078 Phillip and Brian, New York Infrastructure and Idaho National Labs

Phillip Koblence (00:00)
But our goal as a digital infrastructure industry is to make power not the bottleneck. It's going to take some work to do that. It's a fairly significant investment in infrastructure in order to make that happen. And I think everyone is kind of of the opinion when you think about it.

But the thing that generates the most of that, especially in a kind of environmentally conscious way, is nuclear. bringing that, as you get to not just the mega kind of one gigawatt Westinghouse kind of deployments, but these kind of small modular reactors and micro reactors, which give you the ability to really deploy digital infrastructure and any large power generating infrastructure anywhere and not allowing power to be the determining

factor right now you see an entire cottage industry which is you powered land or land adjacent to existing power generation facilities the more work that's done with nuclear to try to to get to the point where you create modularity with with with that technology I think the better served you know kind of digital infrastructure will be to be placed not where the power is but where it's most needed.

shit, we're live! Okay.

Brian Smith (02:12)
Thank

Mark Hinaman (02:14)
Okay,

welcome to another episode of the Fire to Fission podcast where we talk about energy dense fuels and how they can better human lives like making electricity ⁓ cheaper for data centers. Today I'm joined by Philip Koblenz and Brian Smith. Philip is with ⁓ NYI, Data Center and Co-location Group. ⁓ He's co-founder and then Brian Smith, Director of Nuclear Reactor Development at INL. Brian, Philip, how you guys doing?

Brian Smith (02:42)
Doing great, thank you.

Mark Hinaman (02:44)
I'm trying to address you guys by name for folks that are just listening out of the way. yeah, yeah, right. So those that aren't aware, they're both bald and I'm going bald. So, you know, it's just a short, short time behind you guys. So I'm trying really hard. got to turn sideways.

Phillip Koblence (02:44)
doing so well, so well, yeah. No, just say, can the bald guy answer?

It's ⁓ right, right. Yeah, yeah. It does not look that short. It looks like you're trying to be trying to be nice. Right. This is these are

Brian Smith (02:59)
That's right.

Phillip Koblence (03:03)
these are East Coast hairlines. You're you have more of a mountain range hairline.

Brian Smith (03:07)
If that's going bald,

Mark Hinaman (03:07)
Exactly.

Brian Smith (03:08)
then I was looking, I was going bald in the best of my teenage years, so I'm not sure.

Phillip Koblence (03:12)
Right. clearly

clearly my chest is also going bald.

Brian Smith (03:17)
You

Mark Hinaman (03:20)
Cool, well Brian, Philip, it took us a while to get this on the calendar, but really excited to chat with you guys.

Phillip Koblence (03:25)
And for anybody wondering why that is, it's because Brian is super, super busy.

Brian Smith (03:31)
⁓ Sorry to do that.

Mark Hinaman (03:32)
Yeah, I don't think Philip rescheduled once. He's, you know.

Phillip Koblence (03:34)
No, would, I'm just always available. I'm just sitting up tethered to my

Mark Hinaman (03:37)
⁓

Phillip Koblence (03:37)
chair.

Mark Hinaman (03:40)
Cool, let's start with Philip. Philip, do want to give a quick background for the audience, kind set the scene? Who are you? What do you work on? What kind of things do do?

Phillip Koblence (03:45)
Who, yeah, I don't know. Let

me get my box of tissues. It all started in 1978 with a dream. So I am a.

an OG in the data center world in the New York metro area. So I started a company called NYI, New York Internet in 1996 ⁓ as a 17 year old, really because it was the coming of age of the internet. And I was fascinated by the really communication capabilities of the internet and how connected it made the world. And as a result of that, my brother and I, crazies that we are, decided to jump feet first into what

whatever that was that included things like web development systems integration getting people online reselling dial up and and eventually DSL and T1 services and things like that and doing all things internet which has evolved to ⁓ you know ⁓ we operate like five floors within 60 Hudson Street ⁓ focused primarily on interconnection taking legacy facilities making them relevant in and today's kind of AI hype

And given the fact that we started in 96, we've seen like the ups and downs of, everything from the dot com burst to, know, the cloud migration and now the cloud repatriation and now this AI boom and then eventually, you know, which is really more focused on large language models, which will eventually bring things back to major cities with the inferencing revolution, which will be like, let's call it AI 2.0 and then quantum and then all of those things. So, ⁓ thankfully at this

point any hair I lose I want to lose and that's what brings us to to your lovely podcast this morning so all of that to say is my name is Phil great to meet you more

Mark Hinaman (05:33)
Nice.

Awesome. Thanks, Phil. The Hudson Street. That's a building in middle of Manhattan. No windows. ⁓

Phillip Koblence (05:45)
That is a building notes got plenty of windows. It is a historical

building It was originally the headquarters for Western Union built in ⁓ the the late 20s And as a result of having you know, all of those tubes to feed the the horse-drawn carriages with the you know ⁓ The telegrams and whatever it is Western Union was doing to deliver their messages ⁓ Eventually the phone companies came in there started putting copper into those tubes and it just de facto became this hub of connectivity and it's a

Mark Hinaman (05:58)
.

Phillip Koblence (06:15)
evolved over time to be, you know, we call it the most interconnected building in New York, which is for sure, but probably the most interconnected building in the country. Every city has its version of 60 Hudson Street. Chicago has 350 Ceramac. ⁓ L.A. has one Wilshire Street. There's the Weston building in Seattle. So there are these buildings that just over time, by virtue of, you know, certain characteristics of the buildings just made them where connectivity companies tended to aggregate.

And 60 Hudson is one of those buildings.

Mark Hinaman (06:48)
Awesome. Fantastic background. Thanks. Brian, yeah, who are you?

Brian Smith (06:54)
Yeah, well,

Brian Smith, the other bald guy here. It's just easier, right? It's...

Phillip Koblence (06:57)
Is that your real name? There's no way that's your real name. Come on. Yeah, my name is John Doe and I work for the government and I'm here to help.

Mark Hinaman (07:00)
It's like the most generic, generic, yeah. I promise,

trust me.

