Speaker 1: 0:02

Welcome to Uplink, where we explore the world of digital infrastructure, uncovering the technology, fueling AI and cloud innovation, with the leaders making it happen. Good morning, how are you? Yeah, jay, great to have you. Thank you for coming. Yeah, I've got a lot of questions. Do you have a lot of questions? I do. I've always got a lot of questions. So well, firstly, you're at Evocative now, but that's not where you started. But you've been playing in this game for a couple of years now 29 years. I think it was 90s that you were saying where you started, but you've been playing in this game for a couple of years now 29. 29 years I think it was 90s that you were saying that you started out.

Speaker 2: 0:49

Mid-90s I went to work for my business partner. We retired seven years ago.

Speaker 1: 0:55

Yeah.

Speaker 2: 0:56

When we sold Cummy to Evocative.

Speaker 1: 0:58

Beautiful. So, Alan, you're in the 90s Internet's sort of just coming into its.

Speaker 2: 1:03

It was still Wild West when I started, but it sort of.

Speaker 1: 1:06

And was it internet that you're involved with, or was it actually sort of data center or Colo? What was?

Speaker 2: 1:11

it. Then we pivoted to Colo in 1999. Wow, 2000. Okay, and then, before then it was, we sold dial-up and hosting and T1s ands3s and then got into the dsl game and so you were actually um reselling those like as a partner, yeah, and then sort of started to build out well, we always wanted to manage the network right, because they need support and assistance. They didn't understand how any of this worked, and that's what we were for, and so this was like what was that DSL dial-up sort of era?

Speaker 1: 1:49

Well, pivot dial-up to DSL and business or consumer.

Speaker 2: 1:54

Both really. But again we pivoted away in 99-2000 to Cola World.

Speaker 1: 2:00

Okay 99-2000, Cola World. Yeah, everything's exploding, it's about to crash, and then it popped.

Speaker 2: 2:07

You survived, we survived, that's pretty impressive. The funny part is we built our first data center at 6R West 7th Street downtown LA. I was buying gray market cabinets from level three and I had like a bunch of network here that we got you know from like. I had patch panels, I think in the data center they still say Enron on them, I mean.

Speaker 1: 2:36

I've been around a while, I'm a dinosaur. I mean, back then data centers had changed. I mean they've changed and they haven't changed. But what was it? Was the what who's learning. And then, what were you trying to do? Was it more carrier related or no, it was more, you know.

Speaker 2: 2:51

I mean, back then we went to that site, right, the carrier was out yeah there because you know for the cost of a cross connect at the time it was cheaper to just be in the same building with all the carriers, just connect straight straight in.

Speaker 1: 3:03

yeah, which was all it has changed. But that's been this sort of key. It was super key. All that interconnect, all that fiber coming in, all the carriers that became sort of these, I think, the founding part of many data centers now actually. Yeah.

Speaker 2: 3:18

Everybody wanted to be where the carriers were close, right. So you know, customers would buy a cabinet or a cage and we grew there to, uh, almost two megawatts back then, back then this is huge. And then we sort of shrank a little bit, um, and then I want to say the peak, there was probably 2012 2013 got it yeah, and then what was happening then?

Speaker 1: 3:44

the cloud. Aws was born yep, that changed everything it did. Yes and okay. You're at that period you're a data center company and the cloud comes along and everyone tells you that and you're going out of business.

Speaker 2: 3:58

You're right, yeah, I'm still here. Yeah, right, it's so hot right now, and isn't it funny to go I'm still here.

Speaker 1: 4:04

Yeah right, it's so hot right now. And isn't it funny to go back into that time?

Speaker 2: 4:07

You know this business has been cyclical for you know, forever. Right it was hosting, and then I want colo on my own equipment, and then it went to AWS.

Speaker 1: 4:17

Yeah, but are you also doing some hyperscale as well, or is that just On the smaller side right?

Speaker 2: 4:25

Hyperscalers. Don't come to us for 48 megawatts, right they?

