Carrie Charles (00:01)
Welcome to the first episode of Let's Get Digital. I'm so glad you're here. I'm even more glad to see my first guest, Hunter Newby. We have a great show for you today. But first, I'm going to introduce Hunter. Again, my name is Carrie Charles. For those of you who don't know me, probably a lot of you don't know me, but you will very soon because this is going to be an awesome podcast and I just can't cannot wait.
to launch this. So Hunter Newby is an entrepreneur, investor, conservationist, and the owner of Newby Ventures. His primary field of interest has been network infrastructure. As co-founder, chief strategy officer, and director of Telx, he pioneered the carrier-neutral Meet-Me-Room and the development of carrier hotels and data centers in the US, leading to massive value creation.
and economic development throughout the country. Since the sale of Telx, he continues to be a founder, developer, and investor in the creation of multiple network neutral infrastructure businesses all across North America. Hunter is a recognized authority in network neutral interconnection and data center industry. He has served on multiple company boards, is a published author, public speaker.
and has been quoted in several publications, including the Wall Street Journal, New York Times, and USA Today. He has also been the recipient of numerous awards, including being named the Global Telecoms Business Top 40 Under 40 in 2010. In my view, Hunter, you are an icon. Thank you for being on the show today.
Hunter Newby (01:51)
Thank you. I feel the same way about you. You're an icon too.
Carrie Charles (01:55)
Thank you, Hunter. Thank you. So I'm excited. We have a lot to cover. And hunters, sometimes when we talk, we talk for hours and I learned so much and gosh, we don't necessarily have hours, but we do have a lot to cover, probably going to be, let's say around maybe 40 minutes or so, 45 minutes. So we're going to bounce around a bit. If there is, let's say a subject where
Someone in the audience wants to go a bit deeper, can they reach out to you?
Hunter Newby (02:27)
Yeah, sure. That's a good point. what we have lined up to talk about is pretty wide ranging in digital infrastructure. So let's get digital and we're going to touch on these subjects and move through them pretty quickly because it's a wide range. But if there's anything in there that we touch on that's of interest from anyone that's listening and watching, they can reach out to you and we could do this again and just focus on that one subject.
Carrie Charles (02:54)
Great,
awesome. Well, let's talk first about one of your latest projects, Connected Nation. What is the mission and what problems does it solve?
Hunter Newby (03:06)
Okay, so as you mentioned in the preamble about me, I've done a lot of things nationally in the US in the major markets, carry hotel acquisition, Meet-Me-Room development. And I've always long thought about how to physically distribute neutral meet points throughout the US into tier and markets. So basically into cities that don't have meet me rooms. And that's always been part of what I've been driven to do.
Regardless it goes back many years decades So I met Connected Nation. They're a 501 (c)(3) nonprofit focused on rural broadband So closing the digital divide is their mission and I met them many years ago. It's around about 2017 and As I got to know them and ask them what they do and you know what they're working on. It's a lot of grant writing a lot of
FCC broadband mapping, advocacy, and that sort of thing. They have a lot of programs for children, know, K-12 students. They have a big veterans program to train veterans to get them back into the workforce, which is really important. And I asked them more about the network infrastructure side, like what do they know about how the internet works. And they weren't familiar with carrier hotels or meeting rooms. And I thought,
well that's what's wrong with rural broadband in america the preeminent nonprofit that gets all this funding to solve the digital divide doesn't really know how the internet physically works so i invited him to new york i showed them one of the properties i was a partner in at the time 325 hudson st and i gave him a tour from the basement the manholes all the to the roof and the meeting room and everything and
Long story short was at that moment, they asked me if I thought it would be possible to take that design that I have and bring it to rural America. And I said, well, let's find out. And since then, we've worked together very closely drafting the internet exchange bill. And we worked together on a project in Iowa where we helped reduce the cost of transit for school districts by over 90%, which was shocking.
to everyone there that was involved that didn't believe that it was possible. And then through COVID and coming out of it, we formed a joint venture. So one of my entities and the nonprofit own a Delaware LLC 50/50 and that's called Connected Nation Internet Exchange Points, LLC. So that's our venture together. And we've diligent the entire United States.
partly their data research and partly mine. So for my part, I have an API into what's called PeeringDB, which is a database online that's user generated or contributed content that highlights the location of every internet exchange in the world. So the facility it's in, the switch itself, and then the networks that are on that exchange or physically in that site.
