For the past three years, AI investing meant one thing: Nvidia. Buy chips, ride the wave. But that phase is over — or at least, the easy money is.
The AI market is now transitioning into a new phase, and the investment opportunity has quietly shifted from semiconductors to something far less glamorous but potentially just as profitable: infrastructure.
Phase 1 vs. Phase 2: What Changed
Understanding where to invest starts with understanding the two phases of AI buildout:
- Phase 1 (2020–2024): Training Infrastructure — A handful of hyperscalers (Microsoft, Google, Amazon, Meta) spent hundreds of billions building massive AI training clusters. This created the GPU supercycle. Nvidia won this phase decisively.
- Phase 2 (2025+): Deployment Infrastructure — Now, thousands of enterprises are deploying AI locally. Smaller inference clusters. Distributed computing. And the bottleneck has shifted from compute chips to the infrastructure required to run them: power delivery, cooling systems, networking, and deployment services.
The companies that build, cool, power, and connect AI infrastructure are the picks-and-shovels play of Phase 2.
Why the Bottleneck Shifted
AI training clusters are largely built. What remains is deploying AI everywhere — in enterprise data centers, government facilities, and regional hubs. This requires:
- Electrical switchgear and substations to handle power demand
- Liquid cooling systems to manage heat from dense GPU racks
- High-speed optical networking to connect AI clusters
- Construction and HVAC services to build and retrofit the facilities
These are the categories where demand is growing faster than supply — and where investors have not yet fully priced in the opportunity.
The Investment Categories to Know
The AI infrastructure stack breaks into four clear investment categories:
- Electrical Infrastructure: Powell Industries, Eaton, Schneider Electric
- Cooling Systems: Vertiv, Modine Manufacturing
- Construction & Deployment: Comfort Systems USA, Quanta Services
- Optical Networking: Lumentum, Coherent Corp
Each plays a different role in making enterprise AI deployment possible. The next four articles in this series break down each category in detail.
Why Now?
Enterprise hardware earnings are showing early signals: AI infrastructure backlogs are growing, modular data center deployments are accelerating, and utility companies are flagging unprecedented power demand from data centers. These are leading indicators — they’re showing up in earnings calls before they show up in stock prices.
The Risk to Understand
This thesis is not without risk. AI inference efficiency is improving rapidly — models that required expensive GPUs two years ago now run on commodity hardware. If that trend accelerates, infrastructure demand could plateau sooner than expected. Timing matters.
That said, even conservative enterprise AI adoption projections imply years of infrastructure buildout ahead. The question isn’t if — it’s which companies benefit most and in what sequence.
How to Invest
The infrastructure companies in this thesis are accessible on standard brokerage platforms. You don’t need a special account or derivatives exposure — these are publicly traded US equities. Open a brokerage account, research the individual names, and size positions according to your risk tolerance.
Affiliate disclosure: This article contains affiliate links. If you open a brokerage account through our links, we may earn a commission at no cost to you.