Speaker 1: 4:28

oh, we have a network pop and we need, you know, a couple megawatts here, yeah but you're also saying some of these neoclads, all sorts of, I mean, it's so many pieces of the puzzle that you solve for. But before we jump into that, you ended up. I guess that company then was acquired by locativeative in 2018. Okay, but you didn't just sail off into the sunset.

Speaker 2: 4:49

My business partner retired. He sailed off into the sunset. He sailed off into the sunset. I didn't know what to do.

Speaker 1: 4:53

You can always leverage his boat. You know that's important because you'd have to pay for the repairs. Yeah, exactly.

Speaker 2: 5:02

So you stayed, I stayed and you know I I said I'll give it six months. I'm here seven years later, okay, and you're building, you're growing. We've done some acquisitions and we're building today, we're adding in our facilities in Secaucus and Plano. Yes, early stages in facilities in Atlanta and Redondo, so give us the like high level?

Speaker 1: 5:23

how many DCs you got US-based?

Speaker 2: 5:25

15,.

Speaker 1: 5:26

US-based. Any thinking of going outside of the US Exploratory? It's still early. Yeah, I don't know if you need to. I think there's so much opportunity here still, I mean, the majority of the data centers are in the United States anyway. Correct Plus it's hot and actually I think a lot of maybe more pullback for the sort of geopolitical landscape that's going on. I think there's probably more opportunity. At the moment it's changing fast.

Speaker 2: 5:51

Yeah Well, and again, I think we're just starting to scratch the surface, right. Like everybody's training, Everybody's building to train and you know, once they're trained, what happens. On inference yes, edge. Yeah, we're going to you know, once they're trained, what happens on inference yes, edge, yeah, and it's gonna, we're gonna push to the edge and then it's gonna be all I need, you know, 30 to 100 kilowatts of rack, but I only need two racks. Yeah, and I don't really know.

Speaker 1: 6:17

Yeah, I'm sorry it changes what people are gonna need or what companies are gonna. They're gonna want to push that to the edge. Yes, and that's a latency thing uh, they want to you.

Speaker 2: 6:25

everyone on their phones wants to click and get the inference quickly and that game is so. We talked about carrier hotel. Yeah, it's different now.

Speaker 1: 6:43

Your service, sdn, pushes the edge to any data center you're at you, don't necessarily need to be in a carrier hotel anymore that's kind of one of the things I was just thinking. Our value prop is that you actually we provide that capability from any data center that we land in yeah, yeah and connecting you anywhere now.

Speaker 2: 7:01

So that does change that sort of carrier hotel mentality and I I, when I think the training's done, I think we're going to see another explosion of connectivity. That's just, I mean everybody's. You know our last 10 years has been buying gig and then 10 gig, and now you know everyone's, 40 gig or 100 gig, and yes, well, well, it's going to be 4, 8, and I don't even know what's after 8.

Speaker 1: 7:27

Yeah, Well, we're even trying to procure 800. We've got well. I mean, our backbones are all 400 gigs so we can deliver 100 gigs in 60 seconds and we've been just on this note, we've been asking for 800. We want it to be one of the first in the world to roll out 800 gig backbone so we can deliver 400 gig. We're struggling to find people that can deliver that.

Speaker 2: 7:52

We might even cross to 16, you know 1.6 terabyte, you know before we get to 800. You would have asked me 20 years ago. We're going to push 1.6 terabytes down a pair of fibers. I would say you're crazy. And here we are, like every day it seems to keep. I mean, it's just going to keep doubling.

Speaker 1: 8:03

What's weird is when we roll out 100 gigs and we punch them out quickly from your DC, people fill. That's the weird part, because you can talk about speed all you like and until it's been consumed. It's only when you know it's been consumed that someone's doing something with it. It's like all these AI folks A lot of them sort of came on early and we didn't see much happen, but now we're starting to see all this consumption come through and maybe that's the role out of inference, potentially, or it started like I figured it would be another year or two before we got there.

