The Real Finds Podcast, Episode 104: The AI Real Estate Goldmine Everyone’s Missing With Bruce Garrison
A conversation between Gordon Lamphere, J.D. of Van Vlissingen and Co. and Bruce Garrison, CEO of Big Fiber, a dark fiber and underground networking company serving over 100 on-net data centers across more than 600 route miles. Transcript edited for clarity.
Bruce Garrison: We will build to different places in the future. We’re already seeing that start to pick up. The real estate world will have a bigger addressable market to represent if you really understand distance, proximity, how inference is going to work, and where the data sits for enterprise companies. There are four components of AI consumption for enterprises, and if you understand all four, you’ll be better equipped to help market buildings that no one would have been interested in three years ago.
Gordon Lamphere: Hi, I’m Gordon Lamphere, and welcome to the Real Finds Podcast, where we have real conversations with key entrepreneurs, activists, and researchers who are shaping the real estate industry and, as a result, our world. On today’s podcast we’ll be speaking with Bruce Garrison, CEO of Big Fiber, one of the biggest names in dark fiber underground networking. A McNeese State University graduate with more than twenty years in the fiber infrastructure industry, including senior leadership roles at Zayo, Kansas Fiber Networks, and GTS Central Europe, Bruce now leads a company serving over one hundred on-net data centers across more than six hundred route miles. On the podcast, we dive into how fiber connectivity has become a critical piece of data infrastructure and site selection, why AI inference is poised to multiply data center endpoints tenfold, and how this might reshape the commercial real estate industry. We discuss what’s next over the coming decades and what it means for developers, brokers, and communities hosting these facilities. For site selectors and commercial property owners, today’s episode is well worth a listen. Bruce, thank you very much for hopping on today.
Bruce Garrison: Thanks for having me, Gordon.
Gordon Lamphere: So what got you into the world of wiring, data centers, and the infrastructure that drives tech?
Bruce Garrison: It wasn’t aspirational, more like coincidence, maybe. Once I graduated college, I went straight into the industry I’m in today. I met someone who worked for an infrastructure company, and he made a big commission check, and I said, that’s what I’m going to do. It was literally that simple. I never applied anywhere else. I said, give me your manager’s name, and I’m calling.
Gordon Lamphere: That’s not a bad way to start. Let’s tackle the fiber world not just as a line item, but as a mission-critical part of the modern office, the modern data center, the modern R&D facility. What does a fiber deal look like these days, and what are some of the gating factors for fiber development and projects?
Bruce Garrison: Our fiber deals range from big bulk deals with a hyperscaler to a new gaming company that needs to connect from one data center to another. So it’s a range, from two-to-four-pair in our fiber ring in a city, to a two-hundred-and-eighty-eight-pair fiber ring in a city, depending on the customer. But everything still goes back to the dependency I mentioned: interconnect. These interconnect sites, normally the downtown CBD areas, are still the magnets of everything we do in the digital world. They’re pretty repeatable segments that we provide to our customers.
Gordon Lamphere: You mentioned a fiber ring. What’s a fiber ring for somebody who might not know?
Bruce Garrison: You take an A-end and a Z-end, and you have two paths to get there, so it creates a ring. There can be more endpoints that create a bigger ring, but that’s the basic idea.
Gordon Lamphere: Is there a reason they’re mostly developed in rings? What’s the methodology behind that?
Bruce Garrison: The number one factor is achieving one hundred percent uptime. Data centers target one hundred percent uptime, and your fiber has to do the same. There are constructions going on, road work, fiber cuts. So it’s purely redundancy to provide the SLA commitment they make to their customers.
Gordon Lamphere: A lot of folks who listen to this podcast aren’t necessarily data center developers themselves, but they operate on the periphery of data center development, something we’ve talked a lot about on our blog and podcast. One of the big issues with data centers, and I know a developer nearby who’s doing one, comes down to energy, but another is connectivity, getting a site that works for the volume of wiring required. Could you tell me what connectivity looks like for a data center in 2026?
Bruce Garrison: In the current world, a data center campus, more recently the larger campuses built for machine learning and training, those are five-hundred-megawatt, one-gigawatt campuses, generally require three to four routes back to the interconnect sites and back to other sites they may have in the market. The bigger the customer, the more they put multiple regions in a market, call it A and Z, meaning there are clusters, three or four, and you’re connecting those, replicating content, and then connecting to the interconnect of these downtown carrier hotels, because that’s what allows the public to consume the product.
