WEBVTT - Data Centers, Crypto Miners, and Gamers Are All Battling for Semiconductors 0:00:10.640 --> 0:00:15.000 Hello, and welcome to another episode of the Odd Lots podcast. 0:00:15.160 --> 0:00:20.720 I'm Joe Wisenthal. Unfortunately, my colleague Tracy Alloway, she is 0:00:20.800 --> 0:00:24.560 off today and is particularly unfortunate because this is gonna 0:00:24.600 --> 0:00:27.720 be a big one. So, you know, like on this podcast, 0:00:27.880 --> 0:00:31.360 obviously we talk about chips a lot, and I think 0:00:31.400 --> 0:00:35.279 now we've done like eight different episodes on semiconductors and 0:00:35.640 --> 0:00:38.400 who makes them and the shortages and the challenges of 0:00:38.400 --> 0:00:41.279 building them and so forth. So Odd Lots listeners will 0:00:41.320 --> 0:00:44.320 know that we also do a fair number of episodes 0:00:44.320 --> 0:00:48.040 about cryptocurrencies, and of course that's for obvious reasons why 0:00:48.280 --> 0:00:50.880 cryptocracy has been going up a lot and it's a 0:00:50.920 --> 0:00:55.240 fascinating area. Well, this episode, I actually I'm gonna say 0:00:55.240 --> 0:01:00.240 it actually combines the two because it's slowly becoming or 0:01:00.800 --> 0:01:05.319 that you can't really talk about the semiconductor market right 0:01:05.360 --> 0:01:08.760 now without talking about cryptocurrencies because you need chips to 0:01:08.800 --> 0:01:11.919 mind them. And when they said, when the cryptocurrency market 0:01:12.040 --> 0:01:15.880 is booming, people want to acquire more chips to mind them, 0:01:15.880 --> 0:01:19.440 and it's becoming a sort of more significant source of demand. 0:01:19.840 --> 0:01:22.840 If you look in the video gamer community, there's a 0:01:22.920 --> 0:01:27.200 huge frustration about their inability to buy various graphics cards 0:01:27.240 --> 0:01:30.800 from the likes of Nvidia because etherory of miners snap 0:01:30.880 --> 0:01:35.000 them up. Recently, there have been stories out of Asia, 0:01:35.400 --> 0:01:39.640 China in particular, where people are are buying hard drives 0:01:39.680 --> 0:01:43.959 because they want to mind the new cryptocurrency Chia Bitcoin 0:01:44.160 --> 0:01:48.120 is mind using specialized chips, but ultimately that takes a 0:01:48.320 --> 0:01:52.840 chip production capacity to so really the stories are interlocked, 0:01:53.120 --> 0:01:55.600 I think in a way that I hadn't up until 0:01:55.640 --> 0:01:59.000 recently fully appreciated, and I not I would not say 0:01:59.040 --> 0:02:02.880 that cryptos main source of reason why there is a 0:02:02.960 --> 0:02:06.640 shortage or difficulty to get a semiconductors, but everything matters 0:02:06.760 --> 0:02:09.840 at the margin, and so we're going to explore this 0:02:10.040 --> 0:02:13.160 nexus today between what is going on in the world 0:02:13.160 --> 0:02:17.600 of semiconductors, the fight to acquire semiconductors between different players. 0:02:17.639 --> 0:02:20.880 Because you have the minors, you have the gamers, people 0:02:20.960 --> 0:02:24.120 use computers, you have data centers, which of course have 0:02:24.280 --> 0:02:28.000 tremendous chip needs, and they're all fighting it out for 0:02:28.080 --> 0:02:32.600 this semiconductor capacity and it all comes together and it's 0:02:32.600 --> 0:02:36.000 a really fascinating facet of all of these different stories 0:02:36.160 --> 0:02:40.359 brought together. I'm very excited about this episode. I am 0:02:40.440 --> 0:02:42.920 going to be speaking with Brian Venturo. He is the 0:02:43.000 --> 0:02:47.320 CTO and co founder of core Weave, which is uh 0:02:47.320 --> 0:02:50.000 it's a cloud computing company, and we'll get into the 0:02:50.040 --> 0:02:52.680 details of specifically what they do, but as a cloud 0:02:52.680 --> 0:02:56.120 computing company, of course, it's specifically has a lot of 0:02:56.200 --> 0:02:58.959 chip needs itself, and so he's gonna walk us through 0:02:59.000 --> 0:03:01.400 what the world looks like right now. He's also in 0:03:01.480 --> 0:03:03.560 the ethery of mining