Cool Vector Video-Podcast: Charting the rise of data centers and the digital infrastructure asset class.
Episode Title: "The Fiber Optic Future With John Siegel of Columbia Capital"
Episode Duration: 52:04
Originally Posted: September 26, 2024
Listen to the Full Episode here.
Speakers: John Siegel, Partner at Columbia Capital; MODERATOR: David Snow, Cool Vector
Episode Overview:
John Siegel, a Partner at private equity firm Columbia Capital, offers a deep dive into the physical assets necessary to power an AI-driven internet, including the fiber optic cables that connect the growing population of data centers around the world, and the nations jockeying for position to build digital infrastructure hubs. In a wide-ranging conversation, John shares his views on the demand drivers of information sharing, including not only AI but the massive proliferation of devices that connect to the internet. He details the data center build-out across Asia and explains why governments are so eager to develop hubs like his home base in Northern Virginia. A long-time telecom investor and "qualified bull," John also shares his analysis of a wave of bankruptcies in the early 2000s (which lost billions for private equity investors) and what lessons these might have for the current digital infrastructure build-out.
TRANSCRIPT
David Snow: Hello and welcome to Cool Vector. I'm David Snow, your host, and today we're joined by John Siegel of Columbia Capital. Columbia Capital is a specialist in digital infrastructure, data centers, enterprise technology, and mobility. John Siegel, thanks so much for joining Cool Vector today.
John Siegel: Very happy to be here, David.
David Snow: All right, well, we've got a lot to learn from you because you have been in and around the digital infrastructure space for quite a long time, in the telecommunications business. You spent a lot of time in Northern Virginia, which is a hub for telecom, and we're going to learn why. Maybe we can just talk about your experience in digital infrastructure. You've been involved in a lot of different companies, and of course every company is different, but if you could find a common macro theme that ties them all together, what do you think that theme would be?
John Siegel: What I would say is if you look over the past 35 years, you've seen this entire space move from one that was entirely focused on services—providing services to end users. People would invest in their infrastructure and then actually use the infrastructure to provide services. Think about your classic phone line, your cellular service, etc.
What you have seen as the internet has become a utility is that companies' and individuals' relationships with that infrastructure have changed. It has gone from being a service much more to underlying infrastructure that people—or really large enterprises—buy on a long-term basis rather than short-term contracts.
That shift to a utility model really points to the industrial strength and frankly the size and scope of what we now call the internet, which was really a telecommunications network and telecommunications infrastructure 35 years ago. It is now something much different, but it really hasn't changed that much to the end user.
David Snow: I'm glad you mentioned the word infrastructure. Where does the infrastructure end and where do the operating business and services begin?
John Siegel: It really depends on the asset class. I would say there are three primary classes of digital infrastructure, and then there may be some subsectors underneath.
The first would be data center infrastructure. The second would be networks—fiber networks, primarily. And the third would be wireless infrastructure—think wireless towers, DAS systems. So where the lines go from raw infrastructure, if you think about someone who is building and providing data center capacity—people in the data center space would contract it on the infrastructure level, build a building, power it, and contract that with a very large end user who would take down a substantial portion of the building or the entire building. Think about it much as a real estate model.
The same thing on the wireless side, where people would build real estate—i.e. a tower—and then lease space on that tower to a wireless carrier. Again, very much like a real estate-type relationship.
Then if you move towards fiber networks, the shades of gray are probably a lot more subtle. You can start with just a raw infrastructure model. You build the infrastructure and either sell it or lease just the infrastructure. It's again a classic real estate-type relationship to a large hyperscaler, a large carrier, or you would even sell them the infrastructure—sell them the fiber networks—all the way to someone like Verizon, where you are selling someone a phone service.
You pay for that service on a monthly basis and you can switch and go to another provider if you'd like. The fundamental relationship in terms of contract, the ease to break that contract, and the capex required are totally different between the infrastructure side and the services side.
In those classes of infrastructure, wireless really was kind of the original gangster of digital infrastructure. Next comes data centers, and there are infrastructure models. There are also models that are retail colocation, where you can own a data center and lease that space out much more on a services basis. Then when you get all the way to the network side, things get a lot grayer.