What has changed is really two things. One, the fiber count, the fiber requirements. What was built fifteen years ago will not solve AI products and platforms in the future, and it’s not solving it today. All the fiber we’ve built is in the last seven or eight years, so it’s new infrastructure, all the same fiber type, much higher fiber count. Fifteen years ago it was a quarter of what we build. The other thing that’s changed is the chase for power. Machine learning and training, the GPU chips, require ten times more power than a CPU. To operate a big data center, you have to go find power, and that stretches the geographies in a market outside the urban CBD, the highly populated area. So the metro networks are growing, and the fiber count is going up.
Gordon Lamphere: You mentioned three to four points. Why is that necessary for a fiber connection to a data center?
Bruce Garrison: The three to four entrances or segments out of a data center is largely driven by five companies that are the biggest buyers of bandwidth in the world, the hyperscalers, and they’re the ones that require that much redundancy. The gaming example I gave, two paths would be fine for them. But why do hyperscalers require that? It goes back to the SLA. You’re operating a business running on AWS cloud, there are uptime performance metrics and latency metrics they’re committing to, and it can never go down. So is three routes enough? Is four routes enough? Based on how much compute is happening at a campus, that makes the segment number more four than three. It’s to create the redundancies they need to maintain SLA commitments equal to those of a data center.
Gordon Lamphere: One thing we’re seeing is that the next wave of data center development and connectivity isn’t just carrier-to-enterprise, but data center to data center. Can you tell me how that’s playing out in terms of site development and the connectivity issues that come with it?
Bruce Garrison: It’s a pretty interesting time we’re in, the evolution of what’s happening. Five or ten years ago, commercial cloud, GCP or Azure or AWS, was built in a few regions in the US, very centralized, all the scale they could get, and those sites needed to get to interconnect. That’s about as non-distributed as it gets. What you’re seeing now is that the training and inference happening now, and even more so in the future, are going to be very distributed. Our estimation is the number of dots on a map is going to go up tenfold. And those dots are going to be, to your point on data center to data center rather than enterprise, the reality is there are so many digital technologies and apps we consume, and it’s a lot easier for the enterprise to put that in a data center where they can reach others and interconnect, versus building and requiring network to get to their enterprise building. The digital world has centralized everything around interconnect in these cloud regions. While those parts of the map remain, there will be many more distributed points where you have to get from the data in the cloud, to the model, to the inference application, to interconnect. So instead of a majority of two endpoints, it’s going to be tenfold endpoints.
Gordon Lamphere: With tenfold endpoints, does that create a whole new set of problems the real estate world should be aware of?
Bruce Garrison: Yes, though it’s likely a good problem, because inference applications require much less power than training. You have smaller power requirements, but because of latency and being close to the user, it ends up more distributed. What we’ll see in the future, and I don’t know if you’d call them data centers, are smaller points of compute and interconnect in many different places. There’ll be new third-party data centers. A hospital system might have a modular pod on site you have to connect to, where their inference compute is happening. We’ve already seen it in one of our markets where manufacturing facilities are converted to a data center because they had committed power already available. There’ll be AI infrastructure at the bottom of cell towers. Some inference applications will sit on your laptop. And there will be many more we’re probably not even thinking about, because the architecture to support the technology doesn’t exist yet, and we’ll keep evolving and learning. The one thing for sure is that a world of AI creates a much more distributed network architecture.
This distributed, latency-driven shift is exactly what we mapped out in Inference AI Is Rewriting The Commercial Real Estate Site Selection Playbook.
Gordon Lamphere: Let’s talk about solving for that distributed network and for four routes in and out of a building. Not every building can easily have four points of road access. When you’re solving for wiring it’s different, but I’m sure there are problems involved. How do you solve for four paths in and out, and how does that play out in site selection and a project?
Bruce Garrison: It’s very critical in site selection. I do think the real estate world is thinking about it sooner now, but historically they did not. My belief is that fiber’s under the ground, and once it’s built it sits there, so it doesn’t get the same attention or understanding. It’s one of the smaller parts of the total cost of ownership of a campus to get the fiber connected, but it can also be the longest duration. If you’re not planning ahead, someone will build a brand-new, beautiful, shiny data center and it’s going to be an island, not connected to anything. That’s a perfect example of why what we do is so important to the data center world: no matter your uptime, how great your site is, or how big it is, it doesn’t do any good until that compute can get to other compute. So you need fiber.