the networks that are on Internet exchanges are referred to as ASNs or Autonomous System Networks. And that's basically a number that is a given address for all the major IP networks and how they find each other on Internet exchanges. So I don't want to get technical, but the partnership that I've formed with Connected Nation, diligence to all the cities in the country, and it's of a population size of
you know, 200,000 or more, distance of greater than 100 miles to the next closest internet exchange, and a K-12 school district, you know, grouping that pays, on average, more than $2 per megabit for transit. And coupled with the data that I have, which is knowing that there's no IX in that city, and then knowing where the next closest IXs are, a little map just appears. All these dots just come out and say,
We're cities that have big populations that have no internet and we have to backhaul to other cities to get to the internet. And we've mapped 125 of these cities. And we're starting a build in Wichita with Wichita State University. So Connected Nation reaches out to the universities, mostly some municipalities, but predominantly big research at Tech Park institutions, research, know, land grant universities and
We secure land from the universities, usually minimum 40 years for free. It could be up to as much as 99 years, dollar a year kind of thing. And then they commit to work with us to launch it. So a lot of these universities have AS numbers, but they don't use them. So all of their student eyeball traffic is sort of not visible on the internet. It's behind a transit provider.
This will be the first time that these universities use their AS on an IX on their own campus. So that's a start, but that's not the end. Then that site becomes an aggregation point for all the local fiber to the home ISBs and all the
networks to aggregate and deliver traffic there instead of having to backhaul it hundreds of miles away. And localizing the internet to these points, it reduces costs, but it also reduces latency.
And the reduction of latency is now more important than ever with inference and generative AI. But we'll get to that later. So that's where I'm going to stop. That's what my CNIXP partnership is about. It's underway. and we got a $5 million grant from the state of Kansas to build the building. So that's important. We're building a building, a modular neutral colo with manholes and everything to house.
an Internet exchange switch and I'll also mention that the Internet exchange switch itself is operated by DE-CIX, the Internet exchange operator from Frankfurt, Germany, who I have a very long-standing relationship with and friendship with and they're really professionally, as far as I'm concerned, the best in the world. They have an appetite for expansion. They've grown into many cities all over the world and they're in several in the US now. They're the largest Internet exchange in many cities in the world.
So they're our partner for this venture.
Carrie Charles (09:58)
So when it comes to bridging the digital divide, I know that that has been the goal of the BEAD funding, the infrastructure funding. So what are your thoughts about the future of BEAD funding, especially with our new administration?
Hunter Newby (10:17)
So great question. That's another one of those that could go really deep. So I'll just show this level about it. BEAD was tremendous. It's a lot of money. A lot of states have already pushed their BEAD plans through and they've received their allocations. I'll just sort of repeat what I've heard from the Connected Nation guys who are really my lobbyist partners, if you will, in DC. Even though they're based in Kentucky, they're in DC a lot. They know all the governors and all the congressmen and all the senators.
So it's great to work with them. The money's federally committed to the states. The states have possession of it. So it would be very hard to take it back, number one. So that's, I'm repeating what I've heard, but that's, you know, believable. So with the government, you never know. But some of the states have already submitted their plans to NTIA and they've been approved and they're deploying the capital now.
So it's underway and it's a shot in the arm for the economy now because a lot of jobs and a lot of labor, a lot of materials will be put to work to build infrastructure. And generally speaking, B was set up for fiber, mostly. There's some other things, but mostly. And then internet exchange points, I make note of this. There's a delineation between the two and IX is an ethernet switch, crudely speaking.
and an IXP is the room or building that it's in. I know there's been some conflation of those terms over the past decade or so, and some people are a little uncomfortable with the delineation. But if an IXP was an IX, then there would be no DE-CIX It would be DE-CIX puh, you know, or whatever. So, just not to belabor that, in the middle mile grant, which was a $1 billion grant,
IX and IXP for the first time ever in the history of broadband grants in United States were actually defined network elements and grant eligible. Now think about that for a second. That was like 20, 22, 23. All the years of broadband grants, there was never a network map ever submitted as part of a grant application that highlighted an IX or an IXP. It was only about fiber. Broadband equates to fiber.
and broadband equates to internet and it was a catch-all, but it's not. Fiber goes to a home, but where does it go to on the other side? And that's what we do. So introducing the concept of the IX is a destination for fiber to get to to localize traffic and reduce cost and latency. So from my research, I was able to see from the PeeringDB API,
where the internet exchanges in the US and the world but in the US are today and it showed me that 13 states in the United States today do not have a single internet exchange in their state. 13. Yeah. And there's another 13 that only have one and it's very sort of fledgling which means that half of the country is not served appropriately and that those states have to backhaul to other states to get onto what we know as the internet where the concentration of networks interconnect in you know
the legacy big cities New York, LA, Chicago, Dallas, Atlanta, Denver. Nothing wrong with those cities, but what about all the other cities? What about Lubbock and Amarillo and what about Wichita and Lexington and all these other places? So the research provides that the grants were created. It was only a billion. No money went from Middle Mile grant to an IX or an IXP, which is unfortunate, even though we applied. The grant we got from Kansas was a state level grant, not
an NTIA, middle mile or grant. But BEAD speaks to fiber. It does reference the middle mile grant definitions. But as I said, there wasn't a standalone application ability to apply for funds just for IX and IXP. That said, the funds that are going to the states that have already deployed and then the ones that are coming for BEAD, for fiber to the home or whatever.