Speaker 2: 8:34

Yeah, the workloads are getting trained way faster, so it's a matter of, and it's a matter of, and it's everything from your personal assistant bot to image gen video generation. I'm going to plug my daughter who's a graphic artist, right, and she said Dad, I'm not going to be able to be out of a job. Ai is going to put me out of a job and I'm like, no, somebody has to teach AI how to do it.

Speaker 1: 9:02

Why it has to babysit, because AI it's a little bit random.

Speaker 2: 9:06

It's got some hallucinations, but again, so she's dove into that too.

Speaker 1: 9:11

But that's the key. If she can leverage the platform to make and that's what we've found Content's been created so much quicker. But if I try to create content, it's terrible. I'm trying to tell the model or whatever, but it's been able to use this. You know, tell it what you want in a way that it needs, so it gets coding for the AI. I suppose it's the new hot thing. It's communication to it Actually it's going to be fine.

Speaker 2: 9:38

And again, it's not that you have to tell it, you kind of have to coach it along. Yeah, it's coaching Right.

Speaker 1: 9:43

I'm an.

Speaker 2: 9:43

AI coach. What does that mean? I mean, maybe it's a new job. Who knows, that is the new job. Yeah, so you know again, everybody's after all this capacity and everybody talks about it. There's not enough power in the United States. Yeah, there's not enough. The grid can't support it. So now it's oh, maybe I'll do gas generation or natural gas.

Speaker 1: 10:07

I mean gas is quicker, I guess, in terms of what they can roll out it is. I mean, the other piece is not just generation, but distribution is certainly an issue.

Speaker 2: 10:20

But if you can do generation on site, then your distribution is.

Speaker 1: 10:24

Are you guys thinking about that, or is that just something you explore? I?

Speaker 2: 10:28

started digging into it. Some of the byproducts which are beneficial are that when you do generation, you can use it also produces chilled water, so you can leverage that for these high-end cooling capacities you need.

Speaker 1: 10:43

Yeah, so you're getting people turning up trying to, I guess. I guess liquid cooling is that something that you're getting asked for, and the yes, the amount of kilowatts per rack has just gone up.

Speaker 2: 10:53

We've, we've, and and part of it is you got to take an existing data center or retrofit it.

Speaker 1: 10:59

That's how do you do that? Yeah, in a perfect world, water Liquid. Either medium right, yeah, you can use.

Speaker 2: 11:06

DX. The cooling suppliers are rolling out new products every day.

Speaker 1: 11:12

Yeah, and it's again quick, I guess, yeah.

Speaker 2: 11:16

I mean, in a perfect world, you have a chiller plant, you do a CDU liquid to the rack, whether you do RDHX or Are they chilling that right down, or is it just room temperature and that's enough to cool it?

Speaker 2: 11:31

One of the nodes we have running today is in a 20-year-old data center. We're supplying 64-degree air on the cold aisle air on the cold aisle. The AI servers are grabbing that air, pushing it out the back north of 100 degrees into a rear-door heat exchanger and that rear-door heat exchanger is grabbing it basically at the source and spitting out 68 degree air. So you're going from 64 to 68. So now you only have to cool 68 with traditional coolants.

Speaker 1: 12:05

Interesting. So what's that? That's like a chiller unit on the back that chills.

Speaker 2: 12:11

It's a door that's basically a crack unit or a cray unit. Yeah, it's intense.

Speaker 1: 12:17

And so that's been able to. You were able to basically retrofit the data centers that you've had the older data centers, I guess, the ones that would have been around to make them purpose fit for today's requirements. So you don't have to.

Speaker 2: 12:27

At the smaller scale. You can't again. You can't do it at scale Massive, yeah.

Speaker 1: 12:31

Now it's the other one. So we're seeing other data center providers that have just pivoted to focus solely on sort of hyper scale. But what I think that's pushing down is this opportunity for colo, because they're pushing colo out. Some of them are saying we don't want to supply colo, we just want one big facility. But now then what that's from what I can see, is colo is sort of rising.

