WEBVTT - OpenAI Is A Systemic Risk To The Tech Industry 0:00:02.800 --> 0:00:20.239 Zone Media. Try spinning. That's a good trick. I'm at Zetron. 0:00:20.320 --> 0:00:22.600 This is Better Offline and this is the second episode 0:00:22.600 --> 0:00:25.119 of my two part series where I explain how open 0:00:25.120 --> 0:00:27.600 ai has become a systemic risk to the tech industry, 0:00:27.880 --> 0:00:30.440 even with its massive forty billion dollar funding round and 0:00:30.520 --> 0:00:33.239 bird brain benefactor in the form of soft Bank, the 0:00:33.240 --> 0:00:36.959 world's foremost authority in losing money. Now, before I continue, 0:00:37.040 --> 0:00:39.920 shameless request, Better Offline has been nominated for a Webbie 0:00:39.920 --> 0:00:41.560 and I want to win this thing. I've linked to 0:00:41.640 --> 0:00:43.360 it in my Twitter and my Blue Sky and if 0:00:43.360 --> 0:00:44.519 you could vote for me, well, it will be in 0:00:44.560 --> 0:00:48.879 the episode notes two. Moving on back at it. Okay, 0:00:49.360 --> 0:00:53.559 all right, open ai now has forty billion dollars somehow, 0:00:53.920 --> 0:00:57.120 right great? Right, Well, Dolgier horses. As part of its 0:00:57.160 --> 0:01:00.000 deal with soft Bank, open Ai must also convert its 0:01:00.040 --> 0:01:03.000 bizarre nonprofit structure into a for profit entity by December 0:01:03.000 --> 0:01:05.839 twenty twenty five, or it will lose ten billion dollars 0:01:05.840 --> 0:01:08.600 from that forty billion dollars and of funding. And just 0:01:08.640 --> 0:01:10.600 to be clear, by the way, they've only really got 0:01:10.640 --> 0:01:12.800 ten billion dollars of that so far. The rest is 0:01:12.840 --> 0:01:15.600 at the end of the year. Furthermore, and the event 0:01:15.640 --> 0:01:17.919 that open ai fails to convert into a for profit 0:01:18.000 --> 0:01:21.160 company by October twenty twenty six, investors in its previous 0:01:21.200 --> 0:01:23.560 six point six billion dollar funding route can claw back 0:01:23.600 --> 0:01:25.840 their investment with a converting into a loan with an 0:01:25.840 --> 0:01:29.440 attached interest rate. Naturally, this represents a night mess scenario 0:01:29.480 --> 0:01:32.320 for the company, as it will increase both its costs 0:01:32.360 --> 0:01:35.920 and its outgoings. This is a complex situation that almost 0:01:35.959 --> 0:01:38.640 warrants its own podcast, But the long and short of 0:01:38.680 --> 0:01:41.520 it is that open ai would have to effectively dissolve itself, 0:01:41.720 --> 0:01:44.880 start the process of reforming an entirely new entity, and 0:01:44.959 --> 0:01:48.240 distribute its assets to other nonprofits or seller license them 0:01:48.240 --> 0:01:50.760 to a for profit company at fair market rates which 0:01:50.800 --> 0:01:54.280 they would not set. It would also require valuing open 0:01:54.280 --> 0:01:56.360 ai as assets, which in and of itself would be 0:01:56.400 --> 0:01:58.560 a difficult task, as well as getting past the necessary 0:01:58.640 --> 0:02:02.400 state regulators, the IRA state revenue agencies, and the upcoming 0:02:02.400 --> 0:02:05.800 trial with Elon Musk Well, that only adds further problems. 0:02:06.400 --> 0:02:09.000 I've simplified things here, and that's because, as I've said, 0:02:09.040 --> 0:02:12.440 this stuff is a little complex and pretty boring. Suffice 0:02:12.480 --> 0:02:14.680 to say, this isn't as simple as liquidating a company 0:02:14.680 --> 0:02:17.399 and starting a fresh or submitting a couple of legal filings. 0:02:17.480 --> 0:02:19.400 It's a long, fraught process and one that will be 0:02:19.480 --> 0:02:22.799 as as has been subject to legal challenges, both from 0:02:22.800 --> 0:02:25.560 open AI's business rivals as well as from civil society 0:02:25.680 --> 0:02:28.519 organizations in California. You may have heard the lost monologue. 