space, sees it from all the 0:03:03.800 --> 0:03:07.400 all the different angles. We're going to talk about that. So, Brian, 0:03:07.680 --> 0:03:10.600 thank you so much for coming on odd lots. Thank 0:03:10.600 --> 0:03:13.200 you so much for having me. This is a topic 0:03:13.280 --> 0:03:16.519 that I am near and unfortunately dear too, so I 0:03:16.680 --> 0:03:19.200 really excited to dig into it. So why don't you 0:03:19.240 --> 0:03:22.160 just get us started? I mean cloud computing. You know, 0:03:22.200 --> 0:03:26.280 people think of Amazon Aws, people think of Microsoft perhaps, 0:03:26.360 --> 0:03:29.320 but obviously they're not the only players. What is core 0:03:29.400 --> 0:03:35.760 Weave and how is it situated within the cloud computing ecosystem? 0:03:35.840 --> 0:03:40.280 We typically talk about ourselves as a specialized cloud right, 0:03:40.360 --> 0:03:45.800 so we're primarily providing highly scalable, highly paralyzable burst compute too. 0:03:46.480 --> 0:03:50.200 Companies that are in the VFX production and rendering space. 0:03:50.440 --> 0:03:54.680 That's VFX. That's video videographics correct, Like, so pretty much 0:03:54.720 --> 0:03:57.480 anything that you see that's episodic on a streaming channel, 0:03:57.920 --> 0:04:01.160 for that's being produced in long form content like that's 0:04:01.200 --> 0:04:03.600 being touched by a VFX artist somewhere, whether it's on 0:04:03.600 --> 0:04:06.280 the compositing layer or it is actually on the computer 0:04:06.320 --> 0:04:10.240 generated graphics layer. That stuff needs a significant amount of compute, 0:04:10.480 --> 0:04:12.480 and that's one of our focus areas as a business, 0:04:12.480 --> 0:04:15.600 so we're there for them. And then on in addition 0:04:15.600 --> 0:04:18.120 to that, it's AI machine learning, so it's model training 0:04:18.160 --> 0:04:21.359 and serving. There's one additional one that kind of sits 0:04:21.640 --> 0:04:25.960 between the media and entertainment space and the kind of 0:04:26.000 --> 0:04:28.400 batch processing space, which is called pixel streaming, which is 0:04:28.920 --> 0:04:32.479 serving real time experiences in your browser. Right, So if 0:04:32.520 --> 0:04:35.000 you ever go to configure a car online, they don't 0:04:35.040 --> 0:04:38.080 have seven million versions of that car render. They actually 0:04:38.080 --> 0:04:40.120 do it in real time on a GPU someone in 0:04:40.200 --> 0:04:42.520 the data. And then the last piece is kind of 0:04:42.520 --> 0:04:46.680 the legacy HPC, right, So it's highly parallelizable batch processing. 0:04:47.080 --> 0:04:50.240 It's people who are doing drug discovery, if they're doing 0:04:50.360 --> 0:04:53.680 cancer research, kind of everything under the sun. What makes 0:04:53.760 --> 0:04:56.440 us a little bit different is that we're not really 0:04:56.440 --> 0:04:59.840 competing for people to host their word press blugs. Right, 0:05:00.000 --> 0:05:03.200 So we're building for we're building scale compute for large 0:05:03.200 --> 0:05:06.000 customers of compute Um, and then we're kind of working 0:05:06.000 --> 0:05:07.480 with them all along the way to kind of meet 0:05:07.480 --> 0:05:09.359 their needs. And we do a lot of the spoke 0:05:09.400 --> 0:05:12.320 builds for people. So we're constantly in the in the 0:05:12.400 --> 0:05:15.640 chip market and in the component market. And just because 0:05:15.640 --> 0:05:18.159 every industry kind of has their own unique needs. So 0:05:18.720 --> 0:05:21.880 is there such a thing as a sort of a 0:05:21.920 --> 0:05:24.880 normal chip market? I mean right now, and we'll get 0:05:24.880 --> 0:05:27.120 we're going to get very deep into this, and right now, 0:05:27.160 --> 0:05:30.679 we know that there's incredible amount of demand for chips 0:05:30.760 --> 0:05:35.040 for cryptocurrency mining. But you know, go back, I don't know, 0:05:36.520 --> 0:05:39.080 it was a bear market for crypto There wasn't a 0:05:39.080 --> 0:05:42.360 lot of enthusiasm at the time. Talk to us about 0:05:42.400 --> 0:05:46.840 what a