Brian Smith (07:07)
Funny enough, there's another, ⁓ well, he's a ⁓ brilliant person that works for the Nuclear Regulatory Commission named Brian Smith. So for a while when I was still with the federal government, every once in a while, sometimes people would mix us up and we'd get each other's email, but that's happened through my career. Yeah, so Brian Smith, ⁓ today with Idaho National Laboratory.

part of one of the 17 national labs in this country, which is maybe something we could talk a little bit more about for those who, for the uninitiated, those who don't know what the national laboratory system is. It's just a super cool network, very unique ⁓ for the United States, in the world in terms of capability for cutting edge research, development, demonstration across just a whole spectrum of science. Very cool place. Anyway.

⁓ I'm a long time ⁓ fed. I started my career in the federal government in the US Navy, ⁓ active duty and reserve. In fact, I'll retire from my Navy career on January 1st. And interestingly, I got my retirement orders today. Very happy about that. bringing an end to my, ⁓ thank you, thank you. ⁓

Mark Hinaman (08:22)
Congratulations.

Phillip Koblence (08:24)
I think I see a little tear

forming. there a tear forming? we're going to miss you too, Brian. Thank you for your service.

Brian Smith (08:30)
It's been a great career. It is effectively what got me into nuclear though. was assigned, while still on active duty, I was assigned to the US Navy Nuclear Propulsion Program, headquartered in Washington DC. And that was a five-year assignment. you know, the Navy taught me to spell nuclear.

And rather than taking orders and staying on active duty beyond that assignment, I decided to stick around. That's when I transitioned to the reserves and became a ⁓ civilian in that program ⁓ and did a lot of different things in that program, ended up moving into the kind leadership ranks ⁓ in that program. From there, moved on to the National Nuclear Security Administration.

kind of broadened my defense nuclear horizon to include weapons and ⁓ nuclear non-proliferation. And from there went over to the House of Representatives. I was on the professional staff and the appropriations committee ⁓ as the appropriator for the Department of Energy to all of its nuclear programs, both defense

and non-defense and then some other cats and dogs, the Nuclear Regulatory Commission, the Defense Nuclear Facility Safety Board, Nuclear Waste Technical Review Board, a few other things. That was a good time on the Hill. When I left there, I went back to the Department of Energy in the Office of Nuclear Energy. That's the part of Department of Energy that does all of the non-defense or commercial.

nuclear stuff. My last role there, I was acting as the deputy assistant secretary for nuclear reactors. So kind of had purview over all of those programs that are seeking to advance nuclear deployments around the country. A lot of things you read about in the paper, right, or those public-private partnerships that we were doing in the federal government, that was kind of part of my portfolio. And last January.

Phillip Koblence (10:26)
What the hell is a paper? Sorry.

Brian Smith (10:28)
I retired from DOE and joined Idaho National Laboratory. Kind of a short leap in as far as the national labs are owned by the Department of Energy, operated by contractors. So the Office of Nuclear Energy owns Idaho National Laboratory, or kind of owns the work, the work portfolio. It's the primary customer of the laboratory. And so by making that leap, it was really a very short.

leap given that I was already very familiar with the programs being done at the laboratory. So it's been fun now for nearly a year since joining the lab. Just really enjoying keeping this going. And I guess the last comment I'll make, you know, relative to working with Phil here, which has been great, the data center industry, the stuff that Philip's doing to

to ⁓ deploy AI at scale, which is kind of what's on my mind, stuff has to be powered. And so over the last couple of few years, I started leaning into the data center industry as one of the key market catalysts that we needed in nuclear to go deploy those reactors. I wanted to look for something beyond just a standard.

utility purchase, own and operate model for nuclear. Let's find those other off takers. the debut of ChatGPT, frankly, ⁓ in November of 2022 was something that was kind of a seminal moment. Those low growth projections shot up and that told me that was an opportunity for nuclear. And sure enough, the marriage is what it is and here we are to talk about it.

Phillip Koblence (12:09)
Can we agree just as an audience that number one, this would be a fantastic podcast to have a drinking game with the word nuclear ⁓ on ⁓ number two and data centers, but really nuclear. mean, just just in terms of Brian's ⁓ initial introduction and Brian Smith is just far more credible and capable than I am with that introduction. So. ⁓

Mark Hinaman (12:10)
you

And data centers, yeah.

Brian Smith (12:29)
Yeah

Mark Hinaman (12:30)
than both of us. Yeah, yeah. I don't understand how

you retire from an organization and then go work for the same organization. Like, what?

Brian Smith (12:39)
Yeah,

that is a ⁓ mere nuance. It really does get nuanced. The laboratories are owned by the Department of Energy but operated by contractors. ⁓ So each laboratory is contracted by the Department of Energy for a team, in INL's case.

over 6,000 scientists and engineers, all of whom work for a private entity that is hired to operate the labs. So the DOE owns the land, the DOE owns the equipment, ⁓ the microscopes, the hot cells, the glove boxes, but they hire a private company of people to go in and operate it. So I merely joined that contractor workforce.

Mark Hinaman (13:20)
Gotcha, gotcha, okay. Okay, so yeah, I mean, Brian, you alluded to it a little bit, but why are we talking together between Brian and Phil? This all came together.

Brian Smith (13:30)
Yeah, well, maybe I'll be short and Phil, I'll turn to you because I did. You're right, I already kind of alluded to it. I'll say the AI race, but I...

I tend not to use that word because I'm not exactly sure. A race by definition has a ⁓ finish line, right? I don't know what the end of the AI race looks like. I don't know how you win it. I know what the space race was about. They planted a flag on the moon, right? And I'm a triathlete. I know what the end of a triathlon looks like and I'm very glad to see it. so let's say that it's what we're trying to achieve is dominance in AI in the world because those who sort of own

own AI or dominate AI ⁓ have a special place in the world. And I think as we look across the world landscape and those who would seek to ⁓ compete with us, Russia and China come to mind. China is doing certainly big things in AI. Much of that work that they're doing in AI is underpinned by their energy ambitions, ⁓ least of which is their nuclear ambitions. Today, we, the US, produce quite a bit more nuclear energy than

China, about twice as much. We put almost 100 gigawatts out there of nuclear energy, about a fifth of what is produced in the US. China's about half that, but China's got 33 large reactors under construction right now, and they're going to overtake us in just a few years in terms of nuclear energy production.