0:02:29.200 --> 0:02:31.240 Based on discussions with experts in the field of my 0:02:31.280 --> 0:02:33.480 own research, I simply do not know how open ai 0:02:33.639 --> 0:02:37.320 pulls off this by October twenty twenty six, and honestly, 0:02:37.360 --> 0:02:38.880 I'm not sure how they do it by the end 0:02:38.880 --> 0:02:43.200 of this year. It's insane. It's a it's a really 0:02:44.480 --> 0:02:47.400 I just every time I read this stuff and I 0:02:47.440 --> 0:02:49.840 write out, I'm like, how is nobody else reading this 0:02:49.960 --> 0:02:52.040 and going, what the fuck is going on? You See, 0:02:52.040 --> 0:02:55.640 this is a big problem, this nonprofit thing, because open 0:02:55.680 --> 0:02:58.480 eye really has become a systemic risk to the tech industry, 0:02:58.480 --> 0:03:01.519 and anything that increases that is bad news for everybody. 0:03:02.080 --> 0:03:05.519 Open Ai, they've become a kind of load bearing company 0:03:05.520 --> 0:03:07.880 for this industry, both as a narrative, as I've discussed 0:03:07.919 --> 0:03:10.680 multiple times as chat GPT is the only large language 0:03:10.800 --> 0:03:13.560 model company with any meaningful use base, and also as 0:03:13.600 --> 0:03:16.920 a financial entity. Its ability to meet its obligations and 0:03:16.960 --> 0:03:19.800 its future expansion plans are critical to the future health 0:03:19.880 --> 0:03:22.800 or in some cases survival of multiple large companies. And 0:03:22.800 --> 0:03:25.320 that's before the after effects that will affect its customers 0:03:25.320 --> 0:03:28.320 as a result of any kind of financial collapse. The 0:03:28.360 --> 0:03:30.560 parallels to the two thousand and seven the two thousand 0:03:30.560 --> 0:03:34.080 and eight financial crisis are starting to become a little worrying. 0:03:34.320 --> 0:03:37.000 Layman Brothers wasn't the largest investment bank in the world, 0:03:37.120 --> 0:03:39.720 although it was pretty big, just like open Ai isn't 0:03:39.760 --> 0:03:42.600 the largest tech company, though again it's certainly large in 0:03:42.680 --> 0:03:47.360 terms of alleged valuation and expenditures. Layman Brothers collapse sparked 0:03:47.360 --> 0:03:50.080 a contagient that would later spread throughout the entire global 0:03:50.120 --> 0:03:54.120 financial services industry and consequently the global economy. Now I 0:03:54.160 --> 0:03:57.680 can see open AI's failure not having as big an effect, 0:03:57.760 --> 0:04:02.400 but I can imagine a systemic effect. Still, you have 0:04:02.480 --> 0:04:06.280 to realize that the whole AI trade, the narrative, the bubble, 0:04:06.320 --> 0:04:09.240 it's holding up the economy. I think like thirty thirty 0:04:09.240 --> 0:04:11.680 five percent of the US stock market is in the 0:04:11.680 --> 0:04:14.760 Magnificent seven, and all of their bullshit numbers right now 0:04:14.800 --> 0:04:18.240 are held up by this nonsense. And like the financial crisis, 0:04:18.720 --> 0:04:20.640 the impact in this case won't be limited to just 0:04:20.680 --> 0:04:24.440 bankers and insurers. It will bleed into everything else. This 0:04:24.520 --> 0:04:26.680 episode is going to be a bit grim. I'm not 0:04:26.720 --> 0:04:28.680 going to lie. I want to lay out the direct 0:04:28.720 --> 0:04:31.280 result of any kind of financial crisis at open Ai, 0:04:31.320 --> 0:04:34.640 because I don't think anybody is taking this seriously. Let's 0:04:34.680 --> 0:04:36.920 start with Oracle, who will lose at least a billion 0:04:36.960 --> 0:04:42.080 dollars if open ai doesn't fulfill its obligations per the Information. Oracle, 0:04:42.120 --> 0:04:44.640 which has taken responsibility for organizing the construction of the 0:04:44.640 --> 0:04:47.960 