Speaker 2: 12:54

It's a plug bed. Right, it's simple economics, yeah, so when there's less supply, the costs go up. Prices go up and there's still traditional coal. Like people look for small footprints all the time. Yes, two racks, five racks and even the like. Originally, I push was oh, we need five megawatts. Well, now we're starting to see no, we need 100 kilowatts, we only need eight racks. So, and that's a traditional nvidia configuration of eight to ten racks with one storage, one network. Okay, six to eight racks for ai all right, and so how?

Speaker 1: 13:32

how many dc's you guys have now? 15, 15 and regions. How are you thinking about um? Did there a mixture of acquisitions? Or how have you thought about the strategy of where they're landing, where you're building, where you're investing, et cetera, et cetera?

Speaker 2: 13:48

So again it's about near term. Where do we have stable power from the utility?

Speaker 1: 13:53

Okay, great, it goes back to what. Can you get what?

Speaker 2: 13:55

can I get? Where can I get?

Speaker 1: 13:56

it, and can you build more in some of the facilities, or are they sort of cut? Add more to?

Speaker 2: 14:03

that system, plano, texas, prime example. Yeah, we're in construction in Secaucus and in Plano today Cool.

Speaker 1: 14:12

And that's an extension or a new DC we have an empty hall right.

Speaker 2: 14:16

Got a technical warehouse today. Yes, we can get small chunks of power in Skakos, so it'll be effectively a couple megawatt add Turn it into Texas and they say oh yeah, you can have 24 megawatts utility in 12 months. So, eight times what we can deliver in Skakos, we can deliver in Plano.

Speaker 1: 14:41

So you'll build there, yep, and is it a warehouse style facility and you'll extend the warehouse or just inside the warehouse we have built?

Speaker 2: 14:52

effectively data hall that we can expand into and that'll come online.

Speaker 1: 15:00

May of next year beautiful. It's still available. It might change in a week online. May have next year Beautiful. Have you sold?

Speaker 2: 15:03

it already. It's still available right now. All right, yeah, we got some.

Speaker 1: 15:06

Yeah, it might change in a week. Yeah, yeah, that's what we're also finding, yeah.

Speaker 2: 15:11

It's and it's it's purpose built. So you know we're not. We're building and planning these new hulls to be liquid, to the rack ready, right. And then you can go RDHX or you can go with what's RDHX?

Speaker 1: 15:28

Rue Dirty.

Speaker 2: 15:29

Exchangers. Yeah, okay. So put the door, the magic door, on the back of the rack and you're cooling from 70 kilowatts down to what?

Speaker 1: 15:37

is that Just like a big fridge, like a radiator on a car?

Speaker 2: 15:40

It's a radiator on a door.

Speaker 1: 15:42

Yeah, it's a radiator on a door. It's a radiator on a door with a. You're running cold water into it, or cold liquid or something into that?

Speaker 2: 15:46

Yeah, cold water.

Speaker 1: 15:46

It's just coming through, comes out, and then you're grabbing that and just recycling it back through. There you go, just close the loop. Yeah, that's interesting. So that's different to what has been done in the past, like stair cool just came through, as opposed to then having this cooler unit in there, or is that not the case?

Speaker 2: 16:06

Again, you do it in a big box format, right yeah, you surround the edges with all these HVAC units and you're traditionally blowing cold air onto the floor, putting the cool tiles where you want, and it distributes it where you want the cold air to go, and then it sucks it out. And then heat. You know, hotter rises, colder falls, yeah, hotter goes to the ceiling.

Speaker 1: 16:28

To return air, plenum HVAC grabs that air, typically a 20 to 25 degree delta T between what it grabs it in at and what it spits it out at yeah, so now you're delivering and also you're saying, when you say, um, cooling to, that's to the chip, is that like where?

Speaker 2: 16:48

you're running liquid and you're running, so you, the the racks, will have generally stainless steel chilled water to them, okay, every rack, you can tap into it.

Speaker 1: 16:55

You tap into it like you would plug into power. You're now plugging into a water source. There you go. Okay, you try to want to keep them apart as much as you can right.

Speaker 2: 17:03

Yeah, yeah but again the connectors are. I mean, the technology on the connectors is just insane. You literally unplug the connector on the chill water loop.