sort of normal chip market looks like and 0:05:47.040 --> 0:05:50.320 what your needs are as a company on a stay 0:05:50.360 --> 0:05:53.600 annual basis to acquire chips such that you can provide 0:05:53.600 --> 0:05:56.279 cloud services for your customers. Let's start with the normal 0:05:56.279 --> 0:05:59.200 world first. Yeah, So let's let's start off by framing 0:05:59.240 --> 0:06:02.799 the kind of ore and all needs perfect. We're probably 0:06:02.839 --> 0:06:05.320 buying between like let's call it seven and ten thousand 0:06:05.360 --> 0:06:08.800 servers a year. That's seven ten thousand servers is going 0:06:08.839 --> 0:06:11.039 to be built two different ways, right, One portion of that, 0:06:11.080 --> 0:06:14.120 we're going to build a house, which is typically like 0:06:14.160 --> 0:06:17.000 a really dense GPU compute, and then the second piece 0:06:17.080 --> 0:06:19.480 is going to be kind of more generic CPU builds 0:06:19.520 --> 0:06:22.520 that we buy for ancillary services and other things that 0:06:22.560 --> 0:06:25.320 may not be GPU focused, and that's typically from regular 0:06:25.400 --> 0:06:29.320 like O E m's like super Micro, HP, Dell, whoever 0:06:29.400 --> 0:06:32.279 that may be. What does the typical market look like? Well, 0:06:32.480 --> 0:06:34.880 you know, like sitting here today, I would I would 0:06:34.880 --> 0:06:37.680 die to go back to like six months ago. The 0:06:37.680 --> 0:06:40.039 typical market is a place where you can where things 0:06:40.040 --> 0:06:43.240 are in stock, and you may have trouble finding them, 0:06:43.279 --> 0:06:45.599 but it may take you a couple of hours and 0:06:45.680 --> 0:06:47.520 you can find the stock that you need and haven't 0:06:47.520 --> 0:06:50.520 delivered the next day. It's not really like that right now. 0:06:50.880 --> 0:06:54.080 I spend most of it's a little while that half 0:06:54.160 --> 0:06:56.520 my job here is really scouring the internet to try 0:06:56.520 --> 0:06:58.640 to find people that may have stocked. Like if I'm 0:06:58.680 --> 0:07:00.680 looking for a certain item, whether it's a CPU, whether 0:07:00.720 --> 0:07:04.280 it's RAM, whether it's a special motherboard that we're looking 0:07:04.320 --> 0:07:06.240 for you know, I go through a kind of our 0:07:06.279 --> 0:07:09.240 regular partner channels first, which is some of the large distributors, 0:07:09.880 --> 0:07:12.040 and then typically right now, I get told there's no 0:07:12.120 --> 0:07:14.200 chance to get that within sixteen weeks, and then I 0:07:14.240 --> 0:07:16.760 go to eBay, right and I don't go to eBay 0:07:16.760 --> 0:07:19.120 to buy the stuff. I go to eBay to just say, Okay, 0:07:19.200 --> 0:07:21.520 is there a supply out there? How do I source it? 0:07:22.080 --> 0:07:24.240 Because it's really hard to kind of connect buyers and 0:07:24.280 --> 0:07:26.200 sellers in this market because there's so many people that 0:07:26.280 --> 0:07:30.160 provide services and our distributors and resellers. So it's really 0:07:30.200 --> 0:07:32.480 like okay, like get a feel for what's out there, 0:07:32.520 --> 0:07:35.640 and then go hunt it down. So it's it's pretty wild. 0:07:36.320 --> 0:07:38.320 Like last year, it would be no problem for me 0:07:38.360 --> 0:07:41.000 to go out and get ten terabytes of RAM right 0:07:41.040 --> 0:07:43.400 and like sixty four gig dems of ten of RAM 0:07:43.440 --> 0:07:45.560 and build up ten terabytes and get it overnight it 0:07:45.600 --> 0:07:48.440 to me if we needed it for a build. And like, 0:07:48.480 --> 0:07:51.080 I mean, there was a time late last year where 0:07:51.080 --> 0:07:53.240 we actually had somebody get on an airplane from California 0:07:53.240 --> 0:07:55.400 to Chicago because we needed something done in like three hours. 