stargate data centers with unproven data center build a Cruso 0:04:48.240 --> 0:04:50.240 and I quote the Information here, may need to raise 0:04:50.240 --> 0:04:53.240 more capital to fund its data center ambitions. Oracle has 0:04:53.279 --> 0:04:55.640 signed a fifteen year lease with Crusoe, and to quote 0:04:55.640 --> 0:04:57.920 the Information is on the hook for one billion dollars 0:04:57.960 --> 0:05:01.200 in payments to that firm. To further quote information. While 0:05:01.240 --> 0:05:03.440 that's the standard deal length, the unprecedented size of the 0:05:03.440 --> 0:05:06.280 facility Oracle is building for just one customerment makes its 0:05:06.400 --> 0:05:08.800 riskier than the standard cloud data center used by lots 0:05:08.800 --> 0:05:12.200 of interchangeable customers with much more predictable needs. According to 0:05:12.200 --> 0:05:14.400 half a dozen people familiar with these types of deals. 0:05:14.960 --> 0:05:17.599 In simpler terms, Oracle is building a giant data center 0:05:17.600 --> 0:05:20.599 for one customer, open Ai, and has taken on the 0:05:20.600 --> 0:05:23.800 financial burden associated with them. If open Ai fails to 0:05:23.839 --> 0:05:25.880 expand or lacks the capital to actually pay for its 0:05:25.920 --> 0:05:28.480 share of the Stargate data center project, Oracle is on 0:05:28.520 --> 0:05:30.400 the hook for at least a billion dollars, and based 0:05:30.440 --> 0:05:32.400 on the Information's reporting, it's also on the hooked by 0:05:32.440 --> 0:05:35.320 the GPUs for the site. This is me quoting them again. 0:05:36.960 --> 0:05:39.480 Even before the Stargate announcement, Oracle and open Ai had 0:05:39.480 --> 0:05:41.599 agreed to expand their Abelene deal from two to eight 0:05:41.680 --> 0:05:43.920 data center buildings, which can hold four hundred thousand and 0:05:44.000 --> 0:05:46.760 VIDIA Blackwell GPUs, adding tens of billions of dollars to 0:05:46.800 --> 0:05:50.200 the cost of the facility. In reality, this development will 0:05:50.240 --> 0:05:53.360 likely cost tens of billions of dollars, nineteen billion dollars 0:05:53.360 --> 0:05:55.160 of which is due from open Ai, which does not 0:05:55.279 --> 0:05:57.320 have the money until it receives its second tranche of 0:05:57.320 --> 0:06:00.560 funding in December twenty twenty five from SoftBank, and this 0:06:00.760 --> 0:06:02.919 is contingent partially on their ability to convert into a 0:06:02.920 --> 0:06:06.039 for profit entity, which has mentioned is extremely difficult and 0:06:06.120 --> 0:06:10.039 extremely unlikely. It's unclear how many of the Blackwell GPUs 0:06:10.040 --> 0:06:12.320 the Oracle has had to purchase in advance, but in 0:06:12.360 --> 0:06:15.200 the event of any kind of financial collapse, open Ai 0:06:15.279 --> 0:06:18.360 Oracle will likely have to toss at least a billion dollars, 0:06:18.360 --> 0:06:21.279 if not several billion dollars, and then we get the 0:06:21.360 --> 0:06:25.080 core with a companies. So a company whose expansion is 0:06:25.160 --> 0:06:28.239 likely driven entirely by open Ai now and cannot survive 0:06:28.279 --> 0:06:31.240 without open Ai flilling its obligations if it doesn't die. 0:06:31.240 --> 0:06:34.120 O Anyway, Now, I've written and spoken a lot about 0:06:34.279 --> 0:06:36.880 publicly traded AI compute firm Core Weave, and it would 0:06:36.880 --> 0:06:38.560 give me the greatest pleasure in my life. Never think 0:06:38.680 --> 0:06:41.560 or talk about them ever again. Nevertheless, I have to. 0:06:42.080 --> 0:06:45.559 This is my curse. This is my curse. Core Weave 0:06:45.560 --> 0:06:48.000 has become my curse every time I think about this. 0:06:48.080 --> 0:06:52.320 Fuck Okay. The Financial Times revealed a few weeks ago 0:06:52.360 --> 0:06:54.360 that core Weave's debt payments could balloon to over two 0:06:54.400 --> 0:06:56.120 point four billion dollars a year by the end of 0:06:56.120 --> 0:06:59.120 twenty twenty five, far outstripping its cash reserves, and the 0:06:59.160 --> 0:07:01.799 Information reported that its cash burn would increase to fifteen 0:07:01.839 --> 0:07:05.560 billion dollars in twenty twenty five, as Bird's IBO filing. 