And so when you consider what a country can achieve when a country dominates the energy landscape, it is that sort of energy race that I'm really interested in. And then all the things that layer on top of that to include the development of very sophisticated AI tools, that's a big deal. It's a big deal for national security. It's a big deal for our well-being as Americans. And that's where I think nuclear kind of fits in.

Phillip Koblence (15:38)
That was your short answer.

Brian Smith (15:39)
Yeah, so don't get me going on the long one.

Phillip Koblence (15:39)
Geez, Louise, geez, Louise. my God,

don't pull that string. Don't pull that string. Look, ultimately, when ⁓ I started in the data center industry, which is at the beginning of it, it was, to operate a one megawatt site was thought of to be like, that's big. You don't need much more. When would you ever need much more than that for a kind of traditional compute workload in the late 90s, early 2000s? And that was the case really until

maybe the cloud came of age and you had these kind of large aggregation of compute farms that AWS put together, certainly in Northern Virginia and Microsoft with Azure and Google Compute. And you sort of the birth of the hyperscalers, you got to the point where people started talking about, you know, 10 or 20 megawatt sites. And I think what AI really did was it shined a light on the fact that the limiting factor with

respect to like compute workloads with respect to how technology is evolving was no longer space or connectivity which had been you know the the the big issues of you know kind of web 1.0 2.0 and and certain not not 3.0 I 3.0 is where you started getting into cryptocurrency and blockchains and and things like that where the limiting factor started being how much power particularly how much power density you can achieve at these

sites. And that's when we started as an industry making requests of the grids that they could not keep up with. you know, when you you think of and I want to get into the concept of, you know, the environmental benefits of, you know, kind of green power versus dirty power, because we're at the stage when you're in a race where those are nice to haves, but those are not like the focus of of of of the industry.

as much as the industry would like to claim that they are. But once you started talking about ⁓ power being the bottleneck, then we started having to talk about what the ways of generating vast amounts of power quickly are.

⁓ And you know, you have the traditional ones, know, oil, coal, of course, the efficient ones, hydro and wind ⁓ and solar. But the ones where you're going to be able to get the most kind of ⁓ dense capacity are the kind of small modular reactors, the micro reactors, and of course, you know, the large kind of nuclear reactor base. But I still think, you know, one of the reasons why I also assume that Brian doesn't want to talk about the word race is because race

suggests that, you know, speed is the ultimate ⁓ winner. I think when it comes to nuclear is likely the only answer that's going to that's going to solve the problem. It's just not going to solve the problem as quickly as I think a lot of folks want ⁓ to believe initially until you get to the point where a lot of the things that Brian is developing at I &L are starting to be, you know, kind of manufactured quickly. So all of that to say is my name is Phil. Great to meet you guys.

Mark Hinaman (18:56)
I love it, Phil. Phil, what is gonna win that race? ⁓

Phillip Koblence (18:59)
⁓ I think ⁓ the answer is unclear. I my hope is that what wins that race is, you know, the kind of focus on the best underlying use cases for AI ⁓ and not just AI for the sake of AI. And I think you're seeing a lot of that. And I think that that's the problem with just the kind of the corporate mindset that powers a lot of these companies. think somewhere in the area of

80 plus percent of our increase in GDP over the last two years has been based on the back of these almost absurdly growing AI-ish companies and hyperscalers that are kind of building the foundation of where these GPUs get deployed. But I think ultimately what's going to win that race is the best underlying fundamental use case for

this technology, which I'm not sure we really see yet. I mean, you have a lot of, you know, kind of data sets and data analysis. I think, you know, the world has moved to the point where we are the data, you know, the edge, you will, like everybody's household, everyone's person. I'm wearing an order ring. have an Apple watch on. I I am generating data at a crazy pace and that doesn't even take into account my refrigerator that tells me when I need milk. You're going to need the ability to kind of analyze all that data. But ultimately,

Mark Hinaman (20:19)
you

Phillip Koblence (20:27)
the exciting use cases for me are the ones that are like, you know, cancer screening and early detection and trying to really capture that data and work for the benefit of society so we can all collectively as a society work smarter, not harder and focus on some of the fundamental areas of our kind of socioeconomic condition that that, you know, are not just tied to doing repeatable things over and over again and trying to figure out what what what makes, you know, the kind of the economy.

Mark Hinaman (20:43)
you

Phillip Koblence (20:58)
I'm not running for office, by the way. So I want to I don't I don't want to turn this into a Bernie Sanders rally of. Right. I'm Brian. Brian. Brian's got I don't know. I don't know if you guys can. I don't know if you guys see this. He's got many diplomas on his wall.

Mark Hinaman (21:00)
I don't think anyone on this call is running for office. I don't know, Brian's got this politician smile. He might, yeah.

Brian Smith (21:14)
Hahaha!

Mark Hinaman (21:14)
Yeah, yeah.

So, okay, so race is not maybe, yeah, the right framing of it, but ⁓ the whole world, and I'll put this in air quotes, knows that AI needs more power. There's, I think, some nuance around that, but the timing of when nuclear can respond to that, ⁓ I think, is kind of up in the air, right? And I would say even within the nuclear community, a little uncertain. ⁓

Brian.

Phillip Koblence (21:44)
I think it's only uncertain with

those people that are going to lie to you. And Brian's not one of those people.

Mark Hinaman (21:50)
Well, let's double click on them little bit. What do we think the future could look like with timing for new reactors coming online? then, I mean, how much of this AI demand could they meet?

Brian Smith (22:05)
Yeah, great question. And I'm glad you're opening it up to the timeline piece. I do think that that's a critical thing to talk about. In some ways to level expectations, because to your point, you could look out there for information. You'll find all kinds of prognostications, a lot of them driven by the industry, the nuclear industry itself.

And so I do, I appreciate being able to enter conversations like this and be something of an honest broker, you know, because I don't work for any one of those vendors and I don't get commissions on sales of reactors. ⁓ But one thing I want to point out, you know, because some people do ask me when I go around to conferences or I speak on this topic, they say, hey, you know, Brian, when will nuclear be real? ⁓ Because they're asking about the timeline, but that's a good chance. Right.

Mark Hinaman (22:57)
It's super real already. It's already real.

Like, yeah.

Brian Smith (23:00)
Exactly. We've

got 94 reactors operating in this country producing almost a fifth of the electricity. So it's real. And by the way, there is a lot that we can do and that we are doing to ⁓ increase the capacity of the existing fleet. That's a process called upgrades. We have an existing, there's a program, a fundamental program at Idaho National Laboratory.