Speaker 1: 17:13

Well, that's I mean. I know it's been around for a while, but it has never been around at scale like this.

Speaker 2: 17:18

So there's so much change Gaming. To me, that was a big push into cloud gaming, in my opinion. Everybody wanted their own game server to run. Yeah, I'm going to date myself a little bit, but you know, get back to the Counter-Strike. 5.2 beta days right Love Counter-Strike.

Speaker 1: 17:36

We still land that at university.

Speaker 2: 17:38

There you go. So I think that was a big push. And then you know they were. Oh, we got to get more out of these gpus. Right, I gotta run it faster because I gotta have latency. Yeah, and sure enough, they brought liquid to the the chip. Yeah, and I think at that point is when NVIDIA said whoa, this works, how do we do this in the world? And again they're talking. Some racks are doing 275 kilowatts a cabinet, 600 and something kilowatts a cabinet megawatt a cabinet, so it's like megawatt a cabinet.

Speaker 1: 18:19

Yeah, I mean Megawatt data centers. Yeah, and now you're getting that in a cabinet. So the what people probably need to understand, or maybe it's obvious, but if you're punching a megawatt of power, it means you're creating the heat, and so it usually takes one one and a half.

Speaker 2: 18:37

So for every megawatt you consume, you're consuming a megawatt and a half of power, utility power or generator. 0.5. Basically, to call it right is you're consuming a megawatt and a half of utility power or generator power.

Speaker 1: 18:44

So it's 0.5, basically to cool it. Right, is what you're saying? Yeah, got it. Yeah, fascinating, and that's so. You punch that into one rack. There is no choice but to cool it via some sort of liquid. You have to do it. You melt a hole on the floor.

Speaker 2: 18:58

Yeah, yeah, yeah, yeah, yeah, I mean it's pretty much it's, but that's what's changed so quickly, like that's three years ago. You have this conversation.

Speaker 1: 19:09

You say you're crazy, yeah, but that's a big change. That's a big change because it's such a physical like. We change a lot in um technology and software, but what's changed is the physical build, like actual physically having to run liquid to these locations. The connectors that you're plugging in ensure that those things don't leak, don't have an issue. It's obviously running some incredibly expensive tech. How do you do that on scale? How do you choose the right thing?

Speaker 2: 19:33

You put four of those chassis in the cabinet. That's north of a million dollars a rack. That's crazy, just in servers. You're not even talking switchgear, you're not talking storage. It's a millionaire rack. How many racks you got in there? It's a lot. Again, let's say a standard footprint is going to be 10 racks. That's $10 million, $15 million in their gear alone, plus the cooling, plus the power, plus the cooling, plus the power, plus the build to install it, I mean he gets to a billion dollars pretty fast.

Speaker 1: 20:02

Yeah, I think that's what people probably um don't fully appreciate. Now here's the other thing a lot of talk about this, um, we're in a bubble, oh, we're in a bubble everyone keeps caring about. I've heard, um, I'm actually hearing less of that now. I'm hearing that it's actually more of. This is just the way it's going to be ongoing. Um, a little bit, not that long ago, we heard bubble. Then we had this, that little freak out moment when um was the chinese brought out that. Uh, which model did they bring out that scared everyone? Yeah, again it's. I don't even remember which one it is. Yeah, exactly. Yeah, deep sea came out or something yeah it's like we don't need to try anything.

Speaker 1: 20:37

And then that then the chain came back, but I'm hearing that it's getting bigger. You know, like when I just heard the grok so grok, grok elon's one like you had you figured out how to get a hundred thousand gpus, and that was this crazy thing three months ago. Someone said he's at 300 000 and someone said he's trying to get to a million of gpus just on training.

Speaker 2: 20:58

It's like everything's getting you know they're built, and they're building what they did in Memphis. Yeah, it's insane Well they didn't bother with utility power.

Speaker 1: 21:08

Yeah, again, natural gas generation.

Speaker 2: 21:11

Was it gas generation?

Speaker 1: 21:12

was it?