They're going to have a humongous influx of local capacity. And without a local exchange to bring that traffic to, it is only going to serve to cause an increase in the cost of the backhaul bill that that little local ISP that got that grant money is going to have to pay. The problem is the grant money is for capital costs, non-recurring. It is not for operating expenses. So
all the fiber to the home and the increased speed that the end users are going to get, the problem is they're going to use it. And then that's going to end up in the head end of that last mile network, who then has to hand it off to the backhaul provider and the backhaul bill is going to go up 10 X or more. And it's going to be a, it's going to be a financial distress situation. And I've heard that it's not just me positing that I've heard that from the CFOs of the Rolex that are out there that are doing it, that have asked CNIXP
what we do it as i explain it and bring the internet to you to reduce your transit cost basically your back all costs they look at me and say how fast can you bring this to where i am because every great dollar we get we go out of business faster so that's the that's reality whatever anybody else says about this that you think that's reality feels good for the moment when you're spending the money higher big crews and a lot of labor and whatever but then the end of the day
Here's the bad news, the bill comes. And the bill is that you give people a gig at home, they use it. So that's what's going on with B. I think it's going to still be around. I would hope that there'd be a more holistic plan around the deployment of that capital, which would include and incorporate the localization of IP traffic to each state, such that it stays within their borders. And there's reasons for that too. Data.
compliance, privacy, and otherwise, which also ties in a lot to AI. Inference and Gen. AI are going to have, particularly for banking, for healthcare and insurance, they're going to have requirements, sovereign state level requirements, where the data that's generated in the state needs to stay in the state. And the problem is that these people are operating at layer seven and whatnot. They don't know that this layer zero one two infrastructure doesn't exist.
Because most people that operate at the higher layers of the stack don't know how the internet actually works. They just think the internet's everywhere, because I have my phone. But they don't know that that is leaving the state. And there will be, with AI, soon enough, requirements to have data traffic secure, especially things that, you know, from the AI level where it's thinking and it's doing for you, and it's your digital assistant, and you're giving it instructions.
That responsibility, particularly like with a bank, cannot be outsourced like the cloud to Amazon. It has to be done by the bank itself, which means that these banks are going to have to have physical presence in all these states and they're going to need to route bits locally. And when they find out that they can't in 13 states keep the bits local without direct interconnects to every last mile provider, which they don't have, they're going to want to show up and connect to an exchange that already has all the last mile providers connected to it like they would in New York or
you know, LA or San Francisco or whatever. But when you get into some of these other states, it doesn't exist. It needs to be built. We've gotten away with not being, not having to do it as long as we have. And now it has to happen. So we've got to build a consensus that the problem exists, present the solution that's neutral and fair from a pricing standpoint, get those with the need to fund the solution, and then we can deploy the model. Does that make sense?
Carrie Charles (18:31)
Yes, yes it does. Thank you. Let's switch gears to data centers. The power issues keep data centers up at night. Do you see any scalable solutions here? Alright, let's talk about it.
Hunter Newby (18:47)
Yes.
So the power problem for AI, which sort of crept up on everybody, like nobody knew and then all of sudden chat GPT hey, it like, what? And then all these GPUs came out and the power spikes and there's so many things that are broken with the old way, which is only a few years ago, but it was the way it was for a very long time.
that AI has stress tested and broken in certain aspects. when talking about this, you have to qualify which AI you're referring to, because there's different flavors. large language model, machine learning, is what's referred to as training. And that's the stuff that's the macro big power, 100 megawatts to 200 to 500 or whatever. And then there's inference and generative AI.
which is more interacting and generating, which is a different power requirement on a more distributed basis. And the delta between the two in terms of power is also captured and mirrored in the delta between the two relative to latency. So training model AI, which is like the first thing that people sort of heard about and got familiar with, their first impression of it was that it doesn't care about latency, right? It's very latency insensitive.
And that's true, but that's not true for all AI flavors. So a lot of people have this belief that, AI doesn't care about latency, that's not true. AI and the inference in Gen cares a lot about latency. It has to be very low latency in order for it to work. Like a real-time conversation like this between yourself and a digital assistant, you don't want any lag time. You don't want any buffering. You don't want any packet loss. Okay. So there's that. Then...
You could think, large language model, machine learning, training model, AI data centers, the big ones, they basically have chased power. They go to where power is because as we can see, it's been very hard to find power. A lot of the available stuff got snatched up and then land that had power and really macro level things have occurred where certain hyperscalers have literally bought nuclear power plants.
They have no business being in nuclear power business, but they bought the whole power plant because they wanted all the power. And that's not even enough. you turn and you look at inference in GEN, because that has to be distributed in order to effectuate the lower latency, the power requirements or the demands are incrementalized. Albeit, they're still much larger than what we would even consider a big data center a few years ago.