The number of pathways is a challenge. In the CBD you have more roads and streets, but because of this diversity and redundancy topic, you’re not redundant or diverse if you’re all building on the same right-of-way, all on the same side. You have legacy fiber networks that don’t have enough capacity to support AI technologies at scale, and streets that have already been dug up fifteen years ago. In the CBD, while there are more roads and options and we try to build on rights-of-way, I’d say eighty to ninety percent of our network is built on rights-of-way that others are not. It’s easier to solve in the CBD because you have more roads. But the challenge is that a lot of urban, highly populated areas are becoming a little resistant to data centers. I understand that. You probably don’t want to walk out of a restaurant and have five data centers looking at you. So if you can’t do it in a CBD, you have to solve it in more rural areas, sixty to eighty kilometers, maybe a hundred kilometers outside the CBD. But then there are only so many county roads, only so many paths back to the CBD. The data center power is easier to solve, but the fiber requirements get harder. I think the real estate world and site selection teams are understanding that more. When the five hyperscalers will spend three trillion dollars on infrastructure, they’ll try to throw money at the fiber problem, but it’s a real issue that will get proliferated over the next few years, because the bandwidth and throughput required for inference is going to be more than we’ve ever seen. The existing fiber infrastructure built ten years ago will not be able to solve it, so companies like us have to build it, scale in new markets, and change the dynamics.
Gordon Lamphere: How do you build in a world where there’s political pushback, municipal pushback, and regulatory snafus? How do you build effective, scaled wiring?
Bruce Garrison: Building effective wiring and doing it on the timeline we promised is a challenge. Two things come to mind first. Fiber is a very local business, and I probably didn’t appreciate that as much five years ago as I do today. Who’s on one side of the street, relationships with counties. You don’t walk into a municipality and start a relationship and everything’s fine tomorrow; it’s a long-standing relationship. So step one is having a team of local OSP, outside plant, operational groups experienced in the market who’ve been doing it a while, because those relationships and that experience exist. It’s a local, resource-intensive business. Number two, our chief network officer and our founder have been building networks together for probably sixty or seventy years combined. That experience of who built something on what street twenty years ago is still relevant today. That tenure and experience on the construction side is likely the only way to do it successfully, because you have state, county, city, and federal jurisdictions, and one point-to-point segment could go through four different jurisdictions. We’ve now surpassed three hundred miles of new fiber infrastructure in the Bay Area, but the original two hundred and seven was for one customer, and it took over a hundred permits just to do the original network. The timing on those permits varies; some municipalities and jurisdictions are easier to work with than others. Because of all that complexity, it’s much more of a local business. If you build a data center, and I’m not downplaying data centers, you can control the box, the building; the project management team can handle the generators and coordinate everything, so it’s more full control. Building fiber infrastructure at scale, on the number of paths required, has a lot more complexity you don’t have one hundred percent ownership of, and you need experienced people who’ve been in the market a decade or more and have already run into these problems, so they’re quicker to solve them.
Gordon Lamphere: As someone who’s seeing data center sites and wiring them all around, what are you seeing in how site selection is playing out, in terms of the geography of data centers and where they’re locating relative to cities, and any unique features driving site selection?
Bruce Garrison: First, it’s important to state that there are different types of sites, so requirements change. In the real estate world you’ve probably heard of Abilene and everything going on in Texas. Why? Because they can find liquid natural gas pipelines, access to power, cheap land. They build massive campuses for training compute. From an individual’s perspective, when we ask ChatGPT for something and get an answer back in three minutes, that’s fine, and I could have done that in Atlanta, Georgia, going back to Abilene, Texas, and back; the latency is fine. But the real-time applications we’re going to see in more enterprise AI products, autonomous cars, connected manufacturing facilities where you want real-time analytics, robotic surgeries, all of those require much less latency. So it’s going to get compacted into population centers.
The first stage we’re in is machine learning or training compute, and the hyperscalers have requirements of sixty to eighty kilometers, or less than a hundred kilometers; they’re all different, sixty kilometers max for one. What we see changing is they’re getting stretched to the max, a little further past that, because that’s the only place they can find land and enough power for a gigawatt campus. As that evolves, it’s creating the expanded metro I mentioned. But the later changes are going to be inference applications and the fill-in within that. The first thing you solve for in site selection is power, and whether it’s in a distance you can live with. Then, is there fiber infrastructure, and can they get it there? With enough lead time, that’s easier to solve. Nothing will ever surpass the power requirement as first. But then site selection ends up being fiber optionality, number of routes, and expansion capabilities. In a world of inference and more enterprises using AI, you’ll have a lot of density and require a lot more network in these expanded metros than exists today.