David Snow: Interesting. As you're aware more than anyone else, we are in the midst of a huge build-out of physical infrastructure for the internet. AI is driving a lot of that—or at least the expectation of how AI is going to be used. You must have to really understand the physical assets that make the digital world work. Let's talk about one very important connecting asset: fiber optic cables. What are some attributes and challenges of fiber optic cables that you think investors would be surprised to learn about?
John Siegel: I've talked to people about fiber optic cables until they walk away. If you think back to the dawn of the internet, the largest fiber optic cable that people would buy or provision was 864 fibers—864 fiber cores. That was a single cable, and you would basically roll that out and then attach lasers to the end and shoot beams of light, enabling each one of those strands of glass—which are not much larger than a human hair. The flashing of the laser is how the data is transmitted.
When 864-count fiber was put in the ground, people thought there was no way that size of cable would ever be needed again. You put that in the ground and it would last for 30–40 years, basically until the glass degraded.
Now, when people are putting in infrastructure—cables powering AI, cloud instances, really the guts and plumbing of the internet—people are putting in fiber optic cables that have north of 6,900 strands.
What’s happened is, instead of using dense wave division multiplexing—different colors of light over a piece of glass, which is how you transmit data—people have shifted away from that model. That gear can be very expensive. The large hyperscalers have shifted toward just burning fiber. It’s cheaper to put in the ground once the trench is open, and the optronics on it are a lot more reliable and cheaper.
So you've seen this massive oversizing of cables. As you see cloud instances and AI data centers come to play, people are chewing through those cables at a very quick rate. Rather than blue, yellow, red, green all transmitting different types of data or different wavelengths—called waves—the gear that enables that is really expensive versus a simple laser that sends light across a single piece of glass. What people have figured out, especially in dense areas, is it's cheaper to use a low-cost laser and multiple pieces of fiber versus a very expensive laser using multiple wavelengths of light. That’s called burning fiber.
David Snow: Do these less expensive forms of fiber optic simply last not as long? Do they have shorter lifespans? You might have to go back and replace them earlier?
John Siegel: You can replace them earlier. They're easier to upgrade. There's less maintenance and fewer contracts associated with them, especially in areas where there's a lot of density and you're not running extreme lengths—probably greater than 60 kilometers. The large hyperscalers would rather burn glass than use those optronics, because the optronics are really expensive. They use a lot more power, need their own footprint—there are a lot of reasons around that.
David Snow: Was there an underestimation of how much data would be exchanged in the world or transmitted in the world? Was it a failure of human imagination to think about the amount of information we would be sharing?
John Siegel: I wouldn't call it a failure. I think what's happened is as people put infrastructure in place, they found different ways to use it. When the original high-capacity fiber networks were put in place for the internet, no one foresaw this conversation happening—which is video eats up bandwidth.
Think about during the pandemic—the amount of bandwidth that was consumed in a very short period of time. People just couldn't imagine the world would go completely remote and rely on video communications. I'm familiar with one company that spun up a network in Europe three weeks after the pandemic lockdown happened. They deployed a network that was larger than the entire internet in Europe as of 2008, and that was just for one company to support video.
So is it a failure of human imagination? You could say it is. But if you think of the number of devices that are now connected—in your house, the average connected home in the U.S. has over 38 devices pulling data over networks—that data is all being transmitted somewhere, stored somewhere, analyzed, and then coming back to the consumer.
Think of when you hop on your Peloton—you’re getting video coming down, you're sending a huge amount of data back, it's being stored and analyzed. That’s just one device. I don’t think in 2000 we had any idea how we would be using the internet in terms of connectivity.
But we’ve become a lot smarter over the past decade in terms of connected devices. And frankly, this drives the whole cloud and AI craze—how much data is being collected and transmitted between just simple sensors and how we use that.
David Snow: Let’s dwell on fiber optics just a little bit longer before we get to some other topics. Along with everything else, the world is going to need a lot of fiber optic cable as all of these data centers around the world get built out. What physical assets are going to become in high demand—or are already in high demand—as fiber optic cables proliferate?
