[00:00:00] Andrew: How are you?
[00:00:01] Nathan: I'm all right. Not bad.
[00:00:02] Nathan: We were just talking before this, how did we get on [00:00:05] time
[00:00:05] Andrew: I know, right? We were on time, so we intentionally became late now. I have [00:00:10] the chats up,
[00:00:12] Andrew: so we'll be able to actually chat.
[00:00:14] Nathan: [00:00:15] Yes. And that might be a, a good thing 'cause we're, we're just chat- chatting about, yeah, what, what should [00:00:20] we talk about? And
[00:00:20] Nathan: hopefully people in the chat have things to throw in there. We have a, a few topics lined up, [00:00:25] but
[00:00:25] Andrew: I still don't have a producer, so I'm gonna be turning my head quite a bit to make sure everything's [00:00:30] working. But yeah.
[00:00:31] Nathan: Drive
[00:00:32] Andrew: I,
[00:00:32] Nathan: yeah
[00:00:33] Andrew: we were talking a little bit... Last [00:00:35] time we talked about aliens, UFOs, UAPs,
[00:00:39] Nathan: week? That was just [00:00:40] last week, yeah. Hmm.
[00:00:41] Andrew: no, no new news, folks. They're still not real or maybe [00:00:45] they've been here the whole time.
[00:00:47] Andrew: Um, but we'd closed out the stream by talking [00:00:50] through or I guess postulating on what infinite energy would [00:00:55] actually look like. So that's the claim with this, this kind of stuff. [00:01:00] And it starts to get really weird and, and I think there's an interesting crossover with AI like we were talking about [00:01:05] earlier that off stream.
[00:01:07] Andrew: But AI power [00:01:10] requirements and data centers and all this kind of stuff has been in the news quite a bit, and just [00:01:15] resource utilization in general. So I think it'd be an interesting, thing to chat about. And, and [00:01:20] more realistically it's like you were saying last stream, no matter what you always [00:01:25] have some sort of restriction.
[00:01:26] Andrew: So if you have unlimited power, you have space problems. If you have all the [00:01:30] space in the world, you then have distribution problems or power
[00:01:33] Nathan: Yeah. Right
[00:01:34] Andrew: [00:01:35] Yeah, I think maybe just to dispel some of this like AI data center stuff. [00:01:40] I hear a lot about how they use too much power, they pollute, [00:01:45] there's all these kind of issues.
[00:01:46] Andrew: And I think, just to break it down a little bit, [00:01:50] like a data center uses a ton of power to start. Nothing to do with AI, it [00:01:55] just uses a lot of
[00:01:55] Andrew: power.
[00:01:56] Nathan: Yes, in general. Yeah, yeah
[00:01:57] Andrew: Yeah. And [00:02:00] that's never really been a problem until now all of a sudden. So I [00:02:05] feel like it's just like the folks that hate AI for whatever reason are just like into [00:02:10] hating on it for this reason.
[00:02:12] Andrew: The number one utilization of [00:02:15] energy in our society is transportation, like full stop. So if you're really [00:02:20] concerned about the environment, don't drive your car
[00:02:22] Nathan: Yeah, yeah. I was, I was-- I think, I think you've talked [00:02:25] about this before,
[00:02:25] Nathan: but yeah, the whole idea the... basically the whole [00:02:30] resource chain that goes into something. I think a lot of people s- have a perspective of, of [00:02:35] some of these things don't, don't actually exist or, or whatever
[00:02:38] Nathan: Is whatever you get.
[00:02:38] Nathan: But there's a life [00:02:40] cycle that happens with it. And it's one of the arguments behind, I think AI when it first started [00:02:45] happening was a lot of people were talking about how much resources it uses in [00:02:50] general. And I think the underlooked one was like the amount of water it takes to, either cool it or to [00:02:55] produce the chips to actually do the, the AI
[00:02:58] Andrew: Yeah. Water's another one [00:03:00] too that, silicon manufacturing is really water intensive. [00:03:05] Always has been. I don't think the GPU stuff really matters [00:03:10] that much in terms of production. I think it's, yeah, there's more GPUs being produced, but, [00:03:15] we use less CPUs now. So I feel like in, in terms of actual unit [00:03:20] production,
[00:03:20] Nathan: out. Yeah,
[00:03:21] Nathan: yeah.
[00:03:21] Andrew: should be mostly a wash, like no pun intended
[00:03:24] Nathan: No, [00:03:25] you definitely intended the pun there.
[00:03:26] Andrew: I did. I was like, I could totally use that as a pun, so I did. [00:03:30] But, but I, the the data centers use water to cool and the, I think the biggest [00:03:35] change between a traditional data center and an AI data center is that GPUs use a lot more [00:03:40] power, but they also use a lot more water because most of these systems are [00:03:45] water-cooled instead of air-cooled.
[00:03:47] Nathan: Yep
[00:03:48] Andrew: Or I shouldn't say just [00:03:50] water-cooled, I should say liquid-cooled 'cause it's seldomly
[00:03:52] Nathan: Correct. I
[00:03:53] Andrew: Yeah
[00:03:54] Nathan: air [00:03:55] conditioning to keep the environment around it cool as
[00:03:57] Andrew: Yeah.
[00:03:57] Nathan: That's a part of it,
[00:03:58] Nathan: yeah
[00:03:59] Andrew: it, oh, that's ac- so [00:04:00] that's interesting. I'm trying to think of people who haven't been in a data center or have no idea what we're [00:04:05] talking about, like
[00:04:06] Nathan: I've never been in a data
[00:04:07] Andrew: how to maybe back this up a little bit for folks. [00:04:10] Yeah, a, a data center is effectively a giant building that has a lot of computers [00:04:15] in it.
