AI’s Next Chapter Will Need a Bigger Power Cord

Most of us have used an AI tool by now. You type in a question to ChatGPT or Google Gemini and it gives you an answer. Then it waits.

AI is already becoming part of everyday life. Even Kaiden, my three-year-old, has started trying to talk to our Alexa at home, asking it questions and expecting an answer. To him, speaking to a computer already feels normal. Not quite sure how I should feel about that, but I digress!

Agentic or the next phase of AI is different. Think of today’s AI (ChatGPT, Google Gemini) like a student who raises a hand and waits for the teacher to give an assignment. The next phase of AI is more like a helper who helps organize a task, review information, suggest next steps, connect to software tools, or work through a process. It doesn’t just answer one question. It can work across many jobs at once. That sounds exciting but also takes a lot more power.

A normal AI search is a little like turning on the microwave for a minute. More advanced AI workflows are more like running the refrigerator, air conditioner, washer, and dryer all day and all night. One industry report estimates that these AI systems may use 60 to 130 times more energy than a basic chatbot request. That changes the investment story. For the past few years, most of the attention has gone to the “brains” of AI: computer chips, memory, servers, and the companies that build them. Those parts are important but even the smartest computer cannot work without electricity, cooling, cables, buildings, and a strong power grid. A race car may have a powerful engine, but it still needs fuel, tires, a track, and a skilled crew.

This is why we believe the next stage of AI may be shaped not only by valuation debates, but also by real-world bottlenecks in power, cooling, grid equipment, and data-center capacity. A bubble happens when excitement pushes prices far above what the real world can support. A bottleneck happens when real demand is growing, but there are not enough parts, workers, or power to keep up. The move toward agentic AI, sometimes called “AI 2.0,” could create exactly that kind of squeeze. These systems will very likely need to run 24 hours a day. That means data centers will require steady electricity, better cooling, faster connections, more transformers, more high-voltage equipment, and more workers trained to build and maintain it all.

The numbers help show the size of the challenge. Goldman Sachs estimates that U.S. data centers may need 72 gigawatts of new power capacity through 2030. That is roughly equal to the output of 72 large nuclear power plants. Building that much power is not like adding a new app to your phone. It can take years. New power plants need land, permits, equipment, workers, and connections to the grid. Some key parts, such as substations and high-voltage cables, already face long wait times.

There may also be a large gap between where investors are looking and where the real need is growing. According to the report, the combined earnings of chip, manufacturing, memory, and server companies are nearly nine times those of power, component, and data center service companies. In plain English, investors are paying far more attention to the shiny computers than to the systems that keep those computers running. 

That imbalance may continue or it may not. As AI use grows, the companies that build power equipment, improve the electrical grid, provide cooling, connect data centers, and maintain these systems will likely become much more important. They are the picks and shovels of this new gold rush. This thinking is why we have added exposure, where appropriate, to the companies supporting the electric grid and digital infrastructure within our portfolios. Rather than concentrating on one chipmaker or one AI model, the goal is to participate in the buildout underneath the entire system.