We're talking about 10 megawatts, 20, 30, 40 megawatts altogether for these hyperscalers on inference and gen deployments for GPUs. It's still, that is still less on a distributed basis than the macro training data centers that are 200, 300, 400, 500. Where we've reached the point where you, no one can find that. It's very hard to put all the pieces together. So on sourcing that you've got supply chain issues, you got labor shortage issues, but now
The bigger issues is just if you include power in supply chain, it's still part of the same thing. The power is kind of its own category. The power demands for those data centers have exceeded what the utility grid can provide, what it can deliver, because the utility grid model was not built to support AI on multiple levels. The utility grid model is like a sort of load balancing, Rob Peter to pay Paul shared model where if there's like a big spike here, we take it from there and that triggers brownouts and blackouts.
Well, historical averages and I won't name names, but historical averages for grid interconnects by like major utility companies, companies that manage several states. they're like around 60%. You know, draw, so connect to draw. So if you got a megawatt, you only use 600 KW on average, historical, historical. The training model data centers with GPUs use a hundred percent, a hundred percent of the time. So that blows up their whole.
ratio model, yeah. And then it spikes. So when there's a lot of processing taking place in a certain place, that power draw can exceed what it's supposed to even do. That backflows onto the grid and it creates all sorts of havoc. So we've seen utility companies rescind grid interconnection letters that they issued two, three years ago in projects that were given the go-ahead for construction.
But now those projects are done construction waiting for the power and it won't come because the utility said, whoops, sorry, we oversubscribed. We'll deliver it in six years. Yeah. So we got a lot of white elephants out there with no power that don't generate any cash because the hyperscaler or whoever the off taker was won't pay for it until it's finished and it's not finished without power. So it just sits there. There's a lot of them. Yeah, that's bad. You can't find the power.
And then there's permits and approvals to build, okay, new or anything, substations, but the amount of power that's being requested, it's insane. They have to get approvals for that. In some cases, they have to build whole new power plants. And these utility companies have had projections for years that they look out and they project things seven years, 10 years out. And then AI came along and it blew those things like once all that power now and more, and they're not.
They're not able to fund that. can't really work that fast. They don't want to work that fast. It's a lot of risk. So they just said pencils down. We're not doing anything. We have to review. that shrunk the available market in terms of the land power resource even more. And then now you got NIMBY. NIMBY, I've been saying this for three years. No one would listen. I said, there's some people that are really, really, it's not that they are wrong. They have a point.
But if you don't try to contain that, it's going to spread like a brush fire, and it has. So now a lot of governors are here in big data centers, and they're getting a little leery because the local people come to these town hall meetings, and they protest. And they say, I don't want to see the big power poles. I don't want to big buildings with no windows. They don't do anything for us. There's no long-term jobs. Where's the real economic development from this? And why are we, the ratepayers, subsidizing their distribution lines? Why?
Why should we fund their transmission lines? And they're not wrong, okay? But the model was set up that way and the hyperscalers that got in on it early in places like in Virginia around Ashburn, they got a lot of their capex subsidized by rate payers that aren't necessarily benefiting directly from that. And even if it were the case, they would say, hey, why do I have to subsidize it? Why don't you put your data centers in a different state? I could still get the benefit of the internet, but not have to fund it.
And that's where not in my backyard comes from. I'm sure the tower people are happy that they're not the only one subject in in me anymore. But, yeah, that's again, big subject that could go super deep. wanted to delineate training from inference and gen, the power requirements are definitely different. The headwinds are different for each. but now we have a whole new thing that just occurred, you know, yes, yesterday, which is DeepSeek. And I know we're going to get to that, later, but, that
That is going to change the way people think about how much power needs to be consumed.
Carrie Charles (26:31)
Thank you.
What is it and how's it gonna impact AI, data centers, power? Let's get into it.
Hunter Newby (26:41)
Yeah. So on the training model side for the macro level power, the hyperscalers realized that they can't rely on data center providers anymore because they're not reliable. And it's unfortunate, but the data center providers were being given the responsibility to bundle in the grid power to deliver to the hyperscaler finished product. But they've exceeded their ability, you know, to deliver on these requests commitments, even though the data center provider committed the utility company rescinded.