[00:04:15] Nathan: Basically.
[00:04:16] Andrew: And those buildings are...
[00:04:18] Nathan: computers, yeah.
[00:04:19] Andrew: Yeah, [00:04:20] and they're, they might have multiple rooms or multiple locations within the building, but effectively it's like [00:04:25] a giant warehouse with a bunch of computers in it. A lot of cooling, a lot of power. There's [00:04:30] batteries for backup power and, and a, a bunch of other stuff too, but they're really [00:04:35] loud, and they're generally cooled by cooling towers, and [00:04:40] so that's a-- Every commercial building is cooled by a cooling tower typically, and they just [00:04:45] run water through a big heat exchanger, and that gets the heat from [00:04:50] outside and dumps it, or the heat from inside and dumps it outside.
[00:04:53] Andrew: Or if you need to run the reverse and [00:04:55] get heat, you can take the heat from outside, put it inside. And that's, that's how it's been [00:05:00] since, I don't know, probably 15, 20 years. That's just how they are. The, and, [00:05:05] and so the, all the servers, th- which are computers for folks if you're not really [00:05:10] familiar with them, are air-cooled typically.
[00:05:13] Andrew: So they just run air from one [00:05:15] side to the other, and all the hot stuff is in the middle, and then blows out the hot in the back and then they [00:05:20] cool cool that. These thing, these data center aisles, if you can [00:05:25] imagine a library. I know that dates some folks, but
[00:05:29] Nathan: People know what libraries look [00:05:30] like, right?
[00:05:30] Andrew: Do they though? Just Google it, so if you think of the books of [00:05:35] shelves, those could be servers, and if you think of one aisle being cold, the other aisle being [00:05:40] hot, you get the idea. Those aisles can get to be hundreds of degrees if they're not cooled [00:05:45] properly. And so it's a real-- It's actually a danger and it's a hazard to work in because you can get burned [00:05:50] and overheat.
[00:05:51] Andrew: Humidity is a big deal too, so they have dehumidification or [00:05:55] re-humidification depending on what they're trying to do. And so with AI, [00:06:00] the... It's-- there's two pieces of really AI in my mind. It's like the training of the [00:06:05] model itself, and then it's the running of it or The inference.
[00:06:08] Nathan: Yeah, yeah. Yeah,
[00:06:09] Andrew: And it's both [00:06:10] pretty similar.
[00:06:11] Andrew: Power utilization, energy when you're training [00:06:15] is a little bit different, but the idea is the same that you don't have [00:06:20] central processing units, CPUs, which are in servers and are good at doing anything. You have much GPUs [00:06:25] or tensors, which are effectively really good at doing matrix [00:06:30] math, but lots of, yeah, lots of simultaneous calculations [00:06:35] in, in very specific types of calculations versus CPU can, can run your web browser, your [00:06:40] email, play music, et cetera.
[00:06:42] Andrew: And so the move t- for a lot of this [00:06:45] is, is power, like the power densities. If you can, can think about the library like [00:06:50] the shelves tend to like row together, and so there's multiple physical shelves [00:06:55] in a row. In a data center, those are called a rack. And in a [00:07:00] rack, I think the power density for like modern AI [00:07:05] compute can be like three or 400 times what a standard [00:07:10] non-AI GPU rack can be.
[00:07:12] Andrew: You're talking like racks can get [00:07:15] into gigawatts in
[00:07:15] Nathan: Two watts. Yeah, yeah. Wow. Uh-huh
[00:07:17] Andrew: theory. And, and the, one [00:07:20] of the challenges with data centers is they don't actually have enough power in the building to power [00:07:25] a f- a d- a data center full of, of compute, let alone some of the racks. Like you literally [00:07:30] have cables, multiple inches across delivering power,
[00:07:34] Andrew: and that's hard to [00:07:35] deal with in general too.
[00:07:35] Andrew: So th- this is not like a really... They, I think people on it a lot, and they think they [00:07:40] make it sound like, oh, they're just like dumping a bunch of water in the data center and using electrons.
[00:07:44] Andrew: It's, [00:07:45] it's easy or something. No, it's incredibly challenging problem And so the power density is [00:07:50] much larger.
[00:07:50] Andrew: There's a lot more heat. Anytime you have energy, you have heat. And so a lot of these [00:07:55] systems are water-cooled or liquid-cooled. There's exotic gas systems and all this kind of [00:08:00] stuff. But I think that's where a lot of this, water hysteria comes [00:08:05] from. And so instead of just cooling the building, now you're cooling individual systems [00:08:10] with water or liquid.
[00:08:11] Andrew: And, I guess in theory there's more [00:08:15] usage 'cause evaporative cooling in the cooling towers does lose water because it's evaporative. But [00:08:20] every
[00:08:20] Andrew: corporate building does this in the world too. So again, if you're mad at [00:08:25] data centers, you should be mad at corporate office buildings too. But the, I think that the [00:08:30] challenge is that because there's more heat, there's theoretically more use.
[00:08:34] Andrew: But [00:08:35] one myth I kinda wanna dispel is they don't just open the water spigot and just [00:08:40] run fresh water through, and jettison dirty [00:08:45] water out or whatever people are imagining. They're relatively closed loops. They're not [00:08:50] 100% closed, but they're relatively closed. Whereas ironically enough, a nuclear power [00:08:55] plant does just what I explained.
