Dear Valued Clients and Friends,
It is hard to turn on the news these days without hearing some talk about “data centers.” Whether it is financial media, technology coverage, or just broad cultural commentary, data center chatter is all the rage.
The chatter covers a number of topics, though the words “data center” appear throughout. The economic angles, political alignments, and public angst around all of this represent different facets of the topic, and all need some unpacking. Investors, in particular, need to understand the data center discussion, because (a) it really is a big deal, and (b) the bigger the deal, the dumber the things some people say.
The Dividend Cafe exists to offset the dumb things, and offset we shall. Are data centers new? To what extent are they particularly relevant to the economy now? How are investors exposed? And what is the controversy and drama all about? Today, we will unpack all of this and leave you more prepared as an investor and more informed as a reader about the Data Center Drama moment of 2026.
Let’s jump into the Dividend Cafe …
| Subscribe on |
Nothing New Under the Sun
So first of all, let’s just get one thing out in the open … Data centers did not start with ChatGPT. There is a particular AI-orientation to the data center discussion we will get to in a moment, but at the highest level, a data center is nothing more than a physical facility that houses servers, storage systems, networking equipment, and data infrastructure. It is a factory, but instead of it being a factory producing widgets and parts, it produces digital services. The internet exists through the data stored and processed in data centers. Well before the AI context for data centers existed, data centers were built to run the internet in the 1990’s. These really were essentially buildings that looked like warehouses filled with racks of CPUs. Websites were hosted and powered by this, and, more importantly for the early iterations, emails were able to be processed. Good times! It was smaller-ball stuff then, and many smaller businesses could just keep a “server room” (or an “IT closet”) that could handle this task, but the warehousing of such a thing began over thirty years ago.
This went to the next level with the cloud. It was not just the data and processing we all wanted to do with our smartphones, but the entire advent of cloud computing that led to the next iteration. This was the beginning of “hyper-scaler” talk, and these really were massive facilities that housed tens of thousands of servers to manage the data and traffic of the cloud, social media, streaming services, and more.
New and Improved
Now, data center talk usually refers to AI infrastructure. As the generative models of AI (such as ChatGPT, Claude, Grok, and Gemini) take off, they require vastly more computational power than ever before, and therefore more space and infrastructure to support all of this. I do not want to get deep into the tech side of this (which bores me as much as it would bore many of you), but the aforementioned data center categories were largely centered around CPUs, and AI data centers are largely centered around GPUs. The difference between a central processing unit and a graphics processing unit is technically important but somewhat immaterial to my subject here, other than to say that GPUs require at least ten times more power than CPUs. And this all means they require way more cooling than traditional CPUs. So we are all at once talking about:
- Much larger facilities
- Many, many more facilities
- That would require way more power/electricity
- And need a lot more cooling
These are not just “warehouses” like the 2010s, and they are certainly not glorified server closets like the 1990s. They are massive digital “libraries,” if you will, that have to exist at a completely new level of industrial and technological sophistication, size, and scale.
The Economic Reality
Total technology investment was the major driver of economic growth in 2025. Harvard economist Jason Furman, a straight shooter even when he is wrong, calculated that investment in technology accounted for 92% of GDP growth last year. Renaissance Macro Research estimates that AI data center buildouts represented a higher dollar contribution to GDP growth than all consumer spending did last year. Much of the most potent data is not as clean as it may sound. That technology investment Furman references includes equipment and software, meaning some of that would have been spent without data center investment, while some presupposes the data center infrastructure to make it all work.
What we do know is that total construction spending declined 1.4% last year, but data center construction spending increased 30% on the year, leaving mathematically inclined people to calculate how brutal the drop in construction spending would have been had it not been for data center construction. The data center sub-category was 4x higher in 2025 than it was in 2021, and this represents actual spending – not merely planned or desired projects awaiting entitlement. The backlog and pipeline will see that number increase another 10x in the years ahead. We are spending more on data centers than all other manufacturing facility categories put together.
Our Industrial Production for IT equipment is up 80% since the decade began, but for everything else it is … dead flat. S&P Global’s study concluded that 80% of our GDP growth in 2025 is attributable to data center and data center adjacent (tech equipment) spending. Of all the studies I have read and analyzed on the subject, this is the one I found most compelling in terms of the simplicity and clarity of the data presented.
And What it All Leads to
Of course, the construction of data centers (and the accompanying investment in equipment and such) is a one-and-done for contribution to economic activity if one is only considering the construction. But why would hyper-scalers spend billions of dollars on these projects if the construction was the end-game? The entire point is that these investments put assets into place that drive sustained growth via greater productivity. In other words, either these data centers lead to greater productivity via the entire adoption and implementation of AI, or they fail to do so, in which case the whole AI thesis is strained. But the point is that the stakes are very high for GDP growth here: either data center construction is vital to the present and future of U.S. economic growth, or we are about to strand a lot of wasteful assets in never-never land.