And then the hyperscaler was like, hey, we said do this by this day. And they were like, we can't make, the data center provider goes, we can't make them do anything. And then the hyperscaler went to the utility company and said, hey, you will do this by this day. And they said, no, we won't. We don't have to do anything. And it's like, So the hyperscalers went to the DoD. Because let's not kid ourselves, okay? AI is a military operation.
this period. Don't think that this is about business or whatever. Everything is military first. I don't care if it's the highways in the United States or the electrical grid or whatever. Everything is military and then it becomes some consumer thing. The internet, DARPA, it was military. mean, everything, the microwave ovens, Teflon, Kevlar, it all came from the military. So the DoD is frustrated that the hyperscalers can't get this done. So hyperscalers go back to them and say,
You fix it. So if you look back to September of last year, the DoD basically brought the hyperscalers to the White House, not to ask permission to tell them this is what we're going to do. They didn't go to Congress. You go to Congress to ask for permission. You go to the White House to say, we're doing this. So they said, we're doing this. And then about 10 days or so before the inauguration, think it was, Biden signed an executive order that opened up federal land for data center development.
expedited permits, basically no permits, and all these other goodies. You can go read it. It was an executive order. It is there.
how and why, because the hypers go to the DoD. The DoD says, look, we have tech that we're going to license to you and only you, the select few. And they're going to build these data centers to run this AI because it's a military operation. You know, get it done. Nothing can stand in the way. I don't care if it's snowing, raining. I don't care if you feel good or whatever, it's happening right now. So, so one of the things that's not happening,
is some investors concept of return on invested capital. That's not standing in the way of this military operation. So this license for power tech, I'm just going to leave it at that, that the DoD is receiving and they're bringing the DOE in on it for cover. Okay. The DOE and the DoD are working together on this, the department of energy to keep FERC and NRC and the PUC away. They say, we got this. It's, it's good.
And then they put it on federal land, which could be existing federal land, or it could be a little principality carved out within a hyperscalers own land that says this is now federal, you know, like Vatican city or something in the middle. And, and then that is a room within a room within a room that the power comes from and no one's allowed in there. That is exactly the same way it's set up on nuclear subs today. Yeah. The nuke sub power is in a room.
That's not the nuke sub. And that power is a license to power the ship. It's also in other ships, but we'll leave that for later. So the hypers get access to that on federal land, expedited, no permits, know, negative compliance. Like we'll tell you what it isn't, but we'll never tell you what it is kind of stuff. And you can imagine that all these sites will sort of be like the ion colliders, you know, like Brookhaven National Labs or Oak Ridge.
where they have a nuclear power plant that powers the collider. Okay. Or as I like to call it area 51. These will be all area 51 data centers. No one's allowed inside. It's powered. What does it do? No one knows. And you're not allowed to ask behind a fence, behind a fence, behind a fence. Okay. That's for them. So you said, what are the solutions to this macro power problem? That's the one that came up with for the haves. The have nots have to continue to fight in this land powered land permit NIMBY.
you know, street fight knives out. All right. Now we go over to the DeepSeek DeepSeek let me preface it by saying this better math and better code always wins. And the future is better math and better code. So if you could take 2 million lines of code and squeeze it down to one, you can execute on all the commands in a faster timeframe using less energy. If you could take a million lines of code,
and squeeze it down to two, you can get a lot done a lot faster with less energy. And that's sort of like what this is. you know, people go, oh, it's China and they're lying and whatever. Read what they actually did. So what they did was they decided to use a different ratio for probability. So they wrote the code so that it doesn't have to be 100 % accurate. 90 % is good enough.
So in saving that 10%, they can get everything done faster. costs less. Now, how much did it really cost them to build it? I don't know. We can sit here and debate it. I think there's some semantics too. Like they say it costs $5 million to build this thing, which is like equivalent to chat GPT which costs $500 million to build. But how much of that cost is just the chips versus the chips plus the data center? I don't know. No one's said that yet.
So the building still costs real money. Land costs, labor costs, materials cost. Copper isn't free. Getting the price of copper down is not the same as getting the cost of a cycle down. But there's a trend here, and I'll just say again, this is a little macro analogy. 20 megabytes of external hard drive memory in 1988 was 700 bucks.
And I know that because I bought one. It was my external hard drive to my Mac SC that I got with two floppy drives when I was a freshman at Drexel University and I had to buy a Mac. And I'll never forget it because $700 is a big deal. I drove to Delaware to buy it to not pay sales tax because it a lot of money. College kid, no money. So that was then. So I looked up the other day, yesterday, for this other call I did yesterday on DeepSeek. What's memory?
What's external hard drive cost today? Two terabytes, okay, for $60. Two terabytes for $60 versus 20 megabytes for $700. The price just, and the performance like this. Here's the other one. In 1996, a meg of IP transit cost $3,000 a month. Today, five cents. Yeah, wholesale, five cents.
3000 to five cents. So to all you people out there that built your revenue models on AI based upon how much you could generate per hour on an H100 or 200, you learned the day that Blackwells came out, a month later, the price got cut by 60%. And now the people that don't understand the algorithms, the math behind the AI, it's the same people that don't really understand how networks work and that they're actually physical things that have a physical home.