0:07:55.920 --> 0:07:58.600 That's just not possible today. Wait, did that person have 0:07:59.760 --> 0:08:02.000 a chip in or some sort of chip set in 0:08:02.040 --> 0:08:04.560 there that they just traveled. They just got on the 0:08:04.560 --> 0:08:07.200 plane and brought it to Yeah, we needed that. We 0:08:07.240 --> 0:08:09.160 had to deliver something for a client. It was a 0:08:09.240 --> 0:08:12.400 late late on a Friday night and we needed I 0:08:12.400 --> 0:08:14.440 think it may have been like twenty terabytes of ramp okay, 0:08:14.520 --> 0:08:17.120 and we found a supplier in California that was still open, 0:08:17.200 --> 0:08:19.800 that had it in stock, and we we paid them 0:08:19.800 --> 0:08:22.720 extra to put them on a plane. Wow. Right, So 0:08:23.280 --> 0:08:24.960 a lot of what we do is very much just 0:08:25.040 --> 0:08:28.800 in time. Yeah. Right, So we we've done dumb stuff 0:08:28.840 --> 0:08:31.080 like that before. But if I needed to do that today, 0:08:31.120 --> 0:08:32.800 like I don't even think I could source the ramp. 0:08:33.200 --> 0:08:35.959 So again, we'll get to the today. But I want 0:08:35.960 --> 0:08:39.000 to keep diving further into the normal market. When you 0:08:39.040 --> 0:08:45.160 talk about GPUs, CPUs, RAM, talk to us a little 0:08:45.200 --> 0:08:48.600 bit about the difference between the two GPU and CPU 0:08:48.800 --> 0:08:52.080 and how they serve different types of applications for your customers. 0:08:52.400 --> 0:08:54.800 Let's start with the VFX industry. Okay, right there, it's 0:08:54.880 --> 0:08:57.800 really usable there for two different things. The first is 0:08:58.000 --> 0:09:01.800 for virtual workstations where artists are actually now fully migrated 0:09:01.840 --> 0:09:03.800 to the cloud, and working in the cloud on a 0:09:04.480 --> 0:09:07.000 on a desktop that that we've helped them set up. 0:09:07.559 --> 0:09:09.720 So you know, we have hundreds and hundreds of artists 0:09:09.760 --> 0:09:13.440 doing that today. They connect to us via a low 0:09:13.520 --> 0:09:16.600 latency application from their laptop or from their desktop at 0:09:16.640 --> 0:09:18.720 home or whatever it may be, and then they're connected 0:09:18.760 --> 0:09:21.800 to a hardware accelerated desktop in our cloud. That's one 0:09:21.880 --> 0:09:24.160 use for the GPU, right, The GPU provides the hardware 0:09:24.200 --> 0:09:26.400 acceleration of the desktop as well as the graphics for 0:09:26.440 --> 0:09:28.880 what they're doing. And then the second piece for for 0:09:28.920 --> 0:09:31.480 GPU compute on the VFX side is for the actual 0:09:31.520 --> 0:09:34.640 rendering itself. For us, we actually spend kind of two 0:09:34.679 --> 0:09:36.800 different markets in the rendering side, Like one is for 0:09:36.800 --> 0:09:39.840 for large studios where they have people rendering every day, 0:09:40.240 --> 0:09:42.920 and then we also operate a director consumer render farm 0:09:43.160 --> 0:09:47.240 for small freelancers and artists that has like just over 0:09:47.280 --> 0:09:50.480 twenty users at this point. On both sides, though, the 0:09:50.480 --> 0:09:54.400 demand there is is is bursty and crazy, where hey, 0:09:54.440 --> 0:09:56.520 they need to get a shot done, um, they need 0:09:56.520 --> 0:09:58.360 it done in the next two hours to meet a deadline, 0:09:58.480 --> 0:10:00.280 and if they were to do it on our own 0:10:00.280 --> 0:10:02.720 computer in their own farm, and may take them forty 0:10:02.800 --> 0:10:05.520 days we really become the only option at that point 0:10:05.559 --> 0:10:08.160 for some folks. UM, and you know, it's my job 0:10:08.200 --> 0:10:10.440 here to make sure that I have the GPS, the 0:10:10.480 --> 0:10:13.800 scale of GPS available to be able to handle a 0:10:13.880 --> 0:10:16.760 large number of concurrent users like that. I'm looking at 0:10:16.880 --> 0:10:20.199 your website and you know, scrolling down, one of the 0:10:20.200 --> 0:10:22.800 thing is is the GPUs you need when you need them, 0:10:22.840 --> 0:10:26.960 And you say, we currently have over in video GPUs. 0:10:27.120 --> 0:10:30.000 And so when we talk about this battle for chips 0:10:30.080 --> 0:10:34.240 between gamers and miners, these and I don't know if 0:10:34.240 --> 0:10:38.680 they're precisely the same, but generally