[00:08:57] Andrew: It takes in beautifully [00:09:00] cold, fresh, clean water and then jettisons out really hot
[00:09:04] Nathan: it. Cycles [00:09:05] it,
[00:09:05] Andrew: water, and it's really bad for marine life. And so that's, that's generally how it works. And [00:09:10] so I think people need to understand that there's incentives for the operators of the data centers and the people who [00:09:15] use the systems in the data centers to use as little electricity as possible and to use as little [00:09:20] water as possible.
[00:09:21] Andrew: And I think, I don't know if we've talked about it in detail, but [00:09:25] Google rolled out a, an AI algorithm to control heating and cooling in [00:09:30] their data centers, and I think they moved stuff up by a quarter of a degree. So a [00:09:35] quarter of a degree warmer, they turned down the air basically and saved, it was, like, hundreds [00:09:40] of millions of dollars over whatever their budget cycle was.
[00:09:44] Andrew: And I don't [00:09:45] remember how many gigawatts of power because they weren't having to cool it. And they found that [00:09:50] systems were generally fine. Some systems would blow up faster, but it wasn't [00:09:55] any different than before they tested it,
[00:09:58] Nathan: Yeah, yeah.
[00:09:58] Andrew: save a bunch of [00:10:00] energy
[00:10:00] Nathan: Yeah, it's cool. It's cool on a,
[00:10:02] Nathan: on a scale level. I think that's also one thing that gets overlooked a lot [00:10:05] is, when you do something on an atomic level and you scale it out to the [00:10:10] X many places that you have that in in place, it, it adds up, right? And,
[00:10:14] Andrew: [00:10:15] Oh, 100%.
[00:10:15] Nathan: I think is, is overlooked too much
[00:10:17] Andrew: And the same thing, same's true for water, right? [00:10:20]
[00:10:20] Nathan: Mm-hmm.
[00:10:20] Andrew: they think they use a little bit more water initially, and then over the, of [00:10:25] course, evaporative cooling is a thing. And so they use a little bit more. [00:10:30] But the numbers I've seen coming out just don't make sense. They defy [00:10:35] reason. And I
[00:10:35] Andrew: think
[00:10:36] Nathan: the water usage?
[00:10:37] Andrew: for, for consumption, like the purported [00:10:40] consumption, right?
[00:10:41] Andrew: And you look at an almond or [00:10:45] a, a pistachio, right?
[00:10:46] Andrew: Like, it takes nine gallons of water to make a [00:10:50] pistachio, and I don't see people protesting pistachio farms
[00:10:53] Nathan: Because You can eat them and they're [00:10:55] tasty.
[00:10:55] Andrew: Yeah I don't... Yeah, sure. But I for me, I think [00:11:00] it, it's having a realistic conversation around a technology that [00:11:05] is way over-hyped.
[00:11:05] Andrew: We've talked about that a lot, right? It is a very over-hyped thing. But it [00:11:10] also has a lot of potential. It has a lot of use. You can argue with the [00:11:15] benefit i- is there or not, but I think to just be a Luddite and just say we're not ever gonna [00:11:20] touch any new thing because it uses more power is crazy.
[00:11:24] Andrew: Anyone who's [00:11:25] lived for 10 years, remembers laptops used to get 20 minutes of battery [00:11:30] life if you were lucky. Then it was, like, a few hours and you're like, "Wow, this is amazing." And [00:11:35] now my MacBook Pro will run two, three days on a charge [00:11:40] sometimes if I'm not doing...
[00:11:41] Andrew: If I'm just browsing the rat- web and stuff.
[00:11:43] Andrew: And so power [00:11:45] efficient- So if we would've just said, "No laptops because they use too much power," and then just [00:11:50] cut it off we would never have smartphones, we would never have anything else. We do have to [00:11:55] be, I think, to look at a perspective of it's a new technology, we need to figure [00:12:00] out how to make it better and make it more efficient, and I think that needs to be the goal.
[00:12:04] Andrew: I [00:12:05] do think that is the goal with most companies and most people doing this. Yes, I [00:12:10] think there's a little bit too much emphasis on we need more compute, so I'll compute at all costs. I don't [00:12:15] think that's where we need to be. But I also am getting increasingly [00:12:20] fra- frustrated with just the lack of thinking about this kind of stuff.
[00:12:24] Andrew: The, the [00:12:25] one is- interesting issue with data centers, and it, it happens up the road for [00:12:30] me here, is there tends to be atmospheric [00:12:35] changes above them in certain conditions. And so there's data center [00:12:40] clouds, like physical clouds that will manifest over data
[00:12:43] Andrew: centers, and it's [00:12:45] partly due to the evaporative cooling.
[00:12:47] Andrew: There's also,
[00:12:48] Andrew: um...
[00:12:48] Nathan: heat and it's messing[00:12:50]
[00:12:50] Andrew: Yeah, there, there's some questions about how much heat they're dumping into the atmosphere, and if there's an [00:12:55] actual ecological,
[00:12:56] Nathan: Right.
[00:12:57] Andrew: Change. And I'm sure there is, right? Anytime humans [00:13:00] build anything, there's an ecological change. It's just a question of what extent is it and can we d- design [00:13:05] stuff better to not have that problem versus the shut it down, it's [00:13:10] terrible.
[00:13:10] Andrew: And I I'm, I'm... So that for folks who, who don't know, The [00:13:15] Dalles, Oregon has a bunch of data centers. It's a whole thing. But [00:13:20] there was a study that was released oh, I don't know when it was released, but I read about it a couple weeks [00:13:25] ago. And, and so if you Google D-A-L-L A-S data [00:13:30] center environmental stuff, you'll probably find it.