John Siegel: In terms of physical assets, let's break it into terrestrial and ocean-going. On a terrestrial basis, since the pandemic, you’ve seen an explosion of fiber connectivity to homes—both in North America, Europe, and globally.
The type of gear you need for that, if you’re going underground, would be similar to what people use for fracking—hydraulic directional drills (HDD). That’s underground drilling that puts stuff underground so it doesn’t get messed up by ice storms, hurricanes, etc. Those drills—and if you think about rolling out fiber across the United States—there will be some mix of aerial (i.e. hanging off poles) and stuff that is buried underground.
Those HDD machines are expensive and will certainly be in demand. The different types of cables that are made, and that each one is going to have its own specification depending on the use case, is clearly an opportunity. When you flip to the oceanic cables, what you will see is the boats to deploy that—you’re going to need more boats because there's going to be a lot more cables getting deployed.
And then boats that maintain that as well. Again, these are small, niche industries. But once a cable is laid and something happens to it at the bottom of the ocean, you need to have a specialized boat that will go grab that cable, pull it up, and then splice it. You heard the term splicing.
That's where you connect two pieces of glass to prolong that connection. Lasers are used to actually melt that glass and test that the signal is able to propagate across it. That is specific gear for splicing, and there will be a lot of splicing going on.
The only way that cable will be functional again is if you pull it back up and splice it.
David Snow: There's going to be a lot of splicing going on. Talk briefly about the natural resources necessary to make silicon fiber cables. Silica sand, I guess— is there enough of that resource in the world? Is there going to be a battle for who controls it?
John Siegel: I think we're miles and miles away from the lithium wars with silica. I think the silicon is going to be fine. The silica is going to be fine. These are very simple pieces of infrastructure. It's glass, and then it’s wrapped in plastic, effectively.
You’ve got plenty of the materials for both of those, so you will not see a shortage of fiber optic cable based on supplies. Different story for chips, different story for batteries.
David Snow: All right, so I'm not going to invest in a silica sand REIT, according to your recommendation.
John Siegel: I would think there are probably better ways to spend your money.
David Snow: You are based in Northern Virginia. Northern Virginia is a massive hub for telecom assets, particularly data centers. Why is that? What's the history?
John Siegel: If you think of the way the internet was initially architected, it started off as a network built around the defense industry and educational institutions.
There were two points where that network connected: May East and May West. May East was in Northern Virginia. You had the military complex there and DARPA, which was the precursor to what we know as the internet, really started there. Then you had another connection point where networks would shake hands around Silicon Valley.
What happened in those two regions was—Northern Virginia had a lot of land, a very friendly utility, and Dominion Energy. So you had cheap and plentiful land and power. Ultimately, once someone built a data center near May East, if someone wanted to connect to that, they would just build a data center near it.
You started having this kind of cluster of data centers. Other markets had that happen globally, where either networks came together or there was cheap, available land and power. Silicon Valley is one, Chicago, Dallas here in the States, Frankfurt in Germany, London, Singapore, and historically Hong Kong as well. Where you had big quantities of networks shaking hands, that became the fabric that makes the internet.
You're sharing traffic across these networks. People put infrastructure in those markets, and then you had a force of gravity around them as networks were handshaking in major metropolitan areas. What turns it from a natural place for networks to shake hands into a hyper-growth area are the key ingredients: cheap and abundant land, and cheap and abundant power. That’s why Northern Virginia as a data center market is substantially larger than Silicon Valley—because power is cheaper and land is more available.
David Snow: What are some of the incentives or actions other governments are taking to build out opportunities for their countries to become hubs of digital infrastructure?
John Siegel: It’s an interesting question. The first question is which countries are interested in that. If you roll the camera back 15 years, when people were talking about putting internet infrastructure in place, most folks thought that was a strategic asset they wanted to control as a government or country—an advantage not to be dependent on another country for the fabric of your internet.
A lot of countries pursued that. Fast forward to the world of AI and the power consumed by these chipsets and data centers, and suddenly you’ve got a push-pull inside governments. They want data sovereignty—their key data to stay inside their country—but data centers chew up a ton of power. How are they going to power them?