And because you lack that comprehension, but you're going with what everyone else is saying. And it's like, oh, it's AI, it's AI. It's like, oh, it's endless demand. That reminds me of the dot com bubble, know, dot com this and dot com that. And that fueled companies that got funded that went public, but then turned around and bought from the telecom companies who spent all this money building fiber and whatever. So when the dot coms popped, that blew back onto the telecom and they all went chapter 11, 22, 33 and M &A.
But we're still running on all that infrastructure that was invested back then, but a lot of people got wiped out. And unfortunately, DeepSeek has shown that just with the change in the algorithm, can reduce the time and cost to produce the same or produce an output that's acceptable. Let's put it that way. It's the range within a range of acceptability. And for certain things, you don't want any, any deviation.
like, don't know, remote surgery or autonomous driving. It's gotta be 100%. But with other things, it could be 90%. Who knows, maybe it could even be 70%. And it could take longer and not have to be as fast or whatever. So they're gonna tweak the machine. And as they do, again, these are numbers that were stated, but it reduced the cost of power consumption to 1 10th. So that made everybody think, what are we building here?
We're building to something that we think was right a year, a day ago to a couple of years ago, we're all this crazy hyper extrapolated, let's go get land and build gigawatts. But if the same out could be, could be done for a 10th, is it going to be made up in demand or like, what does that mean? So it's like, well, there's, there's a lot of fundamental questions on number one, the revenue models.
of those that base their models on like NVIDIA chips that are still current, but the technological obsolescence, the rapid accelerated depreciation, doom. So what does that mean? You know, you could still use them, they still work, but you can't command the same rent for it. So that just reduced your revenue, which now you can't generate enough cash to service the debt that you use to borrow to buy the chips. problem number one, that makes you a risky credit.
Problem number two, you've just signed leases for white space and data centers that are now thinking, crap, are my customers going to file, which then locks up your space and bankruptcy court. You can't get cash. Your tenant might be out of business and you don't even know if their assets are going to be worth buying for 10 cents on the dollar. And is somebody going to come in and pay the same rent or are they going to negotiate it down? And if you're a data center owner and you just invested millions in capex to upgrade for power, now you're not getting a return on that investment. Maybe you got to wait to see.
Beyond that, there are the people that borrowed billions to build these buildings for off-takers that are either A, these risky credits now, shown to be potentially, or the hyperscaler market, which you think is going to go on forever, but it's going to navigate towards the DoD-DOE deal. Because what they're doing on that level, cost of power, the DoD's licensing technology to the hyperscalers, a license, cost plus.
The construction of the building, it's not going to be a data center provider, Carrie. It's not going to be your run of the mill data center providers. It's going to be Halliburton. It's going to be Black and Veatch. It's going to be Skanska. It's going to be construction companies that are going to do it at cost plus. I say cost plus 3%. People go, no, that's crazy. Cost plus 5%. I'm like, OK, fine. Cost plus 5%. You know any private equity people that want to do anything for 5 %? No.
I don't even know any lenders that want to do anything for 5 % right now. 15 maybe. So the hypers are not going to allow anybody to get rich off of them. And they're getting a path to execution. And their cost of power on a license over the next, know, let's say that these, the energy that they're buying a license to, it'll last for decades.
And if you amortize it over those decades, the cost per KW, if you wanted to calculate it, it's lost in the decimal point. So that's never going to come back onto the grid, by the way. It's all going to be consumed locally because it's either on or off. And they're going to grow into it. There's sort of a minimum threshold too for how much these things are in terms of a deployment. It's well north of 300 megawatts per unit. If you need more, you just add more.
I'd like to see these things get incrementalized because it could help a lot in where I'm going, but if not, that's fine. We could still stick with three-phase, 480 grid power and fight it out. the fact that the hypers will get access to something on a license agreement, it's completely skewed. Again, it's military operations, so don't even think it's apples and oranges. It is. But they're not connecting to the grid and paying five cents a KW or whatever the number is.
All everybody else is. So there's another big question. If you're trying to be in the AI business and you don't have access to that, how do you compete? So there's a lot going on. And I'll stop. I'll stop there. I could go further, but I'm going to stop there because we have other things to cover, but hopefully that answered your question. I know I had to add in the DoD thing. The executive order gives it away, so to speak. And the deep seek thing is just a wake up call to.
better math and better code is coming. And if you haven't figured that out, you think your revenue model is going to stay the same for the next two, three, four year? No way. The transit from $3,000 a meg to five cents, just map that. that's felt like a stone. Now, how do you make up for that? And then last thing, AI is an application. Think of it like this. That'll become embedded in everything. And it's kind of like email. It's emails ubiquitous.
But how much do you pay per email? You don't, right? You don't even think about that. So AI, very generic, but AI will be embedded into everything. Will you pay extra for the AI? No. So how do you get something so ubiquitous that it's used by everybody? The cost of embedding it is incremental to your current operations. That's where the cost of it has to get down to. So it can't be.