[00:13:32] Andrew: Um, interesting... Dalles, [00:13:35] Oregon.
[00:13:35] Nathan: okay, okay, okay, okay.
[00:13:36] Andrew: Looks like Dallas, but it's called Dalles
[00:13:39] Nathan: Okay, I was
[00:13:39] Andrew: [00:13:40] We do things different out here. Yeah.
[00:13:42] Nathan: Okay, got it.
[00:13:43] Andrew: And it's The [00:13:45] Dalles.
[00:13:45] Nathan: Okay, cool.
[00:13:46] Andrew: Yeah. Yeah. And it's interesting because that area [00:13:50] of the Columbia River Gorge is also really well-studied because air currents are very [00:13:55] strange there because of the topography changes.
[00:13:57] Andrew: And totally unrelated maybe not [00:14:00] unrelated I went to a, a talk with a, a [00:14:05] climatologist meteorologist who's looking at climate, but looking at how to apply... And he, he [00:14:10] was working with a energy company to put in wind in the gorge.
[00:14:14] Nathan: Hmm[00:14:15]
[00:14:15] Andrew: And because the wind currents are so strange, they had to, to [00:14:20] build weird models to try to figure out, like, where, where to deploy them and how to best deploy them, [00:14:25] but then also not block the flowage of wind kind of [00:14:30] downrange because that would impact other farms.
[00:14:32] Andrew: So the actual, you
[00:14:33] Andrew: know, the effect of putting a wind [00:14:35] turbine up impacts wind currents. If you put enough of them up, you can actually [00:14:40] impact the next wind farm down the road.
[00:14:42] Nathan: Interesting. Yeah, yeah,
[00:14:43] Nathan: Yeah.
[00:14:44] Andrew: And
[00:14:44] Andrew: it, it [00:14:45] depe- there's a lot... The talk was, like, an hour, and my mind was just continuously blown [00:14:50] how much detail has to be done to just make this work, and it was fascinating.
[00:14:54] Nathan: [00:14:55] Yeah
[00:14:55] Andrew: And my point is, on one hand, it's like there's always some sort of [00:15:00] challenge and downside. There there's no magic technology that's perfect. But also we [00:15:05] learn stuff and then we change, and that's supposed to be what progressivism is about, right? Is [00:15:10] learning and changing. And I, I hear a lot of people who claim to be progressives [00:15:15] really shitting on things that are, are not progressive ideas.
[00:15:19] Andrew: And that you [00:15:20] don't have to be a, in my mind at least, you don't have to be, like, 100% into an [00:15:25] ideology.
[00:15:26] Andrew: But I have a hard time with people trying to, to claim being progressive [00:15:30] and then not doing progressive things or asking progressive questions or, more [00:15:35] importantly, pushing for progressive policies.
[00:15:37] Andrew: And so if these things are dumping a [00:15:40] lot of heat into the atmosphere, that's probably bad. I don't know. Maybe it's [00:15:45] good. Maybe there's a whole new ecosystem that erupt... This has been studied around nuclear plants, [00:15:50] right? Because the water coming in is cold, the water coming out is really, is much warmer, usually 5 or 10 [00:15:55] degrees warmer.
[00:15:56] Andrew: So one, climate change is really bad for nuclear because the water gets so warm that you [00:16:00] can't effectively cool the plant anymore and it becomes e- and they have to dial back power or stop [00:16:05] operations. But two, it changes the ecology around the plants, and [00:16:10] sometimes that's not bad. You get a different biome that can support [00:16:15] different kinds of life or help life that's endangered thrive.
[00:16:18] Andrew: This isn't just some [00:16:20] zero-sum thing, and it, it what is detrimental in one way might be beneficial in [00:16:25] another. And so we we can't,
[00:16:27] Nathan: Have your cake and eat it too thing
[00:16:29] Andrew: Yeah. [00:16:30] We have to look at it on, on net balance. Like anything humans do is gonna be detrimental to the [00:16:35] natural environment because we're changing it
[00:16:37] Nathan: Yeah, yeah. 'Cause we exist, right? Just by existing, [00:16:40] yeah, and doing stuff
[00:16:40] Andrew: Yeah. So I think that, that that needs to be studied. The power [00:16:45] thing's weird 'cause I, I think, yes, it, they do use a lot more power. I think that's [00:16:50] all the more reason to be investing in other kinds of energy sources. They should be
[00:16:53] Andrew: required in my mind [00:16:55] to put solar on the roof and if it makes sense.
[00:16:57] Andrew: Some- sometimes they go to Iceland or whatever because it's [00:17:00] cold, and I don't know if it makes sense to put solar on in Iceland just because of the angles of
[00:17:04] Andrew: the sun [00:17:05] and
[00:17:05] Nathan: Alaska, actually gets a lot of sun, I think y- despite being [00:17:10] yeah,
[00:17:10] Andrew: Yeah.
[00:17:11] Andrew: yeah.
[00:17:12] Andrew: So if it makes sense, I think it makes sense to do that. And I [00:17:15] think it also makes sense for them to guarantee some sort of a positive benefit [00:17:20] to the grid.
[00:17:21] Andrew: So instead of extracting energy out, they should have to supply [00:17:25] their own energy and then put surplus back into the grid so that the
[00:17:29] Nathan: Yeah
[00:17:29] Andrew: [00:17:30] can benefit from, energy growth. The weird part of that is if you put too much [00:17:35] energy into the grid and there's not enough use
[00:17:37] Andrew: for it, it explodes.