There’s that push-pull dynamic inside countries and economic zones. If you look at where data center growth is happening now, it comes down to power and available land.
The next piece is having the fabric of internet connectivity—there are actually networks that want to connect to other networks. And finally, the old gotcha of any international investing: rule of law and property rights. To put a big data center in the ground, you’re talking about a billion-dollar asset. Once it’s built out, it could be anywhere between five to twelve billion for a large AI data center. You want to make sure you actually own that.
David Snow: Glad you mentioned the rule of law. You know a lot about right-of-way corridors—largely back in the States—for digital assets. What is that, and how do you get them?
John Siegel: We're going down a deep rabbit hole here. If you think about how one would build a network connecting Chicago to New York, you basically need to draw a straight line between those two cities. You're not going to let someone just dig through your backyard—or if someone wanted to, they’d have to pay you for that right.
So you need a right of way—a way to get from point A to point B. Think of it as contiguous land corridors. In the U.S., a few holders of contiguous land corridors were granted them either by the government or own them by historical precedent.
You’d look to the highway system, railroads, power corridors, and those allowed for pipelines. Pipelines may also use other corridors, like rail lines, or have their own rights of way. Those rights of way—and the ability to connect point A to point B—are critical to building networks.
When the internet or early telecom networks were built, they used these corridors, primarily the railroads. You weren’t dealing with multiple jurisdictions. One railroad could control from New York to Chicago or from L.A. to Phoenix. If you’re trying to build a network from San Francisco to Los Angeles, you’d rather go down the interstate system or the railroad corridor and deal with one party than contract with 50,000 different landowners and negotiate economic agreements with each of them.
David Snow: You mentioned countries seeing internet infrastructure as a national interest. What's happening across Asia where there's been a boom in data center construction and development, in many cases backed by governments? What do you predict over the next 10 years?
John Siegel: I think of Asia in different buckets. Japan is one market, South Korea is another—North Asia. China is a completely different kettle of fish. Then you have Southeast Asia.
In Japan, they were very early adopters of fiber optic tech and data centers. So when cloud computing became a thing in 2007–2008, there was an ecosystem there. The Japanese have aggressively deployed infrastructure. That market operates in its own ecosystem and is well established.
China is controlled by Chinese companies. Western companies are allowed in by invitation only. The chance of them owning or controlling infrastructure is very limited. The Chinese government controls those assets. At the same time, they’re building out massive cloud and fiber deployments. AI and cloud computing are happening at scale in China—but by Chinese companies. Not much of a Western play there.
Southeast Asia is different. Singapore was a strategic decision by the government in the 1970s to be a hub of communications networks. All networks in that region go through Singapore. Think of it as a “trombone” network—sending traffic from the Philippines to Indonesia usually goes through Singapore.
Singapore had a strong hub market. Rule of law was clear. Easy to do business. A very open regulatory environment like Australia, the U.S., or Western Europe. You had many carriers and infrastructure deployed, and the data center ecosystem exploded.
It grew so much that in 2018 or 2019, the Singapore government called a timeout—no more data centers for a few years. The buildings were taking too much real estate and drawing too much power in a space-constrained country.
Then COVID hit, video calls exploded, and data went exponential. Companies started looking for where to put infrastructure to serve the Philippines, Malaysia, Indonesia, Thailand. You saw small deployments in those markets. The question became: where’s the power, where’s the fiber?
That’s an evolving story. But I can tell you the amount of money being spent in historically underinvested areas is increasing. U.S. hyperscalers are going there, so are Chinese firms, especially in places like Indonesia, which are open to both spheres of influence. So companies are putting significant infrastructure in places with land, power, and increasing comfort with the rule of law.
David Snow: Is there a case to be made in a place like Indonesia or Malaysia, where there's a lot of investment taking place, that the presence of data centers and digital infrastructure benefits the economy? Because there's a bit of a tech transfer, there's learning, these places are staffed with locals, they take that knowledge and create more jobs. Is there a virtuous dynamic at play, or is that not the main reason why one would want data centers in their country?
