In this episode of the REconomy Podcast™, Deputy Chief Economist Odeta Kushi and Principal Commercial Economist Xander Snyder examine the seemingly endless demand for data centers and explain why the story is more nuanced than the headlines suggest. The conversation breaks down why vacancy rates near record lows and surging pre-leasing activity reflect a demand base that extends well beyond AI, and highlights the real bottleneck to new data center supply.
Don’t miss a single REconomy episode, subscribe today.
"There is a diversified demand base that has wind behind its back for data centers that's beyond just AI." — Xander Snyder, Principal Commercial Economist, First American
Odeta Kushi - Hello and welcome to episode 144 of The REconomy Podcast, where we discuss economic issues that impact real estate, housing and affordability. I'm Odeta Kushi, deputy chief economist at First American, and here with me is Xander Snyder, senior commercial real estate economist at First American. Hey Xander, welcome back.
Xander Snyder - Hey Odeta, how's everything going?
Odeta Kushi - Everything's going well and I'm really excited about today's topic because it seems to be one that's everywhere you look these days, and that is data centers. They're being discussed in commercial real estate circles, energy circles, technology and AI circles, mass media, pretty much all throughout the public sphere. And I don't know about you, but if you're ever in social circles, I don't think you can get far without a mention of AI.
Xander Snyder - So lots of circle and sphere metaphors there. But yes, what was once a fairly niche asset class, data centers have now really become the center of a lot of public debate. How many of them do we need? Where should they go? How should costs be distributed? And today we're going to talk about the current state of play of the data center industry and try to separate out a few different narratives that seem to have become conflated. Namely that the data center space is the AI space, and it isn't.
Odeta Kushi - That's right. And data centers are, of course, needed to train and run large language models, but they've been around since long before the advent of consumer-friendly AI. These large buildings full of servers, cooling equipment, backup power systems, fiber connections and an exceedingly complicated electrical infrastructure are in high demand, but not just because of AI, but also because of the broader digitization of our economy. You might say that the cloud is very much on the ground.
Xander Snyder - Ugh, cloud on the ground. Okay, I think I'll go ahead and roll my eyes on behalf of Mark for that one. Yeah, data centers have become some of the most sought-after real estate assets in the country. And the basic story is that demand for data center capacity has surged, driven by AI workloads, but also cloud computing, streaming, storage, financial services, enterprise software and all of the other digital activities that are increasingly surrounding us wherever we go.
Odeta Kushi - The digital world is following you around. It seems like you might need to go outside and touch some grass, Xander.
Xander Snyder - Tell me about it. Yeah. In a world where technology does seem to be increasingly pervading all aspects of our lives, data centers are perhaps the most obvious example of real estate being tied directly to technological change. Every time people upload more data, stream more video, train more AI models, run new software, or just ask a chatbot to generate a vacation itinerary, somewhere there is a server doing work.
Odeta Kushi - And somewhere there is a data center developer asking a utility company when the power will be turned on.
Xander Snyder - Yeah, exactly. And that's easily the most substantial bottleneck to new data center supply being added to the market right now — the interconnectivity to the wider grid. There's a lot of demand for compute, but there are also lots of challenges to building new data centers. Power availability is one of the big ones. You can have the land, the capital, the pre-committed tenant and the shovel-ready construction plan. But if the local grid or the local utility operator can't deliver enough electricity on the timeline that you need, the project won't work.
Odeta Kushi - Which helps to explain why some large data center users have been preemptively acquiring land that could someday become a data center site. Part of it is about future growth, but in the case of certain large data center users, part of it is about keeping competitors from getting there first.
Xander Snyder - It's the real estate version of musical chairs, you might say, except that the chairs are utility interconnection agreements.
Odeta Kushi - Wow. Really catchy, Xander.
Xander Snyder - Yeah, I'm maybe still workshopping that one. This phenomenon of preemptively acquiring space before being ready to start developing, or land banking as it's known in the industry, is prevalent among hyperscale companies, or hyperscalers, which is a somewhat futuristic name given to the small category of very large companies that regularly utilize massive data centers by themselves. So, think Microsoft, Google, Alphabet, Amazon, Meta, Anthropic, OpenAI, these sorts of folks.
Odeta Kushi - Yeah, this sort of preemptive purchasing behavior is indicative of the scarcity of data center space as demand for AI goods and services is accelerating exponentially. This resulting imbalance and the difficulty in overcoming the supply constraints are likely to remain in place for the next few years. And, if that's true, should we assume that if a large tenant in a data center had an economic challenge and maybe abandoned a lease, that there would be multiple tenants waiting to step into that space?
Xander Snyder - I think in most cases, probably yes, but with caveats, because economists are legally required to bring caveats, right? The market is currently very tight across the country. Vacancy rates in major data center markets are around record lows, around one to two percent, depending on the source that you use. So if a facility has available power, strong connectivity and is located in a constrained market, then there's a good chance that the owner would be able to fill the space if it were to become vacant.
Odeta Kushi - Hmm, I can hear the however coming. There's no such thing as a one-handed economist.
Xander Snyder - I don't know what gave it away. However, not all data centers are the same. There's an important distinction between what's called co-location facilities and hyperscale facilities. Co-location centers are typically built for multiple tenants. The owner leases space, power, cooling and connectivity to a bunch of different tenants. Tenant turnover, therefore, is part of the business model and accounted for in the owner's budget. Hyperscale facilities, on the other hand, are sometimes owner-occupied. They're sometimes leased out, but they're very large and often single tenant.