[00:17:38] Nathan: Yep, everything goes [00:17:40] down.
[00:17:40] Andrew: Literally explodes
[00:17:42] Nathan: Yeah. Yeah, basically feeding, feeding something into [00:17:45] a pipe that doesn't have enough Yeah.
[00:17:46] Nathan: to, to, to
[00:17:47] Nathan: take it all in.
[00:17:48] Andrew: it just goes boop
[00:17:48] Nathan: Yeah. [00:17:50] I I mean, I, I think that's the ideal and, and there are some discussions around these conversations [00:17:55] with companies trying to build data centers and not trying to put strain on the grid.
[00:17:59] Nathan: [00:18:00] Um, I think it needs to happen more often, but that's my silver lining in this [00:18:05] is
[00:18:05] Nathan: Given the desire for this compute and energy people who are trying to [00:18:10] build these data centers are actually thinking about, okay, actually, how do you make it a win-win? In order to make it a win-win, [00:18:15] you challenge what people are saying about
[00:18:16] Nathan: what the
[00:18:16] Nathan: data center will do.
[00:18:17] Andrew: thinking about it, to be [00:18:20] honest.
[00:18:20] Nathan: Okay. Though, though
[00:18:20] Nathan: I'm giving people benefit of the doubt here, and I'm
[00:18:22] Nathan: hoping that that's the idea. But i- and in [00:18:25] any negotiation, I think the ideal outcome is, is both parties win something, right? If you're [00:18:30] not getting the approval to build your thing, you're like, "Okay, what do I do to, to satisfy [00:18:35] whatever question they have or whatever doubt they have?"
[00:18:37] Nathan: And if that doubt just happens to be, oh, you'll be taking up more [00:18:40] energy, then naturally your solution is, okay, let me provide energy. And then maybe that sort of solves the, the, [00:18:45] the thing, right? But yeah, I, I do think that's the right way to do it, and that's what I'm hoping to [00:18:50] see come out of this is with the value of compute, that people will [00:18:55] actually want to generate the energy and give back to the people where these data centers are [00:19:00] going, because they do take space and you gotta put that somewhere. And that's hopefully what [00:19:05] we'll see happen more often.
[00:19:07] Nathan: But The, the thing with the-- I, I think [00:19:10] the-- I wanted to say something with this. The, we're talking about resources and the interesting thing about [00:19:15] data centers, whether they're AI or not, you can look at it as like an [00:19:20] optimization problem.
[00:19:21] Nathan: There's a metric where it's like, how much compute per square foot do you actually get? [00:19:25] It's like, how much power can you how closely can you pack your machines such that they don't [00:19:30] all light each other on fire, that you have enough cooling to be able to keep them all at operating temperatures, [00:19:35] right?
[00:19:35] Nathan: Because it takes so much space to, to build this stuff out. And so it's very [00:19:40] similar with, with AI data centers is what's the most compute you can get for the amount of square [00:19:45] footage. I, I think people have, humans in general, have this tendency to, to try and [00:19:50] min-max things, where it's once you have one particular metric,
[00:19:53] Nathan: let's do whatever we can to pump it [00:19:55] all up because, maybe because it, it brings you in money, right?
[00:19:59] Nathan: With the high, [00:20:00] quote-unquote, "demand" for AI
[00:20:01] Nathan: services, you want to put everything you can [00:20:05] into providing compute because that's the most expensive thing, or tokens, let's just say
[00:20:09] Andrew: Yeah,
[00:20:09] Nathan: because that's the [00:20:10] most expensive thing.
[00:20:11] Andrew: Yeah, and, and
[00:20:12] Nathan: need to flatten that out a little bit, right? It, it's, [00:20:15] it's
[00:20:15] Nathan: very much if you're investing in something, you don't want to put all your eggs in one basket, you want to build a portfolio.
[00:20:19] Andrew: And [00:20:20] it, and the, the data center thing, like back to my comment, I, I don't know that they [00:20:25] are thinking about that, but a progressive regulatory environment would say [00:20:30] for every X amount of energy you're gonna consume, you need to bring [00:20:35] back a tenth of a percent or whatever the formula needs to be to make sense.
[00:20:39] Andrew: 'Cause again, [00:20:40] you don't wanna bring so much energy that now it, you'd have surplus energy. That's actually really bad.[00:20:45]
[00:20:45] Andrew: But especially a lot of these smaller communities that have been fighting it in the US [00:20:50] here, there, there's a-- It's always on a resource grounds, which is, I find fascinating [00:20:55] because they don't seem to fight like manufacturing.
[00:20:58] Andrew: And like manufacturing is like [00:21:00] the number one use of fresh water and it becomes unfresh and unusable. [00:21:05] Whereas at least like data center evaporative cooling, as far as I understand, returns to the atmosphere [00:21:10] and theoretically comes back into the water cycle and it's not poisoned with a bunch of chemicals.
[00:21:14] Andrew: Like maybe it [00:21:15] is, I don't know. I don't understand that that's the true. But, so I, I [00:21:20] do think we need to be thinking about this in, in not a zero sum of yes, data center, no data [00:21:25] center. It's how can you use these to benefit your community? And then the data center people are just [00:21:30] needing to shut the up and understand that they are not in an extractive [00:21:35] economy.
[00:21:35] Nathan: I mean, good luck with that.