0:07:05.600 --> 0:07:08.080 Sixty two percent of core weaves twenty twenty four revenue 0:07:08.120 --> 0:07:10.400 a little under two billion, with losses amounting to eight 0:07:10.520 --> 0:07:13.440 hundred and sixty three million was Microsoft Compute, and based 0:07:13.480 --> 0:07:15.880 on the conversations I've had with sources, a good amount 0:07:15.920 --> 0:07:19.200 of this was Microsoft running compute for open Ai. Starting 0:07:19.200 --> 0:07:21.880 October twenty twenty five, open Ai will start paying core 0:07:21.920 --> 0:07:24.280 Weave as part of its five year long, twelve billion 0:07:24.320 --> 0:07:27.840 dollar contract, picking up the option that Microsoft declined. This 0:07:28.360 --> 0:07:30.920 is not great timing, or maybe it's perfect timing, because 0:07:30.960 --> 0:07:33.160 this is also when core Weave will have to start 0:07:33.160 --> 0:07:36.200 making payments on their massive, stupid, multi billion dollar d 0:07:36.280 --> 0:07:40.360 DTL two point zero loan mentioned in previous episodes. But really, 0:07:40.360 --> 0:07:42.760 there's a newsletter if you want to you hear me, 0:07:42.800 --> 0:07:44.320 go mad, You want to read me go mad? You 0:07:44.360 --> 0:07:46.480 read my read my core with piece because it really 0:07:46.560 --> 0:07:51.679 drove me insane. Nevertheless, these core Weave payments, the ones 0:07:52.000 --> 0:07:54.600 from open Ai to core Weave that October, they're pretty 0:07:54.680 --> 0:07:58.320 much critical to Corewave's future. This deal also suggests that 0:07:58.400 --> 0:08:01.840 open ai will become core Weave's life just customer. Microsoft 0:08:01.840 --> 0:08:04.760 had previously committed to spending ten billion dollars on core 0:08:04.800 --> 0:08:06.920 Weave services by the end of the decade, but CEO 0:08:06.960 --> 0:08:09.400 satchly Adella added a few months later on a podcast 0:08:09.600 --> 0:08:13.440 that its relationship with core Weave was a one time thing. Man, 0:08:14.160 --> 0:08:16.720 man a really like really fucking around there, such a 0:08:16.760 --> 0:08:19.600 don't love core Weave. Assuming Microsoft keeps spending it its 0:08:19.640 --> 0:08:23.720 previous rate, So about one point like sixty six percent 0:08:23.880 --> 0:08:26.840 of two billion dollars whatever those I have something that 0:08:26.880 --> 0:08:28.880 isn't guaranteed. By the way, it would still only be 0:08:29.000 --> 0:08:32.000 half of open AI's potential revenue to core Weave. Core 0:08:32.040 --> 0:08:34.960 Weave's expansion at this point is entirely driven by open Ai. 0:08:35.440 --> 0:08:38.000 Seventy seven percent of its twenty twenty four revenue came 0:08:38.000 --> 0:08:40.360 from two customers, Microsoft being the largest and yes, I 0:08:40.480 --> 0:08:42.800 just fucked up a number at sixty two percent and 0:08:42.880 --> 0:08:46.160 using core weaves auxiliary compute for open Ai. As a result, 0:08:46.200 --> 0:08:49.520 the future expansion efforts the theoretical one point three gigawatts 0:08:49.559 --> 0:08:52.120 have contracted, and by the way, that means it doesn't exist. 0:08:52.120 --> 0:08:54.760 Compute at core Weaver are largely, if not entirely, for 0:08:54.800 --> 0:08:57.160 the benefit of open Ai. In the event that open 0:08:57.200 --> 0:09:00.360 Ai cannot fulfill its obligations, core weave will collapse. It's 0:09:00.360 --> 0:09:03.600 that fucking simple, and then the shock waves will ripple further. 