The Quiet Part
We are not merely talking about a hugely important contributor to the U.S. economy from a single subcategory within a single silo. Non-residential private investment is just one element of GDP growth (business spending), and data centers are a very niche element within that sleeve. But if data center projects have become this substantial in their contribution to economic growth, it also speaks to how muted the contributions have become from, well, everything else. Most other construction growth has become a zero. Multi-family has substantially cooled. Warehouse builds apart from data center are minimal. Office construction is down. The utilities, power demand, and energy demand to fuel the power generation have become a whole industrial cycle story of their own! And this is all a real thing, and it is basically a good thing, but it does seem to suggest that we would be facing a pretty muted economic environment without it.
Historical Analogy
I have become fond of using fracking as an analogy in various speeches I have given over the last few months. With a h/t to Strategas Research, I think this historical comparison works, but for more than one reason. I would say the contribution of horizontal drilling and hydraulic fracturing (and adjacent activity) to economic growth in the years following our 2008 economic crisis has never been fully appreciated. When $133.7 billion was invested in the space over a couple of years, at a time that there was virtually no business investment anywhere else, it is hard to overstate how meaningful this was to economic growth when the economy was teetering on a double-dip recession moment (but avoided one). The energy capex story of that period was a massive contribution to economic growth, even if economic growth itself was, shall we say, muted.
Much like today, total economic growth is not substantial, but this AI capex/data center contribution to it is really, really important – perhaps all of it.
Yet what Strategas Research has pointed out in this comparison is less about the economic data and more about the popular, political, and cultural mood that surrounded/surrounds it all. Fracking was not exactly popular with all constituencies. The Obama administration did far more to support it than has been acknowledged by either his supporters or his detractors, but they certainly didn’t go out of their way to promote this! Imagine driving a major catalyst of economic growth, but being afraid to take credit for it because the thing driving such a positive economic activity is so unpopular, or at least controversial, with your own voting base!
In that sense, I see a huge analogy between Data Centers and the shale revolution:
- They both are disproportionately relevant to economic growth, and
- They both are not very popular with a key block of the nation’s voters
(Though I would say that the fracking sentiment was much more identified by partisan affiliation, whereas the data center sentiment seems to be much less partisan)
The Black Hole for Nuance
These debates exhaust a classical conservative like me. I, all at once, support the robust construction of data centers to feed, fuel, and furnish the ongoing power and computing needs of our society, and yet find the idea that the federal government would be allowed to circumvent the zoning ordinances and general entitlement processes of a local community reprehensible. In other words, I believe many localities should support data centers, and at the same time, believe the federal government should not butt in, whatsoever. This is called federalism, and the various economic arguments for progress, investment, and opportunity (embedded in data centers) can all be true without believing that the 10th Amendment is obsolete. I believe many cities and localities have zoning ordinances that are too restrictive or too selective, but that has nothing to do with believing that cities and localities do not have the right to set their own zoning ordinances free of mandates from Washington, D.C.
But AI is too big and too important for the feds not to intervene, you say! The overarching public need is so large here that we can’t wait around for a bunch of nostalgic NIMBYs. I believe this is actually the entire point of federalism and localism. Let one community’s preferences give them (or not give them) what they want (or don’t want) while another does the opposite. Various communities get customized to their own bottom-up preferences, and the shape of society takes hold accordingly. But claiming “this is too important to follow our own laws” is, first of all, utter hogwash. Every commercial interest can lay claim to a priority when they desire to circumvent the processes everyone else has to live by. The idea that the first time we need to throw out our zoning and entitlement laws (which have been wreaking havoc on small businesses, non-profits, and other community actors for decades) is for Silicon Valley trillion-dollar company interests is just outrageous.
And this does not mean I do not want the data centers for the hyper-scalers, because I absolutely do. I just want them to play by the same rules we all do.
Okay. Back to the subject at hand.
So What Happens Next?
Andover Township, New Jersey, announced it will ban data centers. Augusta, GA, turned down a big project. Wisconsin has been fighting data centers tooth and nail. New York State has wrestled with a bill to ban data centers statewide. Northern Virginia already has a vast number of data centers (600+ statewide, providing about 25% of our national capacity), but recent opposition in both Prince William and Loudoun Counties has reached a fever pitch. Despite heavy support in many parts of Arizona, a $14 billion project in Goodyear was canceled after a resident uprising. The Governor in Maine had to veto a legislative moratorium on any new centers. Nearly six dozen cities, counties, or states have imposed some form of ban or restriction.