Speaker 2: 21:12

Yeah, natural gas generation, and they were able to do it at scale, rapidly. And you know, again, the byproduct is you get some cooling out of burning the turbines, yeah, so that's an you know okay. Do you bother with generators at that point? Diesels, yeah Right. Do you just go with natural gas and make them redundant in that configuration? Yeah, I've kind of explored that a little bit Interesting. So rather than, rather than running, having the traditional COLA build with utility feed, with generator backup and EPS, just go natural gas generation with a.

Speaker 1: 21:50

UPS. What do you need?

Speaker 2: 21:53

lots and swaths to do that, or you just get approval from the build or something. Every market's going to be different right.

Speaker 1: 21:57

So then it's just gas prices that you've got to worry about.

Speaker 2: 22:02

You buy RECs to offset that as well. But again these are very, very early stage.

Speaker 1: 22:07

Yeah, well, this is to solve the problem. The problem is that we don't have enough power. So how do you solve that? One thing you can always guarantee if there's a way to solve it, you know, a capitalistic mindset is to go and solve these problems, and if it's not government driving it and it falls into the capital markets, capital markets will solve it if they need it. Yeah, we're seeing that come. That's interesting. On the gas gen. Didn't think about that. You don't need everyone's like running low on diesel well, and diesel is.

Speaker 2: 22:38

I mean? You know I'm closer to retirement college, right? So I'm to the point where it's like, oh, I'm going to leave it, leave the world better than I found it right. We don't need AI to turn into this matrix, right? When it takes over, or if it takes over, I don't think it can take over. You know, it's funny, skynet.

Speaker 1: 22:59

Oh, we have staff running around.

Speaker 2: 23:01

They say Skyn net is my friend they have t-shirts that they wear them. Yeah, it's. Uh, you know, just don't give it the nuclear code.

Speaker 1: 23:07

So I guess this is a way to avoid that put it down on a piece of paper the nuclear codes. That conrad is yeah, yeah. So what else is?

Speaker 2: 23:16

going on in the evolved world, can't? We're just just again Colo's cool again.

Speaker 1: 23:23

Yeah, it is cool, and I think the new, the traditional, so when you say Colo's cool, are you saying companies just expand more? Are they repatriatizing, are they mixing what they put in cloud?

Speaker 2: 23:36

I think the new phase or the new mode is to do you have to do hybrid, right? There are tons of things, tons of workloads, that make total sense to run in the cloud, right, yes, and there are some things that makes more sense to run in gear that you support and you know a lot of, and so you're seeing customers make those choices deliberately.

Speaker 1: 23:57

In the past it was sort of like we're migrating to the cloud whatever's left we're keeping because we couldn't get it to the cloud or we didn't get there or we didn't get to the cloud in the first place at all. Now you're seeing deliberate choice to say, all right, we're putting a whole heap of this in the cloud and, by the way, the pie is just increasing for everyone. It's not like one loses. Whichever way this goes, Everyone wins. But it's more like what? What those decisions are. So you're saying like workload specific.

Speaker 2: 24:25

come to you and say like we're building this here and that there I mean a couple years ago. We got some statistics that said 47 percent of cfos that were interviewed their spend was 20% to 25% higher in AWS and Microsoft and Google than they originally budgeted.

Speaker 1: 24:45

But the question is is it that they expanded 25% more or they just didn't budget enough, or is it more expensive than what they? I think it's a combination of all of those things.

Speaker 2: 24:53

Right, You're going to have your zombie compute instances that somebody spun up because they had to log in, right? Yeah, and it just sits there spinning and nobody's using it.

Speaker 1: 25:05

Yeah.

Speaker 2: 25:06

You know, egress right, Egress is expensive.

Speaker 1: 25:10

Yes.

Speaker 2: 25:11

And you know you can offset some of those things by whether you're leveraging an on-ramp, if you would either way, and pulling it back, or you know, hey, maybe we just build a stack and again it's back to do you want OpEx or CapEx? Yeah, those are real, yeah, yeah. And again, everybody pivots from one to the other. Oh no, we've got to spend more on CapEx.