John Siegel: I think there are a couple of reasons you'd want data centers. One, getting back to data sovereignty—some countries are very particular about making sure their data stays inside the country. You think about defense or intelligence information; people are generally hesitant to let that traverse borders. Controlling that infrastructure and analyzing that information is absolutely a boost to the local economy.
The other thing data centers bring is a significant boost to the tax base. As you see these deployments near population centers—and this ties more into AI, where the training models are developed and where the inference models live—you'll find inference models closer to population centers, where people can use them. The training models tend to be where you’ve got cheap power, cheap land, and more control and certainty, especially in terms of government intervention.
The bull case for data centers is you're doing work in local languages with local data and building up that ecosystem. You're also increasing the tax base substantially. The bear case is that once one of these massive data centers pulling down hundreds of megawatts of power is up and running, you don't need a ton of people. They're automated systems. It's not like dropping a car factory that employs tens of thousands. There are jobs in construction and maintenance, but they aren't huge employment hubs. What happens is knowledge workers build around that, leading to a conversation about training and inference models of AI.
David Snow: Let’s touch briefly on AI. Everyone’s talking about artificial intelligence and how it's driving demand for data centers. We've seen estimates that if there are 10,000 data centers now, the world actually needs 30,000. Do you think that's in the right ballpark? What scope of work is needed to get the physical infrastructure where it needs to be?
John Siegel: Let’s dive into your last two words: expected demand. AI as a technology will be transformative. If you look at the deployment of electrical infrastructure in the early 1900s, globally, it automated a lot of manual tasks. You saw a massive rise in productivity, but it wasn’t immediate. That was a multi-decade run. Infrastructure was built, then people found ways to use it.
I think our species will go through a similar path with AI. We're going to automate a lot of mental tasks. We’re just seeing the tip of that.
At Columbia, we look across digital infrastructure and how large enterprises use technology and ride those trends. There’s a huge amount of capacity coming online for AI. You see it in chip-level investment—NVIDIA being the poster child. People are buying tons of chips, putting them in servers, powering up data centers. They're preparing the world for AI.
How enterprises use that technology in everyday life will be a journey. Will supply and demand be perfectly matched? In any technology deployment I’ve seen, there's always the hype cycle. Then comes the question: how fast does that capacity get used? What are the applications?
It’ll be an interesting two or three years. I can tell you the investment is absolutely being made. People are bullish on AI demand. What I’m watching for is large businesses—who spend a lot on tech—starting to actually deploy and lean into AI. They're just beginning that conversation. We’re on a multi-year journey. Yes, we’ll need two to three times the data centers. The real question is: over what time scale?
David Snow: Roll back the camera to the early 2000s and the fiber optic buildout in the U.S. Is there a risk that we're overbuilding digital infrastructure today, or that it becomes obsolete before it’s fully used?
John Siegel: If we roll the camera back, it starts when MCI rolled out long-distance networks using microwave hops, competing with AT&T, which was a monopoly. They pushed a federal case, and Judge Green broke up the AT&T monopoly.
In 1996, the FCC forced local Bell companies to open up their networks to competition. People could lease parts of the network. That prohibited anti-competitive behavior. This unleashed competition just as the internet became a thing—think America Online, dial-up modems.
Networks built for voice traffic were struggling with internet demands. There was massive demand for telecom infrastructure. New companies were founded. At the same time, the internet bubble hit, and capital flooded the market.
By 2000, the dot-com crash was the first pop of the bubble. Many business models proved unsustainable. Then, telecom infrastructure stocks took a hit. There were 36 public competitive telecom companies—35 went bankrupt. At the bubble’s peak, markets rewarded spending. For every dollar in CapEx, you’d get three times that in market cap. People spent borrowed money. The crash hit, and it was brutal.
David Snow: A lot of private equity firms lost money back then.
John Siegel: Yes, and some lost the right to continue investing because their investors didn’t re-up in new funds.
David Snow: Fast forward to today. The story of fiber optics and the dot-com era is different from AI infrastructure, but where do you see risks of overbuilding or obsolescence now?
John Siegel: When we talk about digital infrastructure—especially data centers—it's important to distinguish cloud from AI.