⁓ that is specifically designed to provide underpinning research and development that can help the Nuclear Regulatory Commission and industry operate existing reactors. So if you've got a reactor today operating, you know, let's say it's putting out a thousand megawatts, ⁓ what kind of ⁓ research can we do to support the case for potentially a license amendment?

that allows that reactor to produce 1,100 megawatts. That's a real thing that has occurred for decades. As we sharpen the pencils, we figure out that there are ways to do that. ⁓ There are also new types of fuel that we can test at Idaho National Laboratory and then can get deployed to the fleet to make the fleet operate in a more efficient manner at a higher capacity. ⁓

So that's important. Upgrades remain important. We have a plan right now, this decade, to add gigawatts of power within the existing fleet. That's without building one single new reactor. So good news there, to the extent that we're talking about a power demand ⁓ of filling a gap measured in gigawatts. So I just want to highlight that piece. ⁓ Restarts is another one.

Right, we do have a handful of reactors that were shut down recently. They were shut down in a way, we don't have to go into the technical details, but they were shut down in a way that makes it fairly straightforward to put fuel back in them and turn them back on. So that's good news because that can quickly bring, again, you know, for each large reactor representing something like about a gigawatt on the grid.

That's good news because you can just refuel say Three Mile Island, the TMI Unit 1 that was shut down in 2019 merely for economic reasons. When Microsoft comes along as a partner, again, kind of that post AI, post chat GPT coming out, you've got these hyperscalers with deep pockets out there just looking for power.

TMI shut down in 2019 for economic reasons. Constellation decided it just wasn't economically viable to remain as an operating asset. It's got a lot of years, it's got decades left to go, but if it's not economically viable, be Microsoft comes along, changes that equation. We're gonna turn back on TMI One. It's the crane clean energy center. That's good news. That's gonna put another gig plus on the grid. ⁓ Palisades, same thing. Dwayne Arnold, same thing. So that's exciting stuff. That's all large scale.

and things that are happening near term. yeah, nuclear is real and nuclear continues to ⁓ get even more exciting with what we've already got. As far as SMRs and micros, ⁓ we're in an exciting time this decade at Idaho National Laboratory and in some other spots around the country.

There are ⁓ small modular reactors and micro reactors. You can kind of group those in, let's call them advanced reactors that are going to start turning on their demonstrations or their pilots, you know, that first reactor. ⁓ Several of them at Idaho National Lab, but there's one by X Energy down in Texas. There's one by TerraPower out in ⁓ Wyoming. Kairos is an exciting company and turned on their reactor in Tennessee. So, you know, it's not

to Idaho National Laboratory, even though we're partnering with virtually all of those developers. ⁓

But they all have plans to turn on reactors. We're citing at INL, Oklo, ⁓ Allo, ⁓ Antares. ⁓ Radiant is putting their microreactor demonstration into a test bed that Idaho National Laboratory has. So all of these things that I'm describing, these are all reactors that are going to turn on this decade. That's important. Now, that's not putting a reactor on everybody's coffee table as soon as those first pilot

that reactors turn on, but it is a critical step to commercializing. You have to turn on the first one before you can turn on the next hundred. And so by partnering with these companies, helping refine their designs, ⁓ assist them through the NRC licensing process, which is critical for commercial operations, we can de-risk the programs.

That can help incentivize investments into those developers so that they can capitalize on turning on the first reactor and quickly move into commercial operations, commercialized scaling those deployments.

Mark Hinaman (28:18)
Gotcha. Phil, were you at the event that INL hosted? I'd be remiss if we didn't mention that.

Phillip Koblence (28:25)
I was not because

Brian must have lost my address when making invitations. Yeah, it's pretty rude. Pretty, pretty rude. And I was just hoping, I was hoping at least I would be able to grow a third eyeball by seeing what you guys are doing over there.

Mark Hinaman (28:29)
Brian didn't invite you directly, like personally, that was kind of rude of him, man.

Brian Smith (28:30)
Hahaha

Yeah, I see.

Mark Hinaman (28:43)
Brian, give us a quick overview of the event that you guys hosted earlier this year, if you don't mind.

Brian Smith (28:47)
sure. Yeah, thanks.

Yeah, we had the second data center workshop at INL. We did the first one in October of 2024. Did the second iteration. Might have slipped into the first week in November. I see there in October, November, but brought together a lot of folks across the data center industry and not data centers exclusively, ⁓ a lot of project developers.

A couple of utilities, just entities interested in this data center building boom, brought them out to Idaho National Laboratory, ⁓ did a tour of the site, mind you, INL.

very cool place, ⁓ lots of history. It's worth Googling and if you're an American, it's something to be very, very proud of, but it's 890 square miles. So it's a huge place with kind of little fence lines all around the property doing some really exciting nuclear research, development and demonstrations. So did a tour of the site and then did a couple days of discussions. Essentially a lot of

researchers coming in and kind of presenting their ongoing research programs ⁓ that have relevance to the digital infrastructure community. And the goal was to create something of a two-way conversation. Ideally, it wasn't just an INL researcher hurting.

the heads of the attendees with a whole bunch of technical stuff and then walking out the door. It was more about presenting these ongoing research programs and soliciting input from the data center community on what is it that is of greatest relevance to you? What should we be doing different? ⁓ What can we be doing in the future that will help ⁓ accelerate this nexus of AI and nuclear? So I think it was a very productive event.

I'm sure we're going to do another one. We're also creating what we're calling a data center playbook ⁓ to impart, kind of coalesce, to bring together a lot of the lessons learned from these workshops and also just to put into one place a lot of that ongoing research, something that can be useful to the community at large. So that was the workshop and its goals.

Mark Hinaman (31:19)
Gotcha. That's really helpful. Phil, do you guys just have stuff in New York? Are you looking at elsewhere? mean the New York internet, right? That is the acronym.

Phillip Koblence (31:27)
Yeah.