[00:21:36] Andrew: I know, right? Uh, I know, but that's [00:21:40] regulation and that's where I would like to see people's energy going towards not just yes or [00:21:45] no to, okay, could be beneficial for our community. They bring in jobs for [00:21:50] construction. I always think construction jobs are bullshit, not because of people who do construction being bullshit, but they're [00:21:55] temporary.
[00:21:56] Andrew: So yeah, you're employing people, awesome, until it's done, and then [00:22:00] what? So it's helpful, but they always we're make, 30,000 jobs are being created. Yeah, and [00:22:05] then they all go away in three years. Great, but also then we have a problem. So [00:22:10] again, it's like proportional growth. The, the thing that I really wanna maybe break [00:22:15] news on here that I have not heard anyone talk about is the [00:22:20] big problem with these-- The resource thing I think we've talked about.
[00:22:24] Andrew: I don't wanna minimize [00:22:25] it, but I think it's a lot less bad than people seem to think that it is, and I [00:22:30] haven't seen evidence To change my opinion, if you, you have evidence, put it [00:22:35] in the comment, email us, I'll read it, we'll talk about it. But I don't believe it until I see [00:22:40] it, and don't give me some fringe bullshit thing.
[00:22:43] Andrew: I want some sort of an empirical [00:22:45] study, that is also verifiable, 'cause both sides put out studies that it's the worst thing in the [00:22:50] world or it's the best thing in the world, and I don't know.
[00:22:52] Andrew: but I need to see evidence.
[00:22:53] Nathan: slicing [00:22:55] signs
[00:22:55] Andrew: yeah. A- and I think the number one thing in my mind that needs to really [00:23:00] change is, drum roll, how we [00:23:05] tax this stuff
[00:23:06] Nathan: Oh, okay. I- I've heard, I've, I've actually heard people talk about [00:23:10] this, taxing AI,
[00:23:11] Nathan: A
[00:23:11] Andrew: no, I don't wanna tax AI. I don't know. I don't really care about taxing AI. [00:23:15] That's not my idea
[00:23:16] Nathan: Okay. All right.
[00:23:17] Andrew: Yeah.
[00:23:18] Andrew: So the, the-- You're on the right track though, [00:23:20] and I think people are on the right track in terms of I don't know. A lot, a lot of the tax AI [00:23:25] arguments I hear are nonsense. So I guess my idea, or I think, not nec- I don't know if I have [00:23:30] necessarily a solution, but I think the problem that we need to reconcile is that [00:23:35] traditional city, town, land, you have land that [00:23:40] is in your city limit.
[00:23:40] Andrew: You basically proclaim, "We control this area," and what happens in there is [00:23:45] controlled through the city. You have zoning, you have planned development, you have [00:23:50] unplanned development, but it's all typically controlled through zoning. You say, "This land can be used [00:23:55] for this kind of stuff." Mostly residential.
[00:23:57] Andrew: Everybody lives in a house and knows what that [00:24:00] is. You, you have all sorts of different types of light industrial where you can have a [00:24:05] shop,
[00:24:05] Andrew: Uh, where you go clothe, shop for clothes or something. That might be in your neighborhood or like a [00:24:10] coffee shop or a restaurant versus a manufacturing plant or like a power [00:24:15] plant or something like that The problem I think with a lot of the data center stuff is that [00:24:20] they're, they don't-- The cities haven't figured out how to zone correctly, [00:24:25] so they, they'll, they'll try to go into areas where they have industrial zoning, [00:24:30] and it's technically an industrial thing.
[00:24:32] Andrew: But when cities zone industrial, [00:24:35] they tend to be wanting to attract businesses that will create value.
[00:24:38] Nathan: 嗯。
[00:24:39] Andrew: And [00:24:40] value tends to be created by people doing work,
[00:24:42] Nathan: Yeah, yeah
[00:24:43] Andrew: right? And so the, they data centers [00:24:45] will pl- employ 30,000 people. Then they all go away and build the next one. You have maybe five, [00:24:50] 10 people who actually work in the actual data center itself.
[00:24:53] Andrew: V- very few people actually [00:24:55] work in these things once they're up and running. So your net employment is quite low. Your [00:25:00] utilization of, of land or investment is quite high. And then you, you [00:25:05] tax the property, right? So you tax the building,
[00:25:08] Andrew: rough- in most [00:25:10] jurisdictions.
[00:25:11] Andrew: And s- and so normally that's fine, right?
[00:25:12] Andrew: 'Cause you have your warehouse, and let's just say [00:25:15] you're, you're taking agricultural product and packaging it up and [00:25:20] making frozen peas, right? So you're, you're taxing the peas that come in when the [00:25:25] farmer dumps them off. You're taxing that product probably going out. And taxes work [00:25:30] differently everywhere, but this is a general idea.
[00:25:32] Andrew: Then you have a bunch of people in the building who are [00:25:35] shifting the peas around and washing them and cooking them and bagging them up, and you [00:25:40] tax their wages.
[00:25:41] Andrew: And then they turn around and go eat lunch at the [00:25:45] cafe. They get groceries in the, the grocery store. They go to the doctor. They [00:25:50] buy new clothes for their kids.
[00:25:52] Andrew: There's a whole economy that happens. With data centers, that [00:25:55] economy is relatively absent, and any economic value [00:26:00] gets extracted out into Silicon Valley for the most part. And so [00:26:05] I think we need to be looking at not necessarily how do we tax AI or agents or [00:26:10] value, but in an information economy, which is what we're in, in a high-tech, [00:26:15] quote-unquote, economy, the inputs and outputs taxation is [00:26:20] not really working anymore.