0:09:04.120 --> 0:09:06.560 In Video relies on corewey for more than six percent 0:09:06.600 --> 0:09:09.800 of its revenue and corwy's future credit worthiness to continue 0:09:09.800 --> 0:09:13.440 receiving said revenue. Well, much of that is dependent on 0:09:13.480 --> 0:09:16.640 open ai continuing to buy services from core Weave now 0:09:16.640 --> 0:09:18.319 and basically this in a comment I received from the 0:09:18.400 --> 0:09:21.640 legendary Gil Luria, Managing director and head of Technology research 0:09:21.679 --> 0:09:24.760 at Analyst DA Davison and Co, I quote him, since 0:09:24.880 --> 0:09:27.599 cor We've bought two hundred thousand GPUs last year, and 0:09:27.640 --> 0:09:30.360 those systems are around forty thousand dollars. We believe cor 0:09:30.360 --> 0:09:32.880 We've spent eight billion dollars on in Vidia last year. 0:09:33.040 --> 0:09:36.400 That represents more than six percent of Nvidia's revenue in 0:09:36.440 --> 0:09:38.680 twenty twenty four, he said last year, but I've just 0:09:38.679 --> 0:09:41.760 wanted to make it sound better. Core We've receives preferential 0:09:41.760 --> 0:09:45.320 access to Nvidia's GPUs, though Nvidia kind of denies that 0:09:45.400 --> 0:09:48.600 and makes up billions of dollars of Nvidia's revenue. Corweave 0:09:48.600 --> 0:09:51.600 then takes those GPUs and then they raise debt using 0:09:51.640 --> 0:09:56.000 the GPU's as collateral as well as customer contracts. Then 0:09:56.320 --> 0:09:58.920 they use the money they've raised to buy more GPUs 0:09:58.920 --> 0:10:01.640 from Nvidia. You may think that doesn't sound right. I 0:10:01.679 --> 0:10:04.440 am being complete, like this is factual information. At this 0:10:04.520 --> 0:10:07.480 point in video was the anchor for Corwave's IPO, and 0:10:07.559 --> 0:10:10.160 CEO Michael and Trader said that the IPO would not 0:10:10.320 --> 0:10:13.240 have closed without Invidia buying two hundred and fifty million 0:10:13.280 --> 0:10:16.680 dollars worth of their shares. Nvidia also invested one hundred 0:10:16.679 --> 0:10:18.640 million dollars in the early days of core weaves, and 0:10:18.679 --> 0:10:21.839 for reasons I cannot understand, also agreed to spend one 0:10:21.840 --> 0:10:24.240 point three billion dollars over four years two and I 0:10:24.320 --> 0:10:27.360 quote the information rent its own chips from core weave 0:10:28.040 --> 0:10:30.880 fum fact. I can't find a single fucking mention of 0:10:30.960 --> 0:10:34.480 core weave in any of Nvidia's filings. Now buried in 0:10:34.559 --> 0:10:37.320 that core weaves s one the document every company publishes 0:10:37.360 --> 0:10:40.440 before going public was a warning about counterparty credit risk, 0:10:40.559 --> 0:10:42.800 which is when one party provides services or goods to 0:10:42.840 --> 0:10:46.320 another with specific repayment terms and the other party doesn't 0:10:46.320 --> 0:10:49.160 meet their side of the deal. While this was written 0:10:49.160 --> 0:10:52.600 as a theoretical as it could, in theoretically speaking, come 0:10:52.679 --> 0:10:55.040 from any company to which core Weave acts as a creditor. 0:10:55.480 --> 0:11:00.280 It only named one open Ai now has discussed previously. 0:11:00.360 --> 0:11:03.760 Core Weaver is saying that should a customer, any customer, 0:11:03.800 --> 0:11:06.320 but really they mean open Ai failed to pay its 0:11:06.360 --> 0:11:09.280 bills for infrastructure built on their behalf or services rendered, 0:11:09.559 --> 0:11:11.880 it can have a material risk of the company. Now. 0:11:12.000 --> 0:11:14.520 As an aside, the information reported that Google and someone's 0:11:14.520 --> 0:11:15.839 going to email me there, so I just want to 0:11:15.880 --> 0:11:18.000 get ahead of it. The core Weave is apparently in 0:11:18.120 --> 0:11:21.400 advanced talks with Google to nd GPUs it Also, it 0:11:21.480 --> 0:11:23.439 also added another thing in this story, just so that 0:11:23.520 --> 0:11:25.280 I don't have to hear from any of you. The 0:11:25.360 --> 0:11:29.400 Google's potential deal with core Weaves significantly smaller than their 0:11:29.400 --> 0:11:31.800 commitments with Microsoft as according to one of the people 0:11:31.880 --> 0:11:34.440 briefed on it, but could potentially expand in future years. 