speaking, these chips that 0:10:38.800 --> 0:10:42.320 you have are like the chips that everyone covets. No, 0:10:43.400 --> 0:10:46.280 they're not, right, So they're very similar in nature, but 0:10:46.400 --> 0:10:49.920 they're similar. Right. So we're in video cloud services provider partner. Okay, 0:10:50.000 --> 0:10:53.280 so we're restricted from offering consumer based GPUs in our cloud, 0:10:53.480 --> 0:10:56.000 got it, got it right. So we're actually we're required 0:10:56.040 --> 0:10:59.839 to buy datas energy pus. So, but from a technological standpoint, 0:11:00.600 --> 0:11:03.160 we're they're talking, we're talking about roughly the same family 0:11:03.200 --> 0:11:06.320 type of technology. Yeah, you know, yes, it's okay. Like 0:11:06.360 --> 0:11:09.640 the data center GPUs typically have a little bit higher quality, 0:11:09.880 --> 0:11:11.840 like they may have error correcting RAM instead of just 0:11:11.920 --> 0:11:15.600 regular video RAM on them, and then typically the frame buffers, 0:11:15.600 --> 0:11:17.720 so the actual amount of video memory on the GPU 0:11:17.720 --> 0:11:19.240 is going to be higher for the data center cards. 0:11:19.800 --> 0:11:23.440 Got it. But technologically what people are lining up for 0:11:23.559 --> 0:11:26.640 in stores outside of video games supply stores and what 0:11:26.720 --> 0:11:28.560 you have a deck of for do you I have thousand? 0:11:28.640 --> 0:11:32.480 Probably more? By now? It's a it's the same kind 0:11:32.520 --> 0:11:37.080 of technology. And it's the artists that would the animation 0:11:37.200 --> 0:11:40.000 artists who probably worked for big studios or television shows 0:11:40.040 --> 0:11:42.680 on Netflix and so forth. They all want access to 0:11:42.920 --> 0:11:46.760 roughly that same technological capability. Yeah, it went from having 0:11:46.840 --> 0:11:50.280 the access to that in their office now that they 0:11:50.320 --> 0:11:52.360 may not have i T. Staff that's sophisticated enough to 0:11:52.400 --> 0:11:54.360 set it up for them, or their office may no 0:11:54.440 --> 0:11:57.560 longer exist. Right, And we've seen this huge rush to 0:11:57.600 --> 0:12:00.040 the cloud, the wild things we get on the on 0:12:00.320 --> 0:12:02.719 with studios and they're like by the end of the call, 0:12:02.760 --> 0:12:05.559 they're like, Okay, when can we start to be honest 0:12:05.600 --> 0:12:07.400 with you. The problem that we have as a company 0:12:07.440 --> 0:12:10.319 is we're actually limited in our ability to onboard, right, 0:12:10.400 --> 0:12:13.000 is there's just so much demand for this cloud migration. Yeah, 0:12:13.280 --> 0:12:15.240 and then when I look at that from a capacity 0:12:15.240 --> 0:12:18.280 planning perspective, it's like you're looking at this environment that 0:12:18.320 --> 0:12:20.960 we have today where the world is in this one 0:12:21.000 --> 0:12:24.600 big shortage of everything possible. Doing the capacity planning and 0:12:24.600 --> 0:12:41.920 scaling for that is this is kind of hard. You 0:12:41.960 --> 0:12:44.560 said something, you said birth capacity, and again I'm looking 0:12:44.559 --> 0:12:46.960 at your website. It says the GPS you need when 0:12:47.000 --> 0:12:49.760 you need them. And so what I take from this 0:12:50.640 --> 0:12:53.960 is that perhaps there are some you know, consumer startups 0:12:54.000 --> 0:12:57.160 or whatever, and they might use AWS cloud and they 0:12:57.280 --> 0:13:01.080 use a sort of predictable on a day to day 0:13:01.120 --> 0:13:05.960 basis roughly share of cloud capacity. Well you're offering, though, 0:13:06.440 --> 0:13:08.880 I guess part of your niche. It sounds like, is 0:13:09.360 --> 0:13:13.640 here's someone who doesn't normally need computational power, but when 0:13:13.640 --> 0:13:17.440 they need crazy computational power, a burst of it. It's 0:13:17.440 --> 0:13:21.480 not regular, you have huge, you have huge scaling ability 0:13:21.520 --> 0:13:23.640 for them. Yeah, you know that. That's the crazy thing. 