I mean, NYI

has ⁓ stuff in New York and outside Chicago. I, as a kind of, let's call it data center OG, ⁓ wear many hats with ⁓ a couple different platforms, consulting and otherwise. So I have a couple different platforms that put me ⁓ as kind of a global digital infrastructure ⁓ kind of Renaissance man. So to the extent that I am part of lot of conversations

for the deployments of hundreds of megawatts of capacity globally and trying to connect the dots, which is really more about the relationships that I've built over the course of the nearly 30 years that I've been doing this. So less specific to NY and more specific to kind of my own place within the ecosystem, which hopefully will lead me to get an invitation to the 2026 version of Brian's Little Party that he's not inviting me to.

Mark Hinaman (32:25)
That was

gonna be my follow-up question. Yeah, you gotta go out to INL and see what it's all about. How do you guys know each other before this call? I guess I should have seemed transparent from the emails, but how did you guys get connected?

Phillip Koblence (32:27)
Right? Right.

Brian Smith (32:27)
Indeed.

Phillip Koblence (32:41)
⁓

I, so I started a, I co-founded, ⁓ a foundation called Nomad Futurist. which really the, the, the initially it was, it was a podcast which grew into a 501 C3 focused on trying to get the next generation excited about looking at digital infrastructure, ⁓ as a potential career path. So not an entry point just in terms of programming or, or computer engineering, but really trying to kind of integrate digital infrastructure into existing

STEAM curricula because in order to really keep pace with the development of all of these kind of technology platforms, we need people fundamentally to do all the work. We have an industry that is largely composed of the same folks ⁓ that were there when I started 30 years ago and that has not necessarily kept pace ⁓ with the evolving landscape and how important we are to the other elements of technology.

process,

⁓ you know, we had one of our advisory board members also, you know, worked for, you know, a couple of other other companies and ⁓ and Brian as a kind of nuclear expert started talking at a number of events that we were speaking ⁓ at as well. And then when I see anybody that could be a doppelganger, except a slightly, slightly better and more accomplished looking ⁓ doppelganger, then I'm like, Brian, I'm Phil. Can you join me?

to make me seem smart on on various things and and I don't you know it seems like we've known each other for decades I think I think we probably said the first word to each other within the last certainly ten months ⁓ and and and then we've seen each other at a number of these events ⁓ and and he's been trying to get away from me ever since

Brian Smith (34:23)
Yeah.

Mark Hinaman (34:31)
Brian, who's the other co-host at the event? mean, he kind of is also a little doppelganger. Yeah. Yeah. ⁓

Brian Smith (34:37)
You're thinking of that person that Phil just described, Mark

Guskov was there, yeah.

Phillip Koblence (34:45)
Mark Gusikoff, yet another

one. yet another. And Mark is on the No Matter Futurist Advisory Board as well. ⁓ we try to make it a safe space for bald middle-aged white men to come together and talk about the future of industry. What could possibly go wrong?

Mark Hinaman (34:49)
Yeah.

Brian Smith (34:59)
Hahaha

Mark Hinaman (35:04)
Fantastic. I love it.

Brian Smith (35:05)
Right?

Mark Hinaman (35:07)
yeah, we talk about nuclear, we talk about natural gas on this podcast. Like, how do we see data centers and nuclear getting married? And not married, but you know, working together to

Phillip Koblence (35:21)
Yeah, look,

I think from the data center perspective, we need to find the best ways to most efficiently generate ⁓ a lot of power. So the power element of it ⁓ is no longer a bottleneck. We'll always have bottlenecks, right? So it was space and connectivity that was the bottleneck. Now it's power that's the bottleneck. Presumably nuclear is one of the ways in which it's not the only way. are lots of technologies happening with respect to hydrogen power generation, leveraging natural gas

existing pipelines, certainly to create bridging power to get a lot of these deployments in place and a lot of interesting ways to leverage different power, fuel mechanisms to generate the type of power that we need. But our goal as a digital infrastructure industry is to make power not the bottleneck. It's going to take some work to do that. It's a fairly significant investment in infrastructure in order to make that happen. And I think everyone is kind of of the opinion when you think about it.

But the thing that generates the most of that, especially in a kind of environmentally conscious way, is nuclear. bringing that, as you get to not just the mega kind of one gigawatt Westinghouse kind of deployments, but these kind of small modular reactors and micro reactors, which give you the ability to really deploy digital infrastructure and any large power generating infrastructure anywhere and not allowing power to be the determining

factor right now you see an entire cottage industry which is you powered land or land adjacent to existing power generation facilities the more work that's done with nuclear ⁓ to try to to get to the point where you create modularity ⁓ with with with that technology ⁓ I think the better served you know kind of digital infrastructure will be to be placed not where the power is but where it's most needed. Brian.

Brian Smith (37:18)
Yeah, no, that's all very well said. I would just add, Mark, one exciting waypoint that we're going to see in this journey is a program that the Department of Energy rolled out this year and the first awards will be made or selections will be made ⁓ in January, I expect. This is a program to site ⁓ data centers on federal lands.

co-located with, ⁓ among energy generating, advanced energy generating capabilities, co-located with nuclear reactors. And so there's four federal sites thus far that have been selected for that program. Idaho National Lab is one of them. The others are ⁓ environmental management, Department of Energy cleanup site in Kentucky called Paducah, ⁓ one in Oak Ridge in Tennessee, and then Savannah River out in South Carolina.

⁓ Idaho National Lab was the first of the four, so there'll be the first selections made, as I mentioned, just next month. ⁓ And those selections will identify teams that responded to requests for applications. ⁓ These teams would be data center developers, ⁓ nuclear reactor developers, and then anyone else that is involved in deploying data center and nuclear projects.

And those teams will be selected to go actually build a data center and data centers. expect there'll be multiple teams. We'll see what Department of Energy selects. But build a data center co-located with the nuclear reactor. And I think that that's going to be kind of the first ⁓ of a wave of data centers co-located with nuclear reactors. And I love the way that Philip mentions it. As much as I tout how...

the importance of large-scale nuclear. And we talked about upgrades, we talked about restarts, building new large-scale reactors. That's a priority right now for the White House. ⁓ But small modular reactors and micro reactors, they still fit into this because there are some ⁓ exciting opportunities, places where you can put those things, maybe where you couldn't fit a large-scale AP1000 reactor. ⁓

I think you'll see the Department of Energy is gonna be looking at some of those smaller reactor sizes under this program to site data centers on federal land. ⁓ But I think it's going to be the push industry needs to partner with the federal government. The federal government's saying, we have lots of land. I mentioned 890 square miles at Idaho National Laboratory. I got a lot of land.