0:11:34.760 --> 0:11:36.720 Do not come to me and claim that Google's going 0:11:36.760 --> 0:11:39.800 to save core Weeve. I'll be so mad anyway. Even 0:11:39.840 --> 0:11:42.679 with Google and open AI's money, Carew's continued the ability 0:11:42.720 --> 0:11:44.959 to do business hinges heavily on its ability to raise 0:11:45.000 --> 0:11:47.599 further debt, which I have previously called into question a 0:11:47.679 --> 0:11:50.320 newsletter that gave me madness, and its ability to raise 0:11:50.640 --> 0:11:54.160 future debt is to quote the financial times secured against 0:11:54.160 --> 0:11:56.880 it's more than two hundred and fifty thousand n vida 0:11:56.960 --> 0:12:02.319 GPUs and its contracts with customers such as Microsoft. Now, 0:12:02.720 --> 0:12:05.280 any future debt that core Weave raises will be based 0:12:05.280 --> 0:12:07.079 off of its contract with open Ai. You know, the 0:12:07.160 --> 0:12:10.800 counterparty credit risk threat that represents a disproportionate sharef it's 0:12:10.840 --> 0:12:13.840 revenue I just mentioned, and also whatever GPUs they still 0:12:13.920 --> 0:12:16.319 have left that they can get debt on. As a result, 0:12:16.360 --> 0:12:18.240 a chunk of a video's future revenue is dependent on 0:12:18.320 --> 0:12:20.640 open AI's ability to fulfill its obligations the core Weave 0:12:20.679 --> 0:12:22.320 both in its ability to pay them, and they're timing 0:12:22.360 --> 0:12:24.760 less in doing so. If open ai fails, then core 0:12:24.840 --> 0:12:27.719 Weave fails, then that hurts and video Jensen's going to 0:12:27.760 --> 0:12:29.360 have to go. He's going to have to go to 0:12:29.559 --> 0:12:34.680 a cheaper leather jacketarium and it gets worse. Open AI's 0:12:34.720 --> 0:12:37.520 expansion is dependent on two unproven startups, one of them 0:12:37.559 --> 0:12:40.920 I just mentioned, who are also dependent on open ai 0:12:41.040 --> 0:12:44.040 to live with Microsoft's data center pullback and open AI's 0:12:44.040 --> 0:12:46.920 intend to become independent from redmen. Future data center expansion 0:12:47.000 --> 0:12:49.400 is based on two partners supporting coll Weave I know 0:12:50.520 --> 0:12:53.560 will get there and Oracle. Now I'm referring, of course 0:12:53.600 --> 0:12:56.920 to Core Scientific, which is the data center developer for 0:12:57.400 --> 0:13:00.280 core Weave, and of course Crusos, the data center develer 0:13:00.400 --> 0:13:03.040 four Oracle. Now, if you were wondering, I can't hint 0:13:03.120 --> 0:13:06.280 it about this earlier how many data center how many 0:13:06.400 --> 0:13:09.599 DIIDENTA centers do you think Cruso's ever built, and the 0:13:09.679 --> 0:13:12.559 answer is none, And of course Scientific, how many do 0:13:12.559 --> 0:13:14.520 you think they've built and the answer is also none. 0:13:14.880 --> 0:13:19.280 These are the fucking companies underpending the AI boom. I 0:13:19.360 --> 0:13:21.920 also really must explain how difficult it is to build 0:13:21.920 --> 0:13:24.319 a data center, and how said difficult he increases when 0:13:24.320 --> 0:13:27.120 you're building an AI focused one. For example, in Video 0:13:27.160 --> 0:13:29.240 had to delay the launch of its Blackwell GPUs because 0:13:29.240 --> 0:13:32.079 of how finnicky the associated