0:13:24.440 --> 0:13:26.720 When we hired a director of sales last year, I 0:13:26.760 --> 0:13:28.760 had known him for like ten years, and I was like, dude, 0:13:28.800 --> 0:13:31.880 like like every customer we talked to tells us they 0:13:31.920 --> 0:13:33.679 just cannot get the capacity they need at the big 0:13:33.679 --> 0:13:36.120 clouds and he's like, yeah, okay, like that's a bunch 0:13:36.120 --> 0:13:39.040 of b s whatever. Okay, you're being like hyperbolic. And 0:13:39.080 --> 0:13:41.319 then he joined us. In the first like fifty meetings, 0:13:41.320 --> 0:13:43.520 he did every one in the exact same thing. So 0:13:43.840 --> 0:13:47.079 I think that the industry really struggles with that access 0:13:47.120 --> 0:13:49.719 to scale. Yeah, right, And the reason for that is 0:13:49.760 --> 0:13:51.839 just I don't like a w S g C P 0:13:51.920 --> 0:13:55.600 ANDAGE or are building these. They're kind of hyper scale 0:13:55.640 --> 0:13:59.360 data center regions two handle every single use case under 0:13:59.400 --> 0:14:02.440 the sun, right, right, So they've got X number of 0:14:02.440 --> 0:14:05.599 types of CPU compute, they've got all different types of 0:14:05.640 --> 0:14:08.000 optimized compute, and you know, at the end of the day, 0:14:08.000 --> 0:14:11.160 there's only a certain portion that's available for the GPU stuff, right, 0:14:11.200 --> 0:14:13.640 And when that's our entire business, like, we can really 0:14:13.640 --> 0:14:16.400 approach it very differently. You know. What we want people 0:14:16.440 --> 0:14:17.839 to do is to come in and say, hey, I 0:14:17.880 --> 0:14:20.160 need to get two thousand GPUs. I need them for 0:14:20.200 --> 0:14:22.880 three hours. If they were to do that at AWUS 0:14:22.960 --> 0:14:24.960 like one, it may not even be possible for them 0:14:24.960 --> 0:14:27.080 if they don't have a good relationship. And then too, 0:14:27.200 --> 0:14:30.000 they probably have to do a lot of like manual 0:14:30.120 --> 0:14:34.280 engineering and reservation planning to get there, we have people 0:14:34.320 --> 0:14:36.880 come in all the time that they they'll burst the 0:14:37.080 --> 0:14:39.960 like to three four or five tho GPUs for like 0:14:40.000 --> 0:14:41.760 six hours and then we won't hear from again for 0:14:41.800 --> 0:14:45.760 three weeks. My commitment to them is that I want 0:14:45.800 --> 0:14:47.800 them to have that capacity without having to talk to me. 0:14:48.280 --> 0:14:50.440 So one thing that and I guess I kind of 0:14:50.480 --> 0:14:53.760 intuitively knew this before, but one thing that's making this 0:14:53.840 --> 0:14:57.400 clear is that cloud computing. You know, you think of 0:14:57.400 --> 0:14:59.560 a cloud like a literal clouds at the cloud up there, 0:14:59.560 --> 0:15:03.120 but cloud computing it's just not a commodity, both in 0:15:03.240 --> 0:15:08.160 terms of the specific types of applications, but also in 0:15:08.680 --> 0:15:12.400 UH scaling and the difference between sort of like standard 0:15:12.480 --> 0:15:17.920 run time consistent demand versus cloud that can be turned 0:15:17.920 --> 0:15:21.600 on at high scale for a client at a moment's notice. 0:15:22.320 --> 0:15:25.120 That's a really good point. I like to think that 0:15:25.640 --> 0:15:28.480 our scale helps us break the commodity label. Yeah, but 0:15:28.640 --> 0:15:30.720 for anybody who goes in then needs like one virtual 0:15:30.760 --> 0:15:33.560 machine with like four CPU corps and sixteen RAM like, 0:15:33.840 --> 0:15:36.560 all right, go get in the cheapest place, right, same same, 0:15:37.080 --> 0:15:38.760 But that's not even the people that we go after 0:15:38.840 --> 0:15:40.920 or talk to, right, I can't help them. I can't 0:15:40.920 --> 0:15:45.080 help them there, right, Why is RAM hard to come by? Okay, 0:15:45.160 --> 0:15:47.520 so let's go, let's go to crypto money, all right, 0:15:47.680 --> 0:15:49.360 I guess where I was. I was trying to like 0:15:49.440 --> 0:15:52.320 hold off the crypto part for as long as we could, 0:15:52.960 --> 0:15:54.680 but I knew eventually