Speaker 1: 25:36

We're now on CapEx. No, we're on OffEx. Yeah.

Speaker 2: 25:39

It's changed all the time.

Speaker 1: 25:40

Yeah, yeah, All right. So Crystal Ball looking forward. You've been in this game since the 90s, so you've seen it all. I don't know what's the next. What comes in the future? Five years out, ten years out? Probably too far to think. Ten years out, I can feel it.

Speaker 2: 25:56

I think it's more than two years 10 years out, I'm thinking, oh, I'm taking my dog to the park every day. It's been long Playing with my tulips. No, I mean, I think if you would have told me three years ago I don't want to think five years, I want to think three years.

Speaker 2: 26:15

Yeah, If you told me three years ago we, we're gonna go liquid to the rack and look into the chip and I was said you're crazy. I'm happy that a year later I'm like no, this is coming standard now hey, we go to have it yeah again, demand is there. Yeah, so and again back to you. Know all these work, those are getting trained and inference is gonna be humongous yes and well, I saw a demo of inference.

Speaker 1: 26:43

I mean, if we went and saw it um at the grok labs and groq not grok, and just the, the speed at which you're getting full responses it's, it's pretty much instantaneous, like it's milliseconds, versus you are noticeably maybe 10 second responses in the past. These things are delivering at milliseconds.

Speaker 2: 27:06

Well, again you're seeing it on site in their lab with very close connectivity low latency. Correct yes, and if you're running it, for instance, and I'm in New York City and I pull out my phone and again that's the other part that I think everyone doesn't realize is everybody's going to go to inference off their phones? Yes, of course.

Speaker 2: 27:35

I mean we all do searches, like somebody brings up a movie at dinner and somebody pulls out their phone and then you have all the data. Yes, well, when somebody wants to ask the AI well, what about this, what about this? Or give me the data on this, and they're in New York City and the inference is pulling from the West Coast, yeah, big delay I had 50 milliseconds, yeah, plus it's got a process and the time and and and as opposed to well, let's build all of our instances all over which is a bit like networks actually.

Speaker 1: 28:13

Um, I mean, you were always trying to design a network so that you weren't tromboning from one side to the other. You're trying to get the content as close as possible. Cdns is a great example. It's almost inference becomes like the content distribution networks. It's like of of all becomes maybe the new inference at the edge as it changes. That's changing super fast and there's some really interesting players sort of jumping into this space. Nvidia is changing, but there's other chip providers that are appearing that are just purpose-built for that. Also, you know, a lot of the hyperscalers are building their own chips, so it's all happening at the same moment.

Speaker 2: 28:48

And again, it's all going to be slightly different than everything else. Yes, you know it's flow rate on chilled water and what temperature do you want it at? These are the specs that they're asking Flow rate on chilled water.

Speaker 1: 29:00

How fast can you pump water into it? Flow rate.

Speaker 2: 29:05

Flow rate and temperature and filtration Interesting. Do you want water that's only been filtered to 25 microns rolling down and settling and some chunk of sediment settles on a chip? Wow, and again that's a $300,000 box. Yes, you just.

Speaker 1: 29:27

I'm going to figure out that. Yeah, and is it oil as well? Or is it water? Water and it's not necessarily oil.

Speaker 2: 29:36

I mean, you know, submersive is a rub, you can go, but we're seeing it not go that route. You're seeing this, I'm seeing it the other way and it's mainly it's water. Yeah, again, water generally. But again there's. You know you're going to treat it with some type of chemical, yeah, glycol or what have you right? But, you know, having filtration, that's 10 microns, yeah, making sure that it's as about as pure as it can be. Or, you know, desalinated water, right, so it's, it doesn't create any sky.

Speaker 2: 30:08

No, and you don't have anything that builds up and it doesn't conduct electrons. Interesting. Yeah, so it's again. We're just starting to. I think we're really starting to just scratch the surface on how crazy it's going to get, because how long before everyone catches up with nvidia? I don't uh, is it a year, is it three?