So let's build a data center. Let's put a nuclear reactor there. Let's have the federal government as a willing partner ⁓ streamlining the environmental ⁓ permitting process, site characterization, et cetera, everything that you would need to go deploy that project. ⁓ That's an exciting thing. And again, for your listeners, know, that's February or January, I anticipate the Department of Energy making their ⁓ selections.

⁓ on the Idaho one with those other three sites I named to follow in succession. So ⁓ it is, this train is moving very quickly.

Mark Hinaman (40:46)
Okay, I've got some follow-up questions on that idea. I and L sounds awesome, right? Like let's just put a nuclear reactor next to a data center. electricity doesn't have to go as far, you know, it'll just power the data center. But when I was there, part of the discussion was like, okay, but some data centers require high reliability and...

Brian Smith (41:08)
Mm-hmm.

Mark Hinaman (41:10)
New technology is notorious for being somewhat unreliable. nuclear, might have to this backup power source. ⁓ Many people have recommended natural gas as a backup power source. There's no natural gas pipeline to the INL site. How then do we provide backup power for nuclear reactors at the site?

And if the answer is well, we can have lower reliability, like what kind of data centers can work intermittently. So Phil, you might be able to comment a little bit on that.

Phillip Koblence (41:48)
I can go with the kind of data centers that. Right. Yeah, so if you're talking about the kind of, ⁓ you know, the.

Mark Hinaman (41:52)
What is the product that is using these data centers, right?

Phillip Koblence (42:01)
the AI factory type data centers that are essentially these large language model farms for the deployment of thousands upon thousands of Nvidia GPUs, whether it's the H100s or B300s or Blackwells or whatever. And that's gonna continue to change, that profile is gonna change, that power requirement, it sounds like, will continue to increase where I come from a generation where five kilowatts in Iraq should really be enough

And you have Jensen Huang at GTC suggesting that one megawatt of Iraq is a realistic expectation, which sounds insane to me. ⁓ But in reality, those large language model farms ⁓ do not necessarily have the same kind of super low failure tolerance that, say, a traditional mission critical workload ⁓ would have. Yes, you want them to

beyond yes you have a lot of equipment and you don't want to just necessarily pull the plug on them so you know in in I think the range of let's call it uptime tolerance for ⁓ for workloads that that are

within kind of, you know, high power data center environment is on the low end of the range. You have a lot of these these these farms that are now being converted into GPU farms, which are like the crypto mining operations, which had no, you know, we would call them a tier zero from an uptime Institute rating methodology. They would just connect. Right. It just you wouldn't even have batteries. Right. Right. You would just connect straight to the utility. Remarkably, it's a similar

Mark Hinaman (43:35)
can turn off anytime. Be mining when... yeah.

Phillip Koblence (43:45)
Kind of hardware use cases. There were GPUs that were mining for cryptocurrency. So a lot of those ⁓ sites, you know There are companies like iron that you may have heard of that that are using former crypto mining sites that are that are well suited to convert into AI sites that do require some level of power conditioning maybe UPS is so you can you can maintain some level of organization of Power conditioning etc, but don't necessarily need to have you know tier 4 to an

methodologies that you would have a critical say like the Morgan Stanley or JP Morgan data center that is operating, ⁓ you know, these hospital grade ⁓ fueling contracts because, you know, the entire like financial market requires this infrastructure be up and running. So there is a range and for the most part, the AI factories are not at the top end of that range. They need more reliability than say a crypto

But I don't necessarily know that they need, know, full tolerant concurrent maintainable ⁓ kind of power ⁓ infrastructure. Brian?

Mark Hinaman (44:54)
And

I'm ignorant, Phil, like what do you know the breakdown and like total data center power demand, like how much of it has to be the super high reliability versus these AI factories? ⁓

Phillip Koblence (45:06)
I would suggest

that ⁓ just in, and I have no actual scientific evidence to support this, right? So I'm gonna, but if you are, I'm sure there is, I'm sure if we went to InfrastructureMasons or we went to IDCA, ⁓ or just ask, or screw that, just ask Phil.

Mark Hinaman (45:13)
We don't know actually, but is there research papers like resource or?

Let's ask chat. Or grok. Yeah. We'll just ask.

Phillip Koblence (45:27)
Just ask Phil. And Phil is going to tell you that ultimately the mission critical business operations is a fraction of the overall kind of deployment, particularly in this new ⁓ AI.

kind of a hype cycle ⁓ element. I'd say like the bulk, bulk, the vast majority of ⁓ the workloads that are coming online in the last, let's call it, two, three years that are 100 megawatts and above are for these kind of AI factory large language model deployments that don't have that same tier four, even tier three requirement for concurrent maintainability.

Mark Hinaman (45:47)
so significantly fewer.

Phillip Koblence (46:12)
a fraction of the ultimate workloads, certainly in the grand scheme of how much data center power is deployed, ⁓ which is to say, you know, several hundred megawatts, know, gigawatt, a low end number of gigawatts of the overall dozens of gigawatts of data center capacity that are deployed that are those critical workloads that can't go down. But even that has started shifting from a kind of vertical like this site can't go down methodology to a horizontal kind of regionally

distributed ⁓ architecture where you're never reliant on one site. If you look at the way Google operates, if Ireland goes down, it automatically just pivots those workloads ⁓ to other regions that are up. So you are having a much more horizontal strategy of systems architecture than you had in the past where one site had to, you know, all eggs in one basket.

Brian Smith (47:06)
The Stargate project, the way it's being deployed in Texas is kind of an example of that. Stargate, that's all about ⁓ AI and learning. The Crusoe deployment is for 100 megawatt buildings essentially to go around a hub. Stargate isn't putting in full redundancy on every 100 megawatt building deployed. They're actually doing

⁓ generators, three megawatts of generators per 100 megawatt deployed. essentially their view is if we lose power from the grid, we'll be okay. The learning process essentially pauses. The three megawatts of diesel backup, generator backup is just to kind of keep the core processing, know, kind of remembering where the learning was taking place when it paused.

And then when the lights turn back on, everything stands back up. So what's that? ⁓

Phillip Koblence (48:03)
There's your answer. 3%. 3%. There's your answer. 3%.