I would ask you a question 0:15:54.720 --> 0:15:56.800 and we'd stumble into the crypto parts. So let's I 0:15:56.840 --> 0:15:59.760 guess we're there now. Yeah, so that there's a lot 0:15:59.800 --> 0:16:02.520 of supply chain stuff here is going to be driven back, 0:16:02.520 --> 0:16:04.600 it's going to be kind of linked back to crypto. Well, actually, 0:16:04.640 --> 0:16:07.800 then in that case, let me so before then, you 0:16:07.840 --> 0:16:09.960 mentioned how you would love to go back like six 0:16:10.040 --> 0:16:12.920 months ago because your job was really easy. So when 0:16:12.960 --> 0:16:15.760 did you notice that things were starting to change and 0:16:15.800 --> 0:16:20.960 that sourcing compute, whether it's GPUs, CPUs or RAM, that 0:16:21.120 --> 0:16:23.880 was suddenly starting to get tighter. You know, every February 0:16:23.920 --> 0:16:27.360 it gets pretty rough because of Chinese New Year. Okay, Right, 0:16:27.560 --> 0:16:29.320 so you kind of rushed to get all your orders 0:16:29.400 --> 0:16:31.640 in before the end of because this is where it's 0:16:31.640 --> 0:16:34.800 all being manufactured. Yeah, before before the middle of January, right, 0:16:34.880 --> 0:16:36.080 and then if you get it in before the middle 0:16:36.080 --> 0:16:37.800 of January, they're gonna do everything they can to shift 0:16:37.800 --> 0:16:40.240 it out for the start of their holiday. If you 0:16:40.240 --> 0:16:42.440 sent it by air, you'll have it early February, sent 0:16:42.480 --> 0:16:45.320 it by boat end of February, but like if you wait, 0:16:45.800 --> 0:16:49.040 there's there's nothing that gets processed in February in China. Okay, 0:16:49.320 --> 0:16:52.480 So that was kind of like normal operating conditions in 0:16:52.600 --> 0:16:54.840 middle of January, and you know, Crypto was kind of 0:16:54.840 --> 0:16:57.080 doing his thing and it was kind of ripping the 0:16:57.160 --> 0:17:01.160 roof off and you saw it coming right, But a 0:17:01.200 --> 0:17:03.520 lot of that was, you know, that was really in 0:17:03.560 --> 0:17:06.520 the consumer side, and so consumer GPUs are really tight, 0:17:06.840 --> 0:17:12.200 didn't really impact us. And then starting in March, orders 0:17:12.240 --> 0:17:14.240 that we have placed in January that we're supposed to 0:17:14.240 --> 0:17:16.960 be in stock in the us UM and we're supposed 0:17:16.960 --> 0:17:19.680 to be delivered the next day, they didn't show up right. 0:17:19.800 --> 0:17:22.560 And this is like, let's call it medium quality server 0:17:22.600 --> 0:17:25.959 grade CPUs, so like a md epic rome platform, and 0:17:26.080 --> 0:17:28.320 we had maybe five hundred or seven fifty at least 0:17:28.320 --> 0:17:31.399 things on order, and it was like, okay, fine, like 0:17:31.440 --> 0:17:33.040 the other components that we need to do our bills 0:17:33.040 --> 0:17:35.119 aren't here yet, Like they arrived in early March, and 0:17:35.160 --> 0:17:37.640 we were actually waiting a GPUs at the time, and 0:17:37.800 --> 0:17:39.520 I kind of wasn't really pressing about it. And then 0:17:39.560 --> 0:17:41.959 the middle of March came and I'm like, where's my stuff? 0:17:42.600 --> 0:17:45.040 And you reach out and they go, oh, we didn't 0:17:45.040 --> 0:17:46.639 first see the semi conductor short. I'm like, what are 0:17:46.640 --> 0:17:48.200 you talking about? You told me these were in stock 0:17:48.240 --> 0:17:52.680 inventor warehouse yesterday like that. It was fine. And this 0:17:52.720 --> 0:17:54.679 is kind of what you start to see when the 0:17:54.760 --> 0:17:57.280 market gets stretched like this, is you have conversations like 0:17:57.320 --> 0:17:59.119 that and people just flat out lie to you in 0:17:59.160 --> 0:18:02.399 the supply chain. That was the first real telltale sign. 0:18:02.920 --> 0:18:05.359 The second one was when I went to buy I 0:18:05.400 --> 0:18:07.800 think it was like eight d sticks of sixty d 0:18:07.880 --> 0:18:10.720 d R four you know for that, like for for 0:18:10.800 --> 0:18:13.480 d d R four what d d R four