling loops, chillers | | IT hardware | 25-30% | GPUs, CPUs, storage arrays, networking switches | | Land and entitlements | 5-10% | Zoning approvals, grid interconnection rights, water access | Notice what is missing from that list: software. The AI infrastructure boom is primarily a hardware and construction phenomenon. That has profound implications for how you build a portfolio around it. ### The Obvious Plays and Their Problems The purest expression of this trend is Nvidia. The company effectively prints money right now, with data center revenue up more than 200% year over year. But here is the uncomfortable truth: Nvidia is priced for perfection. The stock trades at a premium that assumes the AI buildout continues uninterrupted for years. Any hiccup in the supply chain, any shift in chip architecture, any slowdown in hyperscaler capex, and the multiple compresses violently. The semiconductor ecosystem extends beyond Nvidia. Taiwan Semiconductor Manufacturing Company (TSMC) fabricates the advanced chips that make this whole thing possible. ASML makes the lithography machines that TSMC uses. Applied Materials and Lam Research make the deposition and etching tools. These are all critical, but they are also cyclical. When the AI trade wobbles, they wobble harder. ### The Boring Middle Layer Here is where it gets interesting. The companies that actually own and operate data centers are less glamorous but arguably more predictable. Digital Realty, Equinix, and CoreWeave represent different flavors of this exposure. Digital Realty and Equinix operate colocation facilities with diversified tenant bases. CoreWeave is a pure-play AI cloud provider that has signed massive contracts with the big model labs. The risk profile differs significantly. Equinix is a real estate investment trust (REIT) with a long track record of dividend growth. CoreWeave is a growth story with debt leverage that would make a private equity firm blush. Both can work, but they are completely different bets. The real sweet spot might be in the less obvious infrastructure providers. Vertiv Holdings makes the liquid cooling systems that are becoming mandatory for high-density AI racks. Eaton Corporation and Schneider Electric manufacture the power distribution equipment that every facility needs. These companies are not reliant on any single chip architecture or model vendor. They sell to everyone. ## Our Take: What We Recommend We have been tracking this space for three years, and we have learned to be skeptical of the pure hype plays. Here is what we actually like right now. **For core exposure, consider a diversified basket.** The Global X Data Center REITs and Digital Infrastructure ETF (VPN) gives you exposure to the real estate side without single-stock risk. It is not exciting, but it has held up better than most tech funds during drawdowns. **For growth, look at Vertiv (VRT).** The company has quietly become the dominant player in thermal management for AI data centers. Its liquid cooling solutions are being specified into virtually every new hyperscale buildout. The stock has run hard, but the order book remains robust. We think the earnings revisions have further to go. **For value, consider Eaton (ETN).** The power management giant trades at a reasonable multiple relative to its growth trajectory. Every gigawatt-scale data center needs its switchgear, busways, and uninterruptible power supplies. Eaton is not an AI company, but it is an AI infrastructure company, and that distinction matters. **One thing we avoid: small-cap penny stocks claiming to be "AI data center plays."** There are dozens of shell companies rebranding themselves as AI infrastructure providers. They will burn you. Stick with companies that have actual revenue, actual facilities, and audited financials. ### The Power Play Nobody Is Talking About Natural gas is the hidden variable in this equation. Renewable energy alone cannot power gigawatt-scale data centers with 24/7 availability requirements. The grid is not built for that kind of intermittent load. As a result, we are seeing a resurgence of natural gas turbine installations, often at the data center site itself. Cummins and Caterpillar both manufacture the large reciprocating engines used for backup and peaking power. GE Vernova builds the gas turbines that are increasingly being paired with data center campuses. Solar and wind will play a role, but natural gas is the bridge fuel that makes the AI boom physically possible. This is an underappreciated angle that most tech investors completely miss. ## The Risks That Could Derail Everything We would be remiss if we did not flag the bear case. Three risks keep us up at night. First, the power constraint is real. Getting grid interconnection approvals can take five years in some jurisdictions. If the buildout cannot physically proceed at the pace the spending projections assume, those trillion-dollar figures slip to the right. Second, the return on investment for AI itself remains unproven. The hyperscalers are spending billions on infrastructure while their AI revenue streams are still nascent. If the models do not generate sufficient monetization, the capex cycle will eventually slow. Third, there is a concentration risk in the supply chain. The entire AI hardware stack depends on TSMC's advanced packaging capacity and Nvidia's design leadership. A single disruption at either node creates cascading delays across the entire ecosystem. ## FAQ **How do I invest in AI infrastructure without buying individual stocks?** Index funds and ETFs are the safest route. Look for funds that track digital infrastructure, data center REITs, or the broader semiconductor supply chain. The key is diversification across the four spending buckets we outlined above. Avoid funds that are too heavily weighted toward any single chipmaker or cloud provider. **What is the difference between a data center REIT and an AI cloud provider?** A REIT like Digital Realty owns physical properties and leases space to multiple tenants. The revenue is stable and contractual, but the growth is modest. An AI cloud provider like CoreWeave builds specialized infrastructure for machine learning workloads and typically serves a smaller number of large customers. The growth potential is higher, but so is the risk. **Is it too late to invest in AI infrastructure?** The spending cycle is early, not late. We are perhaps 15-20% of the way through the buildout that the Bloomberg projections describe. The first wave of investment was concentrated in the hyperscalers and Nvidia. The second wave is spreading to power equipment, cooling systems, and construction. That second wave is where the opportunity lies for new capital.

Frequently asked questions

How do I invest in AI infrastructure without buying individual stocks?

Index funds and ETFs are the safest route. Look for funds that track digital infrastructure, data center REITs, or the broader semiconductor supply chain. The key is diversification across the four spending buckets we outlined above. Avoid funds that are too heavily weighted toward any single chipmaker or cloud provider.

What is the difference between a data center REIT and an AI cloud provider?

A REIT like Digital Realty owns physical properties and leases space to multiple tenants. The revenue is stable and contractual, but the growth is modest. An AI cloud provider like CoreWeave builds specialized infrastructure for machine learning workloads and typically serves a smaller number of large customers. The growth potential is higher, but so is the risk.

Is it too late to invest in AI infrastructure?

The spending cycle is early, not late. We are perhaps 15-20% of the way through the buildout that the Bloomberg projections describe. The first wave of investment was concentrated in the hyperscalers and Nvidia. The second wave is spreading to power equipment, cooling systems, and construction. That second wave is where the opportunity lies for new capital.

Related articles