I’m a multifamily guy. Twenty-five years now, and every one of them spent underwriting apartment communities, not a single data center deal in the mix. So, it’s a fair question why I’ve spent more time than I expected over the past year thinking about a sector I don’t invest in.
Here’s my answer: evaluating opportunities well means paying attention to where capital is moving broadly, not just inside my own asset class. Knowing why I’d pass on something sharpens the same judgment that tells me why I’d pursue something else. That’s part of how I try to be useful to the people who invest alongside us. And data centers are hard to look away from right now: transaction volume has hit roughly $450 billion over the past five years, with more than $300 billion of that debt-financed, two-thirds of it in just the last two years. That’s an extraordinary amount of capital concentrated in a very short window. And to be clear, the people who got in early have been rewarded handsomely. Data center returns have outpaced nearly every other property sector, and the best performing REITs in the space are up 40% or more in a year. Those are real returns, earned by investors who saw the demand story before most of us did.
The deeper I looked, though, the more it reminded me of patterns I’ve spent my career underwriting against.
I want to be careful here. This isn’t a prediction that data centers are a bad investment, or that the people building them are making a mistake. Plenty of sophisticated capital is behind this trade, and some of it will be right. What I have are reservations, the kind that come from spending 25 years underwriting real estate for downside protection rather than upside potential, and from noticing when a sector’s risk factors are moving faster than its ability to price them.
The demand case is doing a lot of work.
Nearly every dollar of that $300 billion in debt assumes AI revenue shows up roughly on schedule. That’s not unreasonable, but it is an assumption, and a fairly aggressive one. The utility industry, which has no choice but to plan conservatively, since the grid can’t retroactively expand if it under-builds, is currently planning for about 50% more data center demand than the technology sector itself is forecasting. When the people responsible for the physical infrastructure are more cautious than the people selling the platform, that’s usually worth sitting with. Industry analysts have flagged 2025 through 2026 specifically as a window where supply could catch up to, or exceed, actual consumption.
The spillover is already visible.
This isn’t staying contained to data centers, either. Roughly 70% of the world’s memory chip production is now going to AI data centers, and the resulting shortage has pushed memory prices up 80 to 90% in a single quarter this year, with PC makers already warning of 15 to 20% price increases and little relief expected before late 2027 given how long new fab capacity takes to bring online. When a sector’s capital commitments are large enough to move prices on phones and laptops, that tells you something about the scale of what’s being bet on.
Power, not capital, is the binding constraint.
For most of my career, the limiting factor on a deal has been capital access, entitlements, or construction timelines. In data centers, it’s increasingly none of those — it’s electricity. Grid capacity is already constraining development in most regions, and the cost to build a next-generation AI-ready facility has nearly doubled, from roughly $10 million per megawatt for a traditional hyperscale build to $15–20 million today. Rising construction costs eventually flow into tenant lease rates, which compresses the return underwritten at entry, a mechanism familiar to anyone who has had to reprice a deal mid-construction, just operating here at a much larger scale.
Community and regulatory risk are no longer theoretical.
In the first six weeks of 2026 alone, more than 300 data center-related bills were introduced across over 30 states, presenting a meaningful shift from the incentive-driven policy climate of a few years ago. States that once competed to attract these projects with tax credits are now reconsidering them, and in July, New York became the first state to actually act on it, issuing a one-year moratorium on permitting for new large data centers. Meanwhile, federal
oversight only applies above 100 megawatts, while state legislation increasingly targets facilities as small as 10 megawatts, an unresolved gap that tends to surface later as delay risk. Add to that growing local opposition in several states over water and power consumption, along with documented cases of facilities moving ahead of completed environmental permitting, and you have a sector where community and regulatory friction has become a real underwriting variable, not a footnote.
Where the caution comes from.
None of this is a bet against data centers succeeding. It’s a reflection of how I’ve been trained to think about risk: conservative underwriting means treating unproven demand curves, unresolved regulatory questions, and infrastructure bottlenecks as things to size and price, not things to assume away. That habit of mind doesn’t come from any particular asset class, it comes from having underwritten real estate through enough cycles to know that the deals that go wrong are usually the ones where the demand story outran the diligence.
It’s also, frankly, part of the job. The people who invest alongside us trust us to have a view on where capital is moving and why, not just inside multifamily, but broadly. When a trend is large enough to reshape financing markets, construction costs, and even the price of the phone in your pocket, it’s worth our attention and worth sharing our thinking honestly, even when we don’t have every answer yet.
Since I look at this through a multifamily lens, the contrast that stands out to me is time horizon. Most of our hold periods run five to ten years, and nearly all of the risk in that window is knowable going in — job growth, population trends, permitting timelines, rent comps. A data center underwritten today is making a longer, less knowable bet: on power availability that doesn’t yet exist in many markets, on compute architectures that could look different in five years, and on tenant revenue tied to how a new technology adoption curve actually plays out. That’s not a comment on which asset class performs better. It’s a comment on which one is asking the underwriter to be right about fewer unknowns.
I don’t have a definitive call on how the data center story plays out. I have questions I’d want answered before I’d underwrite into it — and right now, more of those questions are open than closed.
Mark Bates is Chief Investment Officer at Security Properties, a multifamily real estate investment, development, and management firm. He is a CFA charterholder with 25 years of experience in institutional real estate investment.