There were 49 rejections of data center projects last year. There have already been 79 this year, with well over half of the year to go. The most recent Gallup survey on the subject found that 70% of Americans opposed data centers near their homes. The problem here is that survey data gets murky as to who is just a classic NIMBY (“I don’t want to look at anything I don’t like looking at”) versus more calculated concerns about energy prices and the reliability of their local electric grid, as well as water systems.
At the end of the day, the aggregate opposition from those who merely view them as a visual nuisance, along with those who have concerns about cost and logistics, is substantial. And because of the economic significance of this entire story (see above) this tension is becoming nationally significant.
Where Do We Go from Here?
The AI companies don’t seem to be much interested in winning the war of public opinion. I could provide all sorts of particular anecdotes, but I don’t want to be accused of picking on one company versus another. Let’s just say that the loudest proponents of AI and AI expansion from inside the AI labs do not always seem like sympathetic characters. But in making the case for data centers, one is making the case (theoretically) for something that is supposed to be benefiting American citizens. Fear-mongering about competition from China is not going to close the deal. Making the case out of fear and panic is not going to work. And making the case by not making the case is also not going to work (“we really need data centers and they are going to benefit your lives so much, but actually we don’t care if we have your approval or not because we are just going to ask the White House to override your city council”).
The need of the hour is a massive public relations effort that is economic, political, and cultural, and done in the spirit of the social contract of our economy. It must, all at once, be rooted to:
- The economic reality that data center build (and AI capex) is vital in the present economic moment,
- The economic reality that IF AI is going to deliver the productivity growth its evangelists have promised, is vital for the future economic moment.
- The local sensitivity that we have zoning laws and a community process for a reason (for good or for bad), and honoring the same process that churches and Costcos and your local town pub have had to follow for decades is important lest we give the impression that rules exist for the little guy but not Silicon Valley billionaires.
- A disdain for cronyism where water, power, and other costs are socialized but profits are privatized, and
- The realization that the energy needed to power this construction and investment requires more natural gas and more pipelines to transport the gas.
Conclusion
I am not against building more data centers. In fact, I would argue that right now it is the biggest thing we have going for the economy. But I assure you, they are not all created equal. Much like the pipeline analogies of fracking ~15 years ago, counter-parties matter, project quality matters, project funding matters, and there will be winners and losers. When it comes to direct data center investment, be discriminating. Sponsor quality is paramount. You will see. I am not wrong about this, and some are going to learn the hard way.
But I do not believe there is any basis for rejecting our own societal norms and Constitutional (federalist) structures to accommodate commercial actors. Let localism play out and let different communities experience the trade-offs (which, by definition, contain both good and bad components) they sign up for.
Think about adjacence to this story for investment opportunity, not the story itself. Power generation. Utility output. Water. Natural gas. Pipelines. There is a lot more to this than mere data centers.
And do not forget that in the end, the data center story is a means to an end, not an end in and of itself. The data centers have to create computing power, and the computing power has to fuel AI tools, and those AI tools have to drive productivity. And in those various dominoes, there is so, so much that can and will go wrong, and so, so much that can and will go right.
Quote of the Week
“ ‘Need’ now means wanting someone else’s money. ‘Greed’ means wanting to keep your own. ‘Compassion’ is when a politician arranges the transfer.”
~ Joseph Sobran
* * *
The data center topic is one that could cover a lot more than 3,000 words and one article. But I do hope this gives a bit of an overview of the fundamental tension playing out in the public square. I do not expect it will be resolved quickly, but I do think it will be resolved over time.
Last week’s topic about the national debt is also not one that could be covered in one piece. I do plan to write a future Dividend Cafe (soon) about a follow-on subject many emailed me about, wondering how I think about the asset side of the equation (that is, whether or not we make the national debt sound worse than it is by not thinking about assets the federal government owns). I have an answer to this, but it is a fair question, and a new Dividend Cafe is forthcoming.
In the meantime, have a wonderful Memorial Day weekend. I am off to Houston Monday, then Austin Tuesday, then Dallas Wednesday, then to LA on Friday to speak at the Reagan Foundation Economic Forum. I will bring you a Dividend Cafe next week, one way or the other, as I do each and every week. In fact, to that end, I work. And reach out with any questions, any time.
With regards,
David L. Bahnsen
Chief Investment Officer, Managing Partner
The Bahnsen Group
thebahnsengroup.com
This week’s Dividend Cafe features research from S&P, Baird, Barclays, Goldman Sachs, and the IRN research platform of FactSet