The power-first, fiber-second sequencing Bruce describes is the same site-constraint logic we unpacked with Whitaker Irvin Jr. in Hydrogen, Data Centers, And The End Of Energy Poverty.
Gordon Lamphere: When we look at that, and inference AI computing is definitely changing the game, are there certain sites that are extremely cost-prohibitive, even if the power is sitting there, for fiber to reach them? And what do those sites look like?
Bruce Garrison: We stay close to the developer and real estate brokerage community in our markets. There are plenty of sites on the market, some under diligence, some under LOI. If there are a hundred parcels being marketed, about forty percent will fall out. They fall out because the power timeline is too long, everyone wants more power now, or the distance issue: if they put a node here, is it too far from the one they already have, and can they get enough segments? It’s a much bigger stress on networks, because in rural areas, and by rural I mean eighty miles or eighty kilometers outside Atlanta, you don’t have the infrastructure built at the scale to support AI. Someone has to build it. It’s really about getting power; if they can get power, they throw money at the problem to solve everything else, and we just get pressed with short timelines to build the fiber segment. The site selection evolution is distance and proximity, water, expandability, and probably fiber after that. I don’t think that dynamic changes for the next two to three years. In three to five years, when there are more inference applications, I think fiber will become a much earlier issue to address.
Gordon Lamphere: We’ve talked a lot about data centers today. One question we always love to ask is: what’s one topic the real estate industry isn’t talking about enough that we haven’t already covered?
Bruce Garrison: I think they’re starting to talk about it, the hyperscalers and the real estate community. A lot of the data center pushback in a municipality, when you read the press, is centered around not wanting the emissions from generators, or why we’re giving a seven-year property tax abatement. These buildings are going to be there thirty or forty years, and these downtown interconnects much longer than that. In our lifetime, these big buildings are never going away. What’s missing, and what would help everyone in the industry, is better education for local communities: sure, you give a seven-year tax abatement, but over thirty years, here’s how much tax revenue it generates. You can look online at the case study for Loudoun County in Northern Virginia, the number of new fire stations, schools, police stations, and libraries they can create. I don’t know of a better funding source for a more rural community that has less access to tax revenue because of population. The education on those longer-term benefits should have already been discussed. A year ago, the conversations are picking up, because as they run into more restriction, it forces the conversation. But to be proactive, to talk about it earlier, to get in front of a city council and explain to City Hall and as many residents as you can the long-term benefits of this building over the next thirty years, that’s just not being done enough. That conversation should have started a long time ago.
Gordon Lamphere: You’re on the forefront of a lot of sections of the commercial real estate industry with your data center work. What are you seeing in your practice, and what do you think the commercial real estate industry will be ten years from now?
Bruce Garrison: Today, the commercial real estate world relative to data centers is finding a lot of land for a lot of compute, so there are bigger sites but fewer of them. What I think the next decade becomes is more of a volume game, where the real estate community has to help find a building with power. Modular data centers are going to pop up because of time to market. It takes a normal third-party fifty-megawatt data center two years to build out, and in the world of inference you’re going to need power quicker. So instead of chasing big campuses, and those campuses will still get added, there’s also going to be a smaller, less costly, but higher-volume marketing approach, finding locations that fit that criteria. The marketing of a big parcel and five hundred megawatts of power turns into twenty more at a smaller scale that have to be replicated for inference. So the volume of transactions will change; they’ll be smaller but higher volume.
The other part is what actually becomes a data center is going to look a lot different. Maybe we don’t call them data centers in ten years, but AI infrastructure will be at the bottom of macro cell towers. You’ll have the hospital system, the shiny enterprise building for a bank. A lot of these companies have on-prem data centers, so can they expand within those buildings? Which enterprise buildings become more attractive because they’re a hub for AI technologies? The inference sits outside on a modular data center in a parking lot, or it goes into a building. When we talked about everything being data center to data center, you’ll now go back to more distributed enterprise locations, cell towers, modular data centers, sites we don’t even think about or know will exist today. You’ll have a different site type for AI compute, AI infrastructure, AI interconnect, just at higher volume, and they’ll be different types of sites we haven’t seen yet. A third-party data center is not going to solve all of this. The carrier hotels are great locations, but they’re hard to expand. So distributed AI is going to sit in places we’re not building to today in our business model. Going from ninety-five percent data center to data center will evolve to still probably seventy-five percent data center to data center, but there will be new location types that are accretive to connect to and a necessity to have a competitive moat in the market. As a result, we’ll build to different places in the future. We’re already seeing that pick up. The real estate world will have a bigger addressable market to represent if you really understand distance, proximity, how inference works, and where the data sits for enterprise companies. There are four components of AI consumption for enterprises, and if you understand all four, you’ll be better equipped to help market buildings that no one would have been interested in three years ago.