Mark Hinaman (48:08)
Mission

critical 3 % to keep things going, right? Yeah, yeah. Yeah.

Phillip Koblence (48:11)
I mean that's I'm just I'm using Brian's example Brian Smith the

Brian Smith (48:11)
Bye.

Phillip Koblence (48:15)
director of nuclear development at Idaho National Labs says 3 % of digital infrastructure deployments are are critical

Brian Smith (48:15)
Yeah.

Hahaha

⁓

But Mark, you ask a good question about, so for the INL ⁓ under this program, Department of Energy is leading. ⁓ I'm excited to see what kinds of deployments ⁓ teams are selected by the Department of Energy because it's not crazy to think that there could be a 50 megawatt deployment that would be grid connected in Idaho.

Mark Hinaman (48:27)
you

Brian Smith (48:47)
But with a, you know, whatever, a 75 megawatt reactor co-located, probably some diesel generators involved in the mix, some batteries to handle load transients, et cetera, right? It doesn't necessarily, don't, Philip already covered it so well, but I think you could translate all of that to site specific for Idaho and determine why that commercial venture, that commercial data center would fit very well within that INLN.

Mark Hinaman (49:17)
Gotcha. Awesome. Okay. ⁓

I said earlier total demand and air quotes and that was kind of my way of asking or saying like how much power is actually needed versus I heard a podcast recently that said that there could be some incentives throughout the industry to use hyperbole and like exaggerate a little bit how much power is needed because like think about if you're a hyperscaler then

and you say there's a power shortage, then other people could jump in and start building power to solve your problem for you. Do you guys think that's in play here? Short answer, right? Yeah. ⁓

Phillip Koblence (50:03)
Yes. Wait, ask the question again. Yes. Yes. It is definitely in play.

I mean, I think, you know, I have some experience with ComEd as an example, which is, you know, the provider in Illinois for specifically for the kind of Chicago sub market. And they are under capacity as of, you know, a couple of months ago of their pipeline of of committed power that people have actually put money down to secure. were under capacity by 29 gigawatts. That is an absurd.

Mark Hinaman (50:32)
Is that breast milk in there?

Phillip Koblence (50:32)
number.

And I think even they are working

under the assumption that regardless of whether someone paid for it or not, a lot of those workloads are chasing the same underlying customer and whether those sites are built or not is really up for debate. And I think any time you have ⁓ your companies whose valuations much like I'm not suggesting we're in a bubble. There is another podcast that I have coming out on Monday where we talk about this with a couple of people. The name of that podcast is Cool Vector.

Mark Hinaman (50:38)
Data farms.

What's the name of it? Can we?

Phillip Koblence (51:03)
I know we have Michael Elias from TD Securities, who's like a big analyst where we talk about whether this is a, you we're in an AI bubble or not. But I think ultimately the one similarity that I can have with the kind of dot com bubble is the focus on companies, not necessarily for how profitable they are, but at the pace with which they grow. And any time you have stock prices that are impacted positively just by saying they have, you know, ⁓

Mark Hinaman (51:04)
you

Phillip Koblence (51:33)
⁓

coming deployments to put online. You see like the ⁓ going public of CoreWeave and companies of that same ilk. You saw Microsoft actually have a stock market dip when they paused some developments in January, which was broken by Michael Elias that's going on that podcast. Anytime the fluctuation of the stock price is based on the perception of just...

Deploying for the sake of deploying and not really focused on what is the underlying workload? Is there a Profitability associated with the pace of this deployment then you have some you know, you know hard to unsee Corollaries between like the pets dot coms and and all of those things that were focused on You just grow get user base at all costs Amazon didn't make money for the first, know 10 years of its operation because it was really just all about getting customers to be

of that ecosystem and I see some similarities here and you'd be crazy not to think that some of the demand is manufactured and ultimately will not I think it will all be absorbed I think one of the corollaries for the dot-com boom was there were there were there were articles that came out in the Times that said you know the internet is a fad I don't think we're at the level anymore where we think digital infrastructure won't eventually be absorbed so I think this is a worthwhile investment

but I think just like the dot-com bubble, there are going to be winners and losers and there are going to be lots of folks that end up building capacity that isn't absorbed on the same timeline that they're underwriting towards. ⁓ And I think that's, ⁓ you know, that's, yeah.

Mark Hinaman (53:15)
Well, so yeah,

that's great segue and great way to frame it actually. So like Phil, you're talking about the demand side of this equation, right? And Brian has ⁓ direct, I won't say access, like insight ⁓ into part of the supply side, which is like new nuclear reactor technology that's getting developed. we said, I don't think we explicitly said, but earlier in for like, these technologies have a long time horizon for their development cycle. And there's,

Phillip Koblence (53:22)
Right.

I think that was inferred

Mark Hinaman (53:45)
There's,

Phillip Koblence (53:45)
by the air quotes.

Mark Hinaman (53:47)
yeah, yeah, exactly. Yeah, yeah. There's a lot of them, right? Like, I mean, I feel like 2025, there's been several new companies that have materialized and like, we're gonna make a nuclear reactor, you know, which is very exciting. And I think helpful for the industry, but I'm curious, Brian, on your take about like, could the sector become oversupplied for relative to demand? And like, what does that look like?

Brian Smith (54:14)
Yeah, that's a great question. And I think right now the count around the world for advanced nuclear companies, somebody in some way, yeah, it's like, think 127 is the last count. so if you asked me, will 127 companies successfully

Mark Hinaman (54:14)
How do you think about that? Supply side technology.

⁓ it's like over a hundred, right? Like, yeah.

Brian Smith (54:44)
deploy commercially their reactors? I think it's fair to say the answer is no. Certainly I don't think so. ⁓ Is there space for ⁓ multiple? Yeah, I also think there are. There are some different technologies out there. They're all splitting the atom. high temperature gas reactors come with high temperatures. ⁓ That heat is very useful beyond just producing

power. Right? And so, I mean, that's the demonstration of the X-Energy reactor in Texas is in a partnership with Dow Chemical that wants the heat from X-Energy's high temperature gas reactor. It's good stuff. I mean, you can use that in steelmaking. That's good stuff. Now, until we move that type of reactor down the cost curve from those early deployments that occur at a higher cost,

down to ⁓ deployments that the market in those other off-take industries would embrace. Let's just say that's gonna take a little bit of time, but that's what we're working on, getting those first deployments made so we can actually start to ride down the cost curve to achieve order books of deployments.