it's 0:18:13.760 --> 0:18:17.600 System Ram okay, right, so it's the it's the most 0:18:17.600 --> 0:18:20.640 recent commodity system rahm um. There's new versions coming out, 0:18:20.680 --> 0:18:23.080 but this is really what's used today. You reached around 0:18:23.119 --> 0:18:24.719 for System Ram because a lot of that stuff comes 0:18:24.760 --> 0:18:27.159 out of liquidation, right, so you can get it for 0:18:27.200 --> 0:18:29.800 this particular component for Ram, you typically buy it used. 0:18:30.480 --> 0:18:33.880 You're not going to go out and spend sixty premium 0:18:33.880 --> 0:18:36.400 on this stuff. Just because you know, that doesn't really 0:18:36.400 --> 0:18:39.639 wear down. So I reached out to to one of 0:18:39.640 --> 0:18:41.639 our main suppliers and I said, hey, man, like I 0:18:41.680 --> 0:18:43.560 need eight hundred sticks and he's like, dude, I got nothing, 0:18:43.960 --> 0:18:45.240 Like what are you talking about? Like what do you 0:18:45.280 --> 0:18:48.440 mean you give nothing? He's like, China? But when was that? What? 0:18:48.440 --> 0:18:50.920 What month? This was in March? Middle of March? Yeah, 0:18:51.560 --> 0:18:53.800 and he goes he goes, dude, like I had some 0:18:53.840 --> 0:18:56.040 guy in China buy all of it last night, Like 0:18:56.080 --> 0:18:57.360 what do you mean all of it? And he's like, yeah, 0:18:57.440 --> 0:19:01.200 he bought like ten thousand sticks. Stupid question and stupid question. 0:19:01.240 --> 0:19:04.880 I don't know if this is you know, proprietary sticks, 0:19:04.920 --> 0:19:06.800 Like what are we talking about from a dollar amount 0:19:08.160 --> 0:19:10.240 two and a half maybe I think it's maybe two 0:19:10.280 --> 0:19:13.640 and a half million bucks? Okay, okay, right, but ten 0:19:13.640 --> 0:19:16.920 thousand six a ram is not really a small order, 0:19:17.080 --> 0:19:18.399 right you don't just like buy them and throw it 0:19:18.440 --> 0:19:20.720 in the closet for rainy day. Well I guess maybe 0:19:20.720 --> 0:19:22.880 this year you do. And I said, okay, like what's 0:19:22.920 --> 0:19:24.960 going on and then said that's he was like, you know, 0:19:25.040 --> 0:19:26.720 some one buyer came in from China and bought it 0:19:26.800 --> 0:19:29.680 all and and you know, I kind of give him 0:19:29.680 --> 0:19:32.040 the the expletive late and you gotta be kidding me. 0:19:32.320 --> 0:19:33.679 A couple of days later he called me and he 0:19:33.720 --> 0:19:35.359 was like, hey, I got like four hundred sticks, Like 0:19:35.359 --> 0:19:36.719 do you want him? And I said, yeah, I'll take 0:19:36.880 --> 0:19:39.199 everything you have. So that that was kind of my 0:19:39.280 --> 0:19:41.919 reaction to that first one. Then a week later on 0:19:41.960 --> 0:19:43.280 the fact of the same guy and said, hey, I've 0:19:43.280 --> 0:19:45.359 got to I've got a need for like Intel's on 0:19:45.440 --> 0:19:49.240 V four, which is chips that probably stopped production like ten, 0:19:50.000 --> 0:19:52.159 Like what do you got? And he's like, dude, like, 0:19:52.200 --> 0:19:53.600 are you gonna kill me if I tell you that 0:19:53.600 --> 0:19:57.240 I sold them all to China last night? Like okay. 0:19:57.320 --> 0:19:59.560 So this is at the point when when final coin 0:19:59.640 --> 0:20:04.040 was going from like bucks. So just for so just 0:20:04.080 --> 0:20:06.399 for listeners that don't know, and why do you explain 0:20:06.440 --> 0:20:09.040 it further? But file coin is a new cryptocurrency is 0:20:09.080 --> 0:20:11.880 actually I think first conceived of back in like seventeen. 0:20:11.920 --> 0:20:17.760 It only recently launched, but the idea is essentially cloud files, 0:20:17.800 --> 0:20:22.000 distributed file storage in the cloud. Rather than hosting your 0:20:22.200 --> 0:20:25.400 documents with a company like dropbox or box, you hosted 0:20:25.400 --> 0:20:28.840 on a network of a decentralized blockchain network of computers 0:20:28.880 --> 0:20:31.600 all around the world. Yeah, and all around the world. 0:20:31.640 --> 0:20:34.480