Speaker 1: 30:27

years. Those guys move so fast too. Um yeah, the tech they're coming out with in the iterations just insane. It's amazing all so one last piece Advice for folks looking to get into the industry Is it a good industry to land in?

Speaker 2: 30:45

Absolutely. Whether you go to college or not, it doesn't matter.

Speaker 1: 30:49

That's a good point.

Speaker 2: 30:50

Think about this. So every data center has got to have security. It's got to have technicians. It's got to have security. It's got to have technicians. It's got to have facilities engineers. Right, and you know, I call myself a dinosaur, right? But I'm not the only one right. I'm again. We're all close to retirement, to college, and there's a huge opportunity for students right out of high school to go to trade schools and colleges.

Speaker 1: 31:21

That's an interesting point. Like you think of being in high school and you go. I've got a choice I go to college or I do maybe like a trade. Right, I might do like a plumber, electrician, a carpenter, whatever. There's a space here as well that you don't need to go to university or college. You could go and actually become Electrician.

Speaker 2: 31:42

HVAC technician. Yeah, we need welders. We need you know there are, if you think about it, for every data center that gets built, there's construction and you have to have all of these specialized trades.

Speaker 1: 32:00

Yeah, and even to the technicians that that have to do the you know customer opens a ticket or yes, when you're replacing a chip or you need to get something sent out there.

Speaker 2: 32:13

Remote hands and now we're going turning it on its head and saying oh, by the way, this server is also liquid-based.

Speaker 1: 32:20

On top of yeah, you have to understand and train it, yeah.

Speaker 2: 32:24

And again oh, I need to upgrade RAM in that machine AI is not doing that.

Speaker 1: 32:29

I'm sorry. Ai is not going to install your chip. No, ai needs humans. Yeah, yeah, thankfully.

Speaker 2: 32:35

So it's a great career path, I'm sure. Yeah, yeah, and again, you go the college route, you can learn. Yeah, even I think there are enough trade programs now, even on, say, on the networking side, right, where, like a lot of us are in the industry, didn't finish college and I mean I was learning more in the office every day.

Speaker 1: 33:00

Yeah, on the job, on the job, figuring it out.

Speaker 2: 33:03

Well, they were teaching me at university was yeah, it's probably.

Speaker 1: 33:07

Ten years prior, yeah, token ring or something, I don't know. As-400s, and that's the funniest part, like that's.

Speaker 2: 33:13

As-400s at the time were the cult, so they went through cycles right and everything goes back to the way it is here, we go right. It's just, you know, I mean, there's more computing power in your phone than there is on an AS400. That's good, yeah, and that's the other way to go, like could you program inference on a phone?

Speaker 1: 33:35

Yeah, that's the next phase. Can you get a chip that's punching out inference on that? Can you get a?

Speaker 2: 33:39

chip on your phone that can do the inference for you, and how is it trained?

Speaker 1: 33:44

How big is it? Can you stick a model on it? Can you keep up with any of the models? Maybe, yeah. Maybe, Apple becomes the inference.

Speaker 2: 33:52

Who knows? Yeah, I mean, these are crazy ideas, that. Yeah 20 years ago, you said you're not apple well, liquid cooled, you know smartphone, yeah, like that's probably not very unlikely but it's funny anyway, yeah, oh, cool, well, I appreciate you.

Speaker 1: 34:09

Uh a the partnership be coming on the show. Um, really interesting times. I mean, the last two years sounds like it's changed. I know I think it's done changing. Yeah, I think it's just gonna keep again, but it does feel like it's changed faster than the last three years, sounds like it's changed. I don't think it's done changing, I think it's just going to keep, but it does feel like it's changed faster than the last three years and it maybe didn't in the previous 10, but maybe not, I don't know.

Speaker 2: 34:27

Things were crazy. I would say going back to 2011, 2012,. Things were crazy yeah.

Speaker 1: 34:35

And then the push was to go to the cloud. 'm sorry, it's cool to be in the data center space, right? Oh yeah, yeah, kolo's cool again. Yeah, kolo's cool, awesome man, yeah, appreciate it.

Speaker 2: 34:46

Thanks, jack, take care.