[00:26:22] Andrew: And I'm not convinced that it's the right way to do [00:26:25] it. Maybe it is. I'm happy to be wrong on that. I think we need to [00:26:30] look at generated output or economic value of the data center, 'cause the data center's probably [00:26:35] generating hundreds of millions of dollars, if not billions of dollars in, in value [00:26:40] to the world, your company, to multiple companies.
[00:26:44] Nathan: Hmm.
[00:26:44] Andrew: out [00:26:45] a way then for that, some of that value to come back into these communities.
[00:26:49] Nathan: Yeah, [00:26:50] yeah.
[00:26:50] Nathan: I I, think, yeah, I think your, your idea is, is on... [00:26:55] We need to do something around it, right? I can't proclaim to be any tax expert at all,
[00:26:58] Nathan: but I think [00:27:00] it's, like whether you're changing the tax, tax law or the zoning, which then applies to [00:27:05] tax, right? Because you can't blink a tax.
[00:27:06] Nathan: And the other thing is you can't really know what the economic [00:27:10] value of a data center would be ' cause it depends on what, what you're serving, right? So it's like, how do [00:27:15] you even know what tax, at what tax percentage you should even... Or [00:27:20] maybe it's not even a percentage anymore. I don't know. But I would actually be
[00:27:23] Nathan: very curious if anyone knows more [00:27:25] about this stuff because I, I, I...
[00:27:27] Nathan: This is not, this is something I know
[00:27:28] Nathan: nothing about, but
[00:27:29] Nathan: it certainly [00:27:30] has an
[00:27:30] Andrew: have the, the idea of a use tax. A lot of states will have that [00:27:35] where, the equipment that you use to generate value or revenue or just [00:27:40] work, is taxed.
[00:27:42] Andrew: And so I think-
[00:27:43] Nathan: be, these are all like state, [00:27:45] on st- on the state level, right? So then if you're doing something that is very mobile in terms
[00:27:49] Andrew: I've lived in cities that have use [00:27:50] tax. It's weird.
[00:27:52] Nathan: Yeah, yeah.
[00:27:52] Nathan: But, it's all fancy ways to skirt [00:27:55] property tax or perhi- prohibitions on property tax. But my point is [00:28:00] that's usually where people go is we're gonna tax the servers." I, I think, again, the [00:28:05] use tax makes sense for a car, which is where they often are applied, because you're driving on a [00:28:10] road, you're causing damage to the road, you need to build new roads and bridges, you have to build new infrastructure, so everybody contributes.
[00:28:14] Andrew: [00:28:15] That kinda makes sense, but taxing the server, it doesn't, 'cause that server could be [00:28:20] generating $100 million in value by itself, one little server.
[00:28:23] Andrew: And, so but my, [00:28:25] my argument is that then the people in that community with the data center should get [00:28:30] value returned, and then also compensated for the lack of water or power or [00:28:35] whatever their purported problems are.
[00:28:37] Andrew: ' Cause if your pea factory, for example, [00:28:40] was taking up all the town water, you can go to the pea factory and be like, "You're making less [00:28:45] peas," or, "You have to pay us back for the harm that you're causing to the well," or [00:28:50] whatever, right? Then there's also accountability then
[00:28:53] Nathan: Again, I would, I would love to [00:28:55] talk to an expert on this, but w- can you just simplify all that without doing the tax stuff [00:29:00] with just something like a co-op model where you have a co-op or a, a community which [00:29:05] decides or, or all agree that a data center can go there and any proceeds that that [00:29:10] data center generates can be split amongst the people in that community?
[00:29:13] Andrew: Yeah you could do that. It won't [00:29:15] happen.
[00:29:15] Nathan: Sure.
[00:29:16] Andrew: I don't care how it's done. I think the [00:29:20] idea is that you tax the building and then you just move [00:29:25] on is outdated. And I think we need to be having the conversation [00:29:30] around what is the value of your work generating, and how do you return some of that [00:29:35] value to the local community? And it's not a redistribution of wealth because this is how it [00:29:40] works everywhere else.
[00:29:41] Andrew: The fundamental problem being is the electrons I'm pushing and the power [00:29:45] that I'm computing on is not easily [00:29:50] understandable or producing physical outputs that we would then [00:29:55] generate value from
[00:29:57] Nathan: But in, in the
[00:29:57] Nathan: context of something like a tax, [00:30:00] yeah the con- because we live in a space of like the, the services that [00:30:05] you're providing through a data center are virtual, but your data center itself has to reside [00:30:10] somewhere physically. if you h- unless you did every place, there's always gonna be [00:30:15] some space or some, some country arbitrage that you can do, right?
[00:30:19] Nathan: So if, if [00:30:20] your specific state has this use tax for AI or for data centers or whatever, as an [00:30:25] org- as an, as an organization who's building this, I'm, I'm just not gonna go to your state, right? I'm just gonna build somewhere else. I can still export the compute to your state[00:30:30]
[00:30:30] Andrew: Yeah, no, and that's, again, taxing [00:30:35] AI is not the right way to do it.
[00:30:37] Andrew: We have to figure out a better way [00:30:40] to recoup money for communities
[00:30:44] Nathan: I agree with
[00:30:44] Andrew: based [00:30:45] on modern outputs, right? And that's the core problem.
[00:30:48] Andrew: And, and you, you've totally [00:30:50] nailed it. The reality though is these things exist and they have to be somewhere.