So high temperature gas reactors, kind of have that use case. That's good news. mean, heck, any of the advanced reactor fleet could be used for its heat. There's a lot of industries out there that just need heat. Desalination operations, right? You can move to very clean desal operations in parts of the world that you'd never really think about. I microreactors using their heat on the coast in Africa, providing fresh water, like Olympic-sized swimming pools, multiples of them every day of fresh water.

to communities that don't otherwise have access to fresh water. That's a great use case. And so there are a lot of other use cases for advanced nuclear. We have to remain laser focused on the power piece for all the reasons we've just talked about. But we can't lose sight of all of those other exciting use cases that nuclear brings along with it. And again, as Phil said at the top, those SMRs, those micros.

Those are a lot of those other use cases. Yeah, you're probably not gonna build a large scale gigawatt size reactor on the coast in Africa to make fresh water for desal operations. ⁓ But you might do that with a couple of micro reactors that take up as much space as a tennis court. So that's good stuff. That's exciting stuff. And I think that's why this is such an exciting place to be right now.

Phillip Koblence (57:15)
I think I think I think

the way we're looking at it now is that we don't measure things in tennis courts. We measure them. We measure them in pickleball courts. So it's for pickleball courts. Yeah, it's.

Mark Hinaman (57:21)
Oh man, that's a trend. We just posted some visuals on the Fire Division website that have

Brian Smith (57:28)
you

Mark Hinaman (57:28)
nuclear waste

and then some oil and gas waste, but they're on a tennis court, not a pickleball court. But we'll have to update that, right? Put them on pickleballs. That's wonderful. Brian, I really like that framing and a great counterpoint of like, electricity isn't the only product that is being offered here. electricity is a commodity and a service and it's an easy solution.

Phillip Koblence (57:32)
Right, no, no, no, I mean, come on, no doubt, no doubt, right.

Brian Smith (57:33)
Yeah.

Yeah.

Mark Hinaman (57:52)
to use this technology for, but there are other solutions and if your technology can solve multiple problems inexpensively or cheaper than competition, you stand a chance to win. It's still a competition. ⁓ Okay, guys, we're coming up on our time. This has been fun because I can tell already we're gonna have to you guys back on either together or separately. ⁓

But how about parting thoughts and I'll ask the question, you know, if you had to say by 2030, what has changed? Like let's say we're having this discussion end of year 2030 instead of 2025. Like what's changed? What steps have been taken? How has the landscape evolved over those five years?

Phillip Koblence (58:38)
Look, my sense will be that ⁓ I'm not sure we will have made a significant change in the overall complexion from a nuclear perspective. think a lot of those technologies will have evolved, ⁓ but I don't think there'll be enlarged commercial use ⁓ at that point in the way that they may in the future. That's not to say the upscaling of the existing sites, but I'll let Brian touch on that stuff. think from a data center perspective, I think

The thing that I find super unique about this moment is that so much of our planning is based on a spec of a kind of chipset that exists today. And I think the one thing we have recognized over the course of the last five years or even three years, there was this quote at the OCP conference in October that the last year has been the most impactful decade in compute in history, which just suggests that things are evolving at such a

quick

timeline that the idea that a data center that is designed today and deployed today is going to be relevant to the exact use case that we have in 2030 is a misunderstanding of the way you know this technology evolution cycle is is working and it's only proving itself to be more rapid. And I think the one thing that the past has shown to us certainly about compute is that things tend to

Once we solve for the bottleneck, was operation cycles, which is what these GPUs are doing, we start focusing on how do we create those operation cycles more efficiently from a power perspective. So when you had like spinning disks being the prominent way to store data, you ended up with a fairly rapid evolution to solid state drives because you were able to solve the space limitations that solid state drives had, which

created a vast reduction in the amount of power that was used for solid state tribes. I see the same thing happening in this kind of compute cycle where if you asked me to look into crystal ball in 2030, I would say that you have a significant, refer to it as, you know, kind of overbuilding. I think there is going to be a significant amount of overcapacity available in the market of sites that, you know, are going to only come online in 2028 and 2029.

because that's the cycle with which companies are developing now. They have sites that are under development now that are going to be ready for service, RFS, in 2028, 2029. And I think in 2030, it's going to be a completely different landscape for what sort of power requirements, what sort of deployments, how the kind of large language model experience has shifted towards a more inferencing focused

methodology which is going to move away from some of those large giant campuses in the middle of nowhere and maybe focus back towards the edge, which is still going to require some infrastructure investment in kind of major population areas, but in kind of a different way. So I don't know if that answers the question, but all of that to say is my name is Phil. Great to meet you.

Mark Hinaman (1:01:58)
I think that was excellent. Yeah. Thanks, Bill. How about you, Brian?

Brian Smith (1:02:03)
Yeah, I

think in 2030, we'll have put at least five gigawatts more on the grid from the existing fleet of 94 reactors. I think we'll have at least 10 large scale reactors under construction. Now that might be at four or five sites. So, you know, that could be places where we're building anywhere from two to four.

but I think at least 10 total large-scale reactors will be under construction by 2030. And in 2030, we'll have turned on multiple advanced small reactors starting next year. And that plan is already baked. So in 2030, we'll just be counting just how many of the small demonstration reactors we have turned on.

And by 2030, we will have powered a data center using co-located advanced nuclear power generation. Hopefully even more than one. Again, I hope in 2030 we're counting just how many of those we've turned on. Again, and I think, you know, it's important to have that caveat. And I think Phil hit on this. That doesn't mean that we're talking about the entire data center industry as powered by nuclear in 2030, but we're hitting those

critical milestones this decade and celebrating them in 2030 because they are the key milestones we need to advance into the 2030s with a plan to deploy at scale. And that's why I keep saying it's such an exciting time now and in 2030, we're just gonna be looking forward to more excitement.

Mark Hinaman (1:03:51)
exciting times. Awesome. Phil, Brian, thanks so much for taking the time to chat with me.

Brian Smith (1:03:57)
Thanks for having us.

Phillip Koblence (1:03:59)
Thanks for having us, Mark.

‍