infrastructure, so the servers and 0:13:32.120 --> 0:13:35.400 the cooling and such is for customers. This was for 0:13:35.480 --> 0:13:39.000 customers that had already been using GPUs and therefore likely 0:13:39.120 --> 0:13:41.920 knew how to manage the temperatures created by them. Also 0:13:42.080 --> 0:13:44.400 is another reminder open eyes on the hook for nineteen 0:13:44.400 --> 0:13:47.520 billion dollars of funding behind Stargate, and neither of them 0:13:47.640 --> 0:13:49.760 have that money. I just want to remind you of that, 0:13:49.880 --> 0:13:52.320 because it costs so much money to build a fucking 0:13:52.440 --> 0:13:55.760 data center. And imagine if you didn't have any experience 0:13:55.920 --> 0:13:58.079 and effectively had to learn from scratch, how do you 0:13:58.320 --> 0:14:03.640 think it would be building these data centers? Let's find out. 0:14:04.120 --> 0:14:06.959 So let's start in Abilene, Texas with Crusoe and the 0:14:07.040 --> 0:14:10.360 Stargate Data Center project. Now. Cruso is a former cryptocurrency 0:14:10.400 --> 0:14:12.439 mining company that has now raised hundreds of millions of 0:14:12.520 --> 0:14:15.400 dollars to build data centers for AI companies, starting with 0:14:15.440 --> 0:14:17.920 a three point four billion dollar data center financing deal 0:14:17.960 --> 0:14:20.960 with asset manager Blue Owl Capital. This yet to be 0:14:21.120 --> 0:14:23.640 completed data center has now been leased by Oracle, which 0:14:23.680 --> 0:14:26.160 will allegedly fill it full of GPUs for open AI. 0:14:26.720 --> 0:14:29.640 Despite calling itself and I quote the industry's first vertically 0:14:29.720 --> 0:14:33.360 integrated AI infrastructure provider, with the company using flared gas 0:14:33.400 --> 0:14:35.560 as a waste by product of oil production to power 0:14:35.640 --> 0:14:38.240 I infrastructure, Cruso does not appear to have built a 0:14:38.320 --> 0:14:40.520 single AI data center and is now being tasked with 0:14:40.600 --> 0:14:45.360 building one point two Gigawatt's had a data center capacity 0:14:45.440 --> 0:14:49.120 for open Ai. It's just so fucking Cruso is the 0:14:49.240 --> 0:14:52.080 sole developer and operator of the Abilene site, meaning, according 0:14:52.080 --> 0:14:54.360 to the information that it is in charge of contracting 0:14:54.440 --> 0:14:57.560 with construction contractors and data center customers as well as 0:14:57.640 --> 0:15:01.280 running the data center after it is built. Oracle, it seems, 0:15:01.280 --> 0:15:03.720 will be responsible for filling said data center with GPUs 0:15:03.720 --> 0:15:08.080 as mentioned. Nevertheless, the project also appears to be behind schedule. 0:15:08.440 --> 0:15:11.240 The Information reported in October twenty twenty four that Abilene 0:15:11.280 --> 0:15:13.720 was meant to have fifty thousand of Nvidia's Blackwell AI 0:15:13.920 --> 0:15:15.960 chips in the first quarter of twenty twenty five, and 0:15:16.000 --> 0:15:18.320 also suggested that the site was projected to have a 0:15:18.440 --> 0:15:20.400 whopping one hundred thousand of them by the end of 0:15:20.440 --> 0:15:23.360 twenty twenty five. Now you can join me back here 0:15:23.440 --> 0:15:26.000 in reality, because a report from Bloomberg in March twenty 0:15:26.080 --> 0:15:29.040 twenty five said the Open AI and Oracle we're expected 0:15:29.040 --> 0:15:32.080 to have sixteen thousand available by the summer of twenty 0:15:32.120 --> 0:15:35.000 twenty five, with and a quote open Ai and Oracle 0:15:35.080 --> 0:15:38.000 expecting to deploy sixty four thousand then video GB two 0:15:38.160 --> 0:15:41.200 hundreds at the Stargate Data center by the end of 0:15:41.280 --> 0:15:48.240 twenty twenty six. That's that's very delayed. That's really delayed. Again. 