[00:30:54] Andrew: [00:30:55] And you're not gonna just go build it in I- Idaho or Montana or something because you have to [00:31:00] have a lot of other infrastructure required, and a lot of that infrastructure also [00:31:05] supplies other things,
[00:31:07] Andrew: right?
[00:31:08] Andrew: And so-- And I think [00:31:10] this kind of hits on a, a larger societal problem of the information economy is there's a group of [00:31:15] people like, we're in there, you and I,
[00:31:17] Andrew: that we work in the information economy [00:31:20] and it's just different.
[00:31:21] Andrew: And whereas before I can only [00:31:25] make so many pea bags in a day before I pass out or whatever [00:31:30] with... And, but there's only so much direct value I can contribute.
[00:31:34] Nathan: Yeah.
[00:31:34] Andrew: And [00:31:35] so all of our models--
[00:31:36] Nathan: by the way.
[00:31:37] Andrew: I know, right?
[00:31:37] Nathan: tea bags before you pass out.
[00:31:39] Andrew: Hell [00:31:40] yeah. And so the, that output is, is [00:31:45] also finite and localized, and then it-- And our, all of our [00:31:50] systems of, of paying for the society we all live in, everybody hates the T [00:31:55] word, but the reality is that's how we pay for stuff.
[00:31:58] Andrew: This idea that malevolent [00:32:00] people with means are gonna somehow make everything better is insane. It's never worked that way ever.
[00:32:04] Nathan: [00:32:05] that's how countries look, right?
[00:32:06] Andrew: But that model's outdated because you and I can sit on a keyboard, I can work [00:32:10] for 10 minutes and produce hundreds of millions of dollars in value,
[00:32:13] Nathan: Yeah,
[00:32:13] Andrew: right? And we don't [00:32:15] have a system to really deal with that.
[00:32:17] Andrew: And so my c- if I'm doing that, my community should also [00:32:20] benefit from that. We've-- That's a whole problem with the, the T word is that [00:32:25] people seem to think that if you win, you should just be able to take all your chips home and then never play the game [00:32:30] again. That's insane. And so we have to figure this problem out, and I think it's only [00:32:35] getting worse, and we're s- we're s- there's a lot of talk about income redistribution and wealth.[00:32:40]
[00:32:40] Andrew: That is, that's a whole thing. But I think a tangible thing folks can actually work on that's an, [00:32:45] an immediate problem is these data centers, and figure out not ways to punish [00:32:50] them, but figure out ways that you would adopt your regulations so [00:32:55] that you can have a data center built but also not lose all the value from [00:33:00] that.
[00:33:00] Andrew: 'Cause I think a, a city a city that has a data center [00:33:05] in it has police, has fire protection. Like, all of these services that would go to the [00:33:10] P factory are the same services that a data center also needs so that the data center, if it does catch on [00:33:15] fire, they need someone to come put it out. But they need to contribute back to that.
[00:33:18] Nathan: Yeah, yeah, yeah. [00:33:20] And the other way around too, right? We may not think of places like fi- police departments or fire [00:33:25] departments needing technology, but yeah, you-- they can certainly leverage the compute that comes from that, [00:33:30] that data center,
[00:33:31] Andrew: 100%. And, and I guess you give credits or something. Like the-- I [00:33:35] think the companies who are building these need to be doing more, and they won't voluntarily, which is again why I'm [00:33:40] like if, oh, they're just gonna give us stuff," that. You have to make them. They, they, they won't, [00:33:45] otherwise they would be.
[00:33:46] Andrew: And even then, it all goes away at some point, right? So you, you need to [00:33:50] like... In a way for that, that's an interesting idea. It's like a way you could do it is you [00:33:55] have free AI in your school, right? All the students get [00:34:00] credits
[00:34:00] Andrew: to to learn AI right? Just like the pea factory, like they're probably [00:34:05] selling peas in your local grocery store at a much lower price than the big [00:34:10] supermarket, and the community then gains a better, cheaper product because [00:34:15] it's more localized.
[00:34:16] Andrew: We need to return these ideas, and we need to adopt them for the digital [00:34:20] age instead of just trying to invent a whole new system or have some magical [00:34:25] benevolent thing come along and That's-- 'cause that's what it always comes to, right? Is [00:34:30] yeah, aliens, communism, fascism is [00:34:35] everybody wants some magical thing to happen, and it's okay, or we could just adopt the system that we [00:34:40] already know works.
[00:34:41] Andrew: But it's so hard to understand this stuff and [00:34:45] I'm trying to play my part in this to give ideas and, and give hopefully [00:34:50] inspiration for folks. Because I watch this we've talked about this too. I watch people in Congress, they [00:34:55] don't have any clue how the economy actually works.
[00:34:57] Andrew: They're all 80-something years old, and they [00:35:00] can barely run a phone. And I'm like, I literally contribute massive amounts of [00:35:05] value by not leaving my office. And that value is perpetual, [00:35:10] right? That's on a whole 'nother-- and these are concepts that most humans are really-- well, all humans are really hard at [00:35:15] getting, but the-- it's really, really hard to, to grok what this even means
[00:35:19] Nathan: Yeah, [00:35:20] yeah, yeah
[00:35:21] Andrew: Should we leave it there? Aliens.
[00:35:23] Nathan: doubt it. Can't doubt it. Yeah, [00:35:25] yeah
[00:35:25] Andrew: Cool. Happy Friday.
[00:35:27] Nathan: All right. Yeah, you guys all have a nice weekend
[00:35:28] Andrew: All right. Take care