Cloud—Oracle Cloud, Microsoft Azure, AWS—is compute as a utility. Think about all the connected devices and the data they generate. That data needs to be stored and analyzed. AWS, Azure, Oracle, and Google Cloud are the receptacles.
Also, there's been a transition of enterprise workloads to cloud. Think SAP systems. That transition is ongoing and increasing.
Cloud continues to grow and has driven the data center ecosystem for 15 years. AI is different. It uses different chipsets and algorithms to optimize these large data pools. AI still needs cloud—those repositories of data.
What we’re seeing now is servers packed with chips learning from unstructured data, with the goal of generating productive commercial output—rules and behaviors.
Training models are in place. Inference models get pushed out regionally. AI is different from cloud. Commercial applications are still being developed.
I’m bullish on AI, but I’m a qualified bull. I just don’t know the timeline. A friend once said: being early is the same as being wrong. There’s massive AI spending. It’s going to happen. It’ll be transformative. But I can't say when or what the killer app will be—like video was for bandwidth.
David Snow: Looking across digital infrastructure, are any technologies particularly vulnerable to obsolescence? Certain chips? Data center models?
John Siegel: I’m not smart enough to answer that. The question is whether something from left field—like quantum computing—comes and disrupts the compute-store-analyze model developed over the last 15 years.
We’re not seeing infrastructure turned down. We're seeing layers added. It's like powering down a plant—you better have new power already online. We haven’t reached a point of stable compute demand that allows us to shut things off.
David Snow: A big question—maybe you can give a short answer. You mentioned national security. What could competition between nations for digital infrastructure look like? Could it turn nasty?
John Siegel: Shades of gray—there’s jockeying for position, and then there’s conflict.
Around the Olympics, there were protests involving fiber cuts that disrupted providers in France. That’s one kind of threat.
At the extreme, ocean-floor cables carry vast amounts of data. They’re vulnerable and hard to repair. In a conflict, those cables are Achilles heels connecting allies. That’s a real concern.
On the more benign side, nations recognize digital infrastructure is something to have within borders. So, you’ll see more favorable power availability, zoning rules, and rights of way.
If you put up monopolistic barriers to competition, capital and workloads will go elsewhere.
David Snow: I’ve tried to get you to outline the worst-case scenarios in digital infrastructure that might keep investors up at night. And yet, you remain a bull. You said it yourself. Maybe we can close with a question, which is: I'm a qualified bull. Qualified bull. Qualified bull. You know it's going to happen, you just don't know exactly when.
John Siegel: I'm going to be right, it's just a question of which decade.
David Snow: Right? The optimist is always right in the long term, right?
Yeah. Good things will eventually happen. So if you could summarize what makes you excited to be an investor in the data center, digital infrastructure space—what continues to make you excited as you look at the opportunity ahead?
John Siegel: You know, there are very few times in human history where deployment of a specific technology happens at such a rapid, global, and transformative scale. So, a rapid drumbeat of what's happening right now is unlike anything we've seen since the deployment of the electrical grid.
The impact it's having on people in terms of just this conversation—this would have been impossible to do 10 years ago at this quality, at this fidelity. And now we just take it for granted. What that technology allows is just transformational to us as a species. I've got the opportunity to sit here on the front row and watch it happen across a number of geographies. And, you know, I'm a student of history. I just think it's stimulating and feels historic.
So, I think it's a great time to do what I do, and I like doing it. You're actually building something—I can actually go and point to the assets that are getting put in the ground or the building that's getting built and say, “This is going to enable telemedicine. This will enable distance learning.” Now, it also enables a lot of bad stuff too, to be clear, but it's changing the way our species relates to itself. And I just like that.
You know, it sounds dorky, but super cool to be there on the front row watching it happen.
David Snow: Well, John Siegel from Columbia Capital, thanks so much for sharing your expertise with us. It is indeed exciting to hear from you, who has a front row seat at the buildout of digital infrastructure all over the world. So thanks so much for talking to Cool Vector today.
John Siegel: Thank you. Really appreciate the conversation, and thanks for having me on.
TRANSCRIPT END