Odeta Kushi - Hyperscale facilities are also different in that because they are so big, it can limit the depth of the potential tenant pool. So, if a leased hyperscale facility is vacated, it may not be immediately usable by another without significant additional investment.
Xander Snyder - Right, another large tenant could very well have different requirements, perhaps greater power density, requiring more modern cooling systems or different network or grid connections. And, in these cases, re-tenanting risk is still low compared to a lot of other types of commercial real estate because there's so much demand for data center space. But it isn't zero. Retrofitting may still need to occur before a new tenant can occupy that vacated space.
Odeta Kushi - So data centers are a lot easier to lease than, say, office buildings, but they aren't magic power boxes that can be instantly reused by anyone with a laptop and a dream.
Xander Snyder - Correct. If you build it, they might come.
Odeta Kushi - But what if you build it with AI?
Xander Snyder - These are the great questions of our time, Odeta. I do think that counts as our contractually obligated 1980s reference. The classic movie, of course, Field of Dreams from 1989. If you build it, they will come, et cetera, et cetera.
Odeta Kushi - I think Mark would be very proud of that one, maybe a little too proud. Indeed. Speaking of how it's built, we've heard people describe the data center and AI industries collectively as a house of cards, but how legitimate of a characterization do you think this is?
Xander Snyder - Okay, okay, I'll take it. So I don't think it's accurate to describe the core data center market as a house of cards. As we discussed, the physical data center market remains very constrained. Vacancy is low, pre-leasing and pre-commitments are high, rents have been rising and demand for data center compute is broader than just AI.
Odeta Kushi - So, even if AI demand slowed, new data center demand would not disappear.
Xander Snyder - Right. And I think this is an important point that gets lost sometimes in the most recent AI and data center narratives that float around. Though AI is an important and fast-growing source of demand for data center compute, data centers also support other workloads like cloud services, software, hybrid IT and data storage, so on and so forth. There is a diversified demand base that has wind behind its back for data centers that's beyond just AI.
Odeta Kushi - But, and I'm going to channel my inner Debbie Downturn here, might there still be some house of cards element in the broader AI ecosystem?
Xander Snyder - I think you could make that argument, but I'd argue that it has more to do with AI growth expectations rather than demand for data center compute. Even if AI turns out to be a transformational technology, which I certainly think it will be, that doesn't mean that every growth forecast, valuation, or financing structure will prove correct. After all, expectations can change.
Odeta Kushi - You might even say that you should expect expectations to change.
Xander Snyder - How do I roll my eyes on an audio recording?
Odeta Kushi - You know, at least we make ourselves laugh. And practice, Xander. That's how you do the eye roll. It's all in the tone. All right, so if data center demand is strong and supply is currently constrained, what could happen that could change demand over time and how much time might that take?
Xander Snyder - Well, I think the two largest risks are end user AI prices rising materially, or AI productivity gains disappointing. If either of those were to occur, then those growth assumptions could come under pressure and we could see demand soften, or at least for growth to slow from prevailing expectations.
Odeta Kushi - That makes sense, but it's also worth noting that there's a credible optimistic case that could also play out. Model efficiency could improve, hardware could get better, inference optimization could reduce costs and the AI cost curve could keep falling.
Xander Snyder - What happened to Debbie Downturn?
Odeta Kushi - Even she needs to take a break sometimes, and often should. But coming back to the demand risk, rising prices could eventually put downward pressure on compute needs, which could slow demand for data center space, at least over a longer-term horizon. Our base case for the data center market over the next several years then is that demand continues to exceed deliverable supply through the latter half of the decade, but risks increase gradually as more supply is delivered and as AI economics mature. Does that sound about right?
Xander Snyder - I'd say so. Yeah, markets with constrained power and strong connectivity should remain well positioned. But, by the early 2030s, we will have a clearer view of how durable AI-related demand really is, at least at price points that are sustainably profitable. Additionally, if AI productivity gains do disappoint, or if AI services become more expensive and demand slows, the urgency around acquiring land and locking up future development opportunities — which we mentioned earlier is an increasingly common trend for large data center owners and owner-occupiers — could diminish.
Odeta Kushi - So, while data centers may power the ethereally named cloud, the economics are still grounded in familiar real estate fundamentals: supply, demand, location, tenant quality, financing and execution risk.
Xander Snyder - Yes, plus actual power. Lots and lots of power.
Odeta Kushi - Details, details. Yes, of course. But that is a topic for a future episode. Suffice it to say, with data centers becoming an increasingly important asset class powering one of the fastest-growing sectors of the broader economy, we will certainly be returning to this topic in the near future. And that's it for today's episode. Thank you for joining us on this episode of The REconomy Podcast. If you have an economics-related question you'd like us to feature in the future, you can email us at economics@firstam.com. And, as always, if you can't wait for the next episode, you can subscribe to our Econ Center at firstam.com/economics or connect with us on LinkedIn. Until next time.
Thank you for listening, and we hope you enjoyed this episode of The REconomy Podcast from First American. We're pleased to offer you even more economic content at firstam.com/economics. This episode is copyright 2026 by First American Financial Corporation. All rights reserved.
This transcript has been edited for clarity.