We have to absorb the cost of billion dollar buildings and billion dollar power and billion. No, we need better math to bring the functionality down to it can fit within what we have. Otherwise, sure. People can go build multi-billion dollar buildings and power plants and everything else and they can lose all their money. It's like, sure. Fine. You want to make it happen a lot faster and it doesn't fit anybody's revenue model or you're competing against people that have better tech that aren't charging for it and you are.
You're dead in the water.
Carrie Charles (41:44)
Right. Okay. Well, it's interesting because this actually affects the question I was going to ask you about workforce potentially because there's major talent issues and labor shortages in the data center space. But will that still be the case with, let's say, said better tech and
you know, things are changing obviously rapidly. But first let me ask you, what roles are right now and will be in the highest demand with the data center space? And I'm curious if you see any solutions for automation and technology and how that would reduce the need for humans or will all these humans, hundreds of thousands of humans still be needed?
in these data centers or to operate the data centers.
Hunter Newby (42:40)
The answer
is a little bit of both. So construction labor is paramount and skilled labor, electricians, know, skilled labor and others. AI can be used to model and plan a more efficient site, which could mean that it takes less time to build and maybe less materials and parts of different things, which might require an aggregate less labor.
either less people or less hours, but you can't replace the people and the time. Today, that's necessary to construct anything, a data center or anything, a bridge, plumbing in your house. There's no AI robot yet today, although Elon's probably working on it. That's going to come in and take out your Len the plumber or whatever. Len, I hope it doesn't happen.
We're Mr. Sparky or whatever. Imagine an AI robot showing up to wire your 200 amp panel in your house. We're far from that. But data centers could be, there could be more intelligence on the front end used to do the design that causes less hours to be needed. That's the only impact that I see. You know, it does feel a little Terminator, Cyberdyne systems, you know, where we're building the machine that eliminates us. But I'll say this.
for all the people that are concerned about that, because we are all science fiction buffs, right? At the end of the day, anyone in this industry, we have to learn how to adapt the technology into our lives, because where we're going without that, we can't compete with whatever's out there, because it uses it we don't, we need to, and this is a transitional period. We're getting acclimated. And yeah, sure, a lot of the things that were done by hand,
by humans can be replaced by machines. But that's any different than saying, I have an iPhone, and on my phone, I have a calculator. And this is in mind, somebody else said this, but somebody was like, whatever, some meme or cartoon about kids in math class and like whatever, sixth grade, and the teacher's like, you have to learn how to do math by hand.
because you won't always have a calculator in your pocket. Well, now we do. When kids have the last laugh and go, I do have a calculator in my pocket. That's basically like using AI. It's cheating. I don't know how to do the math. I just put it the thing and it gives me the answer. We've been using AI in a very rudimentary form for a long time. It's scaling that up and how that helps us. Now, what's nefarious is the code. What is the intent of the code? And I have a whole nother two hours with you on this, but.
Carrie Charles (45:05)
So
Hunter Newby (45:31)
Do I trust these big guys that are getting this deal from the DoD that their intention is pure and good for the good of betterment of humanity? No. Based on what we just went through in the past several years of censorship and everything, I don't trust them. Zuckerberg was like, we had to censor people, but we were made to, but now we're good guys. Dude, no. Come on, man. Like, you're a good guy or you're not a good guy. So the code of the intention behind the code, behind the AIs, is really important because
If you had your calculator and you put in 1 plus 1 and it said 3 and you just blindly believed it, you would think 1 plus 1 equals 3. That's kind of like an AI that has mal-intent. It could give you an answer that's kind of true but skewed-ish. And, you know, that would be like 1 plus 1 equals 2.01. It's 2, close enough. And you're like, all right, fine. I'll let it get away with that. And over time, it works on you until you believe that it's true but you're off. And now it's got you going in the direction it wants you to go in.
That's what have to watch out for with the code and the AI and how it's written. And there's battles on that. know, Elon's talked about it and, you know, open AI and all this other stuff. And I'm not going to, I'm not going to weigh in on that because it's, it's, it's happening, but it's a race. It is a war. Like it's a war because if China can win, you know, they're authoritarian from the top down. Power, not a problem. Cost, not a problem. Everything's free. Build it. Make it happen. And then.
all of a sudden they have the biggest AI and they have all the data, then their intent is what gets implemented. You see, it's really interesting. It's like the TikTok ban. How's that impacting people's minds by what they feed people based on different algorithms of who they are, their age, where they live. And over there, they get sent math problems to do. And over here we get cat videos, you know, that that's, that's real. And
All of that requires infrastructure and here we are digging holes and building buildings to create the prison we all live in. So I'm very cognizant of that and I just keep an eye on it while I still build out network infrastructure because it's a double-edged sword. It's a love-hate, you know. I need it but I don't want it to be used against me. But if I don't build it, I'm not going to have it and I'll be weak anyway so I might as well build it and I got to make sure it works out. It's complicated.
