The largest data center operators are beginning to treat power as part of the infrastructure they build, finance, and control, not simply a utility service they purchase.
That is a meaningful shift.
Rather than relying exclusively on the traditional utility path, they are taking greater control over how power is sourced and managed today while making long term commitments to support demand well into the future.
What interests us is what that shift sets in motion.
It broadens the AI investment story beyond computing and into the infrastructure that makes computing possible at scale. The opportunity is not confined to producing more electricity. It also sits in the systems that move and manage that power, and in the technologies that help data centers use it more intelligently.
One of the clearest responses is the move toward more onsite generation.
Natural gas has emerged as a practical near term option because it can provide firm, dispatchable power and give developers more control over when capacity becomes available. Fuel cells are also gaining attention as a modular source of onsite generation, particularly where developers want more control over how quickly new power can be added.
Microgrids take that idea further. Rather than treating the data center and its power supply as separate projects, the two can increasingly be designed as one coordinated system. That gives operators greater control over reliability and how different sources of power work together.
The investment implications extend beyond the generation source itself. They also reach into the infrastructure and services required to deliver, manage, and maintain that power over time.
As more capital moves into these projects, we think it is important to look beyond the most visible technologies and evaluate the businesses becoming increasingly important to making the buildout possible. We are particularly interested in areas where capacity is difficult to expand and practical alternatives are limited.
The traditional grid remains an important part of the picture as well.
A significant amount of future data center capacity will still depend on the broader electrical system. Even facilities producing some of their own power may continue to rely on the grid for additional capacity, balancing, or redundancy.
That keeps transmission and electrical infrastructure firmly tied to the AI buildout. But the investment question is more specific than simply assuming more spending will create opportunity. We are interested in where persistent constraints create strategic importance and where the market may be underestimating the value of infrastructure that is difficult to work around.
There is another side to the equation.
The industry is also becoming more sophisticated about how electricity is managed. Battery storage can add flexibility to the system, while certain computing tasks can be shifted to better align demand with available power.
That distinction matters. Some of the solution may come from adding generation. Some may come from making better use of the power already available.
As power supply and computing demand become more closely connected, the technologies that help coordinate the two become another area worth evaluating.
Looking further out, advanced geothermal and nuclear tell a different part of the story.
Geothermal has the potential to provide reliable power over long periods, and advances in drilling technology are expanding where it may eventually be practical. Large technology customers can also provide the kind of long term demand commitment that improves the economics of projects that historically faced meaningful development hurdles.
Nuclear may prove even more consequential.
We are already seeing long term power agreements support efforts to return previously retired nuclear capacity to service. At the same time, small modular reactors are progressing toward deployment and could eventually provide another path to reliable power in a more standardized and potentially scalable format.
What matters here is that large technology companies now have the scale, balance sheets, and long term demand profile to support projects that may not have been viable under a more traditional customer model.
That can change what gets built. It can also change which technologies attract capital and which projects become commercially viable.
The power strategy around AI already spans multiple time horizons. Some solutions are addressing more immediate capacity needs, while others are being financed today because the companies driving this buildout are planning around power demand that could extend for decades.
Nuclear deserves a deeper discussion of its own, particularly as existing plants and small modular reactors develop along very different paths. We will return to that in a dedicated Perspective.
What we are watching
This market will continue to evolve as more projects move from planning into operation and the industry gets a clearer view of what works economically at scale.
Some of today’s solutions may prove most valuable over the next several years. Others could become durable parts of the infrastructure supporting AI for decades. Continued innovation will almost certainly change the economics along the way.
For us, the opportunity is in following how that landscape develops. We are watching where capital continues to move, where scarcity remains difficult to resolve, and which businesses become more strategically important as data center operators take greater control over their power strategy.
That is where we believe some of the more interesting opportunities may be taking shape.
Sources: U.S. Department of Energy, U.S. Nuclear Regulatory Commission, Ontario Power Generation, and public energy and infrastructure research.
Securities and Advisory Services offered through LPL Financial, a Registered Investment Adviser. Member FINRA/SIPC. www.finra.org, www.sipc.org.