Bruce’s thesis, that ordinary buildings with power become valuable AI-adjacent assets, is the same revaluation we examined in Valuing Chicago Data Centers and Adjacent Properties.
Gordon Lamphere: Let’s go back more than three years, to the start of your career. We have a lot of younger listeners, and one question they always love, and why we keep asking it, is: if you could go back to the start of your career and give one bit of advice, what would it be?
Bruce Garrison: I actually talked about this with a mentor at lunch today. One, swallow your pride and ask for help sooner. Related to that, number two, understand that the learning process is continual; it never stops. As a result, you’re never going to have all the answers, so you have to know how to find them. Asking for help is one thing I didn’t understand the value of, even into my career today as CEO, leveraging the expertise of board members. If college graduates can do that sooner, their skill set and capabilities will accelerate much faster than those of us who didn’t swallow our pride and didn’t ask for help until the second half of our careers.
Gordon Lamphere: Those are both suggestions we’ve gotten before from some pretty great folks, and I’d definitely recommend both. Having the ability to swallow your pride can be critical in this business.
Bruce Garrison: We all have to understand that never stops, the need to ask for help, the need to learn. You asked what point I’d make to myself earlier in my career; some of these conversations I’m having, I should have told the guy two years ago the same thing. It’s still evolving, and it never stops. Ego gets in the way of good decisions. Swallow your pride, ask for help, and learn something.
Gordon Lamphere: There is one thing we’ll ask for help on, and we always ask this question, it’s the whole reason for the podcast: who’s the next person we should bring on? We believe the men and women in the arena tend to know who the next voice should be. So who’s the next person we should ask to hop on?
Bruce Garrison: Everyone building chips competing with NVIDIA. You have LPUs from Groq, which is now part of NVIDIA. Cerebras Systems develops a chip for inference. Those developing the chip that’s going to make it work know where all these endpoints will have to be, know how much power will be required, and that tells you where the locations are going to be. Cerebras Systems comes to mind because they have an office close to ours in Santa Clara, California. Companies like that, building chips for enterprises to consume AI technologies, know where the puck is going next.
Gordon Lamphere: If you could ever give me any contacts there, we’d love to interview anyone on their team. If somebody wants to reach out to you, what’s the best way?
Bruce Garrison: You can go to our company website and submit an inquiry. Every customer we have has my contact information, and it’s generally easiest by email. We have a company LinkedIn page, and I have one as well, but the one that works best for me is someone reaching me by email. I’m happy to help anyone, figure out some things, ask some questions, because I got a lot of help, so I should pay it forward. Even when someone’s asking me questions, I’m still learning from my own answers.
Gordon Lamphere: We can put your contact information below on the podcast. Bruce, thank you so much for hopping on today. We really appreciate it, and we have to have you on in the future. Thanks again to Bruce, we appreciate his insights. If you enjoyed the podcast, please give us a five-star rating and a review. Your comments, interactions, and subscriptions truly matter and help us continue to provide quality guests. You can find us on YouTube, Spotify, or wherever you get your podcasts. I’m Gordon Lamphere, the Real Finds Podcast, and thank you for listening.
Van Vlissingen and Co. has been the Midwest’s oldest commercial real estate brokerage, development, and management firm since 1879, and today is independently ranked the #1 commercial real estate agency in Chicagoland, home to the #1 independently ranked agent, Gordon Lamphere, and the region’s #1 ranked commercial property management team. If you own, manage, or invest in data center, industrial, or AI-adjacent property across Lake County, the North Shore, the Northwest and O’Hare corridors, DuPage and the I-88 corridor, Will County, or southern Wisconsin’s Pleasant Prairie, Kenosha, and Racine markets, contact Van Vlissingen and Co. at 📞 847-634-2300 or 🌐 vvco.com. For a market-wide view of where these dynamics sit today, see our State of the Chicagoland Commercial Real Estate Market for Q2 2026.