Carrie Charles (47:53)
Boy, is it complicated. But the news sure is interesting lately, right? I am curious your thoughts on one last question about Trump's $500 billion AI venture, Stargate. I read that it's supposed to create hundreds of thousands of jobs, transform the tech industry. What will be the impact of this on digital infrastructure, on
I mean, really on jobs as well, is that true? Are we going to see hundreds of thousands of jobs?
Hunter Newby (48:27)
I hope so. And I believe so. Trump likes to make big statements and get people's attention. And he says things that get headlines and captures their attention. It's like a distraction. you know, here, look at the ball, go get it. You know, like, Gulf of America and we're going to buy Canada and Greenland. I mean, that's all great. And it could all very well be true. We'll see. It's not without precedent. You know, the United States took over Hawaii. It was a sovereign nation. It became a state.
There are other examples. But as it relates to Stargate, first of all, pay attention to the name. Stargate. So Stargate is a movie, a show, but it's also indicative of something else that Trump did in his first term, which is the creation of Space Force. And yeah, that's a big deal. A big, huge deal. Big, big, super big, really big deal. Pay attention to that.
And then behind that is SpaceX. And then behind that is Starlink. Pay attention to all these things. They're all very important things and they're all interrelated things. Stargate's more of like a terrestrial thing that ties to star things like space related things. But it's under the guise of an investment from Masa Son, which is nice and you know, OpenAI and Oracle also great. Bring an Oracle into the fold, trying to bring them up to
hyperscale level. $500 billion is a number. I know Elon said that they don't have it, but whatever. They'll get it. Is it going to require that much capital? Sure. At the rate that money was being raised in all these crazy energy funds and infra funds to do data center AI, AI data center, yeah, sure. But again, when you look at better code, better math,
And then you also look out, the hyperscalers go into construction companies to build these things, not data center providers. When you take out the factor of a 30 % IRR on invested capital for a data center, you make it 3%. A lot of the capital won't be investing in that area. They'll go chase where they can get those returns, because that's how they get paid. Not to say that there isn't a need for investment. It's what the return.
expectations are on that $500 billion. So where can it be deployed to get a return that would satisfy the investor over the next however many years? Because, you can see the landscape is changing like on a daily basis. It's like a mirage. I looked at a thing. had all of all the guys that, you know, investment community, they they like to write big checks. We write big checks. Why? Because they make more money.
More bigger the fees, the bigger the returns or whatever. They're not writing big checks into things like make 3%. They're not. They're by T-bills. how does the capital get deployed and into what? Maybe there's some new energy technology that's going to come out that requires a big capital investment that's going to help AI, for example. Maybe it's an investment into, I don't know what, chip fabrication, know, back on shoring factories that make the stuff here. Trump likes that.
all the above in the resources and materials, and if there could be margins generated that satisfy the investor, then we can have that level of capital deployed from, you know, this particular venture. But for sure, it all needs to get built. We don't have it. What we have is like the starter kit. have long haul fiber from 25 years ago plus that has to change. We've got distributed interconnection facilities, neutral that don't exist that need to.
for cities like Wichita that have the population that justifies it and poor Wichita has to backhaul all the way to Kansas City, Missouri and to Denver every second of the day. Yeah, because the internet's not in Wichita and other cities just like it. So if you have a realistic expectation for return and I think in infrastructure, if you're not ready to be in it for at least 10 years, then don't get in. And then if you're in it for 10 and you like clipping a coupon for a really long time, you're in a great place.
because this infrastructure is needed and it's going to be around just like the infrastructure that's been around already for decades. So I think it's a really great announcement. It's sort of like a moonshot kind of deal. And we'll see how it goes. As they deploy it, I think it'll shift here and there. But the commitment is definitely to data centers and AI and fiber and technology and network and communications. That area, I can't say it enough.
The return expectations versus what reality is, is the rub. That's the shift, right? And then ultimately who is supposed to make the capital investment? You can look at that DoD, well, the executive order linked to federal land, linked to, read the names of the people that were in that meeting. And that'll tell you everything that you need to know. Yeah.
Carrie Charles (53:44)
Hunter, I love our conversations. I have to close and I love comments and suggestions of future shows. I mean, we could talk about so many different subjects. I want to thank you for coming on the show, especially on the first episode of Let's Get Digital.
Hunter Newby (54:04)
My pleasure.
Thanks. An honor to be the first.
Carrie Charles (54:06)
Yes, yes, I appreciate you and I'm sure I will see you soon at Metro Connect. Wonderful. How can you be reached?
Hunter Newby (54:11)
yeah, absolutely.
Just through my website newby-ventures.com. That's fine. Go there. There's a about thing and their contact thing
Carrie Charles (54:23)
Sounds good. Thank you, Hunter. I know I'll see you
Hunter Newby (54:25)
Thanks, Carrie. Okay, bye.