0:15:48.440 --> 0:15:50.520 How I run a PR firm in that I record 0:15:50.520 --> 0:15:53.040 a podcast, I've write a newsletter, I have a book. 0:15:53.040 --> 0:15:54.760 I'm right in. I got all this shit on and 0:15:54.800 --> 0:15:58.160 I'm the asshole who notices this anyway, has discussed previously. 0:15:58.240 --> 0:16:01.680 Open ai needs this capacity very bad. According to the information, 0:16:02.000 --> 0:16:04.640 open ai expect stargate to handle three quarters of its 0:16:04.680 --> 0:16:07.120 compute by twenty thirty and these delays call into question, 0:16:07.160 --> 0:16:09.320 at the very least whether this schedule is reasonable, or 0:16:09.400 --> 0:16:13.680 logical or even possible. And I actually really question whether 0:16:13.800 --> 0:16:16.800 Stargate itself is possible at this point. But it can 0:16:16.880 --> 0:16:19.480 get dumber because we're about to talk about Core Scientific, 0:16:19.880 --> 0:16:22.960 and they are core Weave's friends. They're the people building 0:16:23.040 --> 0:16:36.280 data centers for core Wave in Denton, Texas. Now, as 0:16:36.360 --> 0:16:38.520 you can probably tell, I've written a great deal about 0:16:38.560 --> 0:16:40.200 core Wave in the past. That got a monologue, got 0:16:40.240 --> 0:16:42.960 a newsletter, and I got a therapy bill for it. 0:16:43.080 --> 0:16:46.560 And specifically I've written about their build out partner, Core Scientific, 0:16:47.000 --> 0:16:51.040 a cryptocurrency mining company, yes, another one that has exactly 0:16:51.120 --> 0:16:53.360 one customer for its AI data centers, and you'll never 0:16:53.440 --> 0:16:56.160 guess who it is. It's Core Wave. Now here's a 0:16:56.200 --> 0:16:59.800 few fun facts about Core Scientific. Core Scientific was bankrupt 0:16:59.840 --> 0:17:02.720 lotat year. Course Scientific has never built an AI data center, 0:17:02.760 --> 0:17:05.640 and its cryptocurrency mining operations were built around a six 0:17:06.040 --> 0:17:09.400 specialist computers for mining bitcoin, which led to an analyst 0:17:09.480 --> 0:17:12.520 to tell CNBC that said data centers would and I 0:17:12.600 --> 0:17:14.440 quote need to be buildozed and built from the ground 0:17:14.520 --> 0:17:18.520 up to accommodate AI compute. That's the stuff. Course Scientific 0:17:18.560 --> 0:17:21.080 also does not appear to have any meaningful AI compute 0:17:21.080 --> 0:17:24.000 of any kind. It's AI slash HPC, which is high 0:17:24.040 --> 0:17:28.639 performance computing. Revenue represents a teeny tiny, teeny little percentage 0:17:28.680 --> 0:17:31.840 of overall revenue, which mostly comes from mining crypto, both 0:17:31.840 --> 0:17:35.879 for itself and other parties. Now, hearing all of this, 0:17:36.920 --> 0:17:41.639 would you give this company your your compute? Would you 0:17:41.720 --> 0:17:43.960 think these are the people that I am going to 0:17:44.080 --> 0:17:47.680 call to build my data centers. If you said no, 0:17:48.080 --> 0:17:50.520 you are smarter than Corewave, who has given their entire 0:17:50.560 --> 0:17:55.160 one point three gigabat build out to Core Scientific. Course Scientific. Also, 0:17:55.280 --> 0:17:58.159 it seems they seem to be taking on like one 0:17:58.200 --> 0:18:00.880 point one four billion dollars of capital expenditures to build 0:18:00.920 --> 0:18:03.040 these data centers, which, by the way, is not enough money. 0:18:03.080 --> 0:18:05.600 But nevertheless, Corwave has promised to reimburse them at eight 0:18:05.720 --> 0:18:08.760 hundred and ninety nine point three million of these costs. 0:18:09.160 --> 0:18:11.320 This is all from public filings. By the way, it's 0:18:11.359 --> 0:18:14.240 as one clear house course Scientific actually intends to do 0:18:14.320 --> 0:18:16.640 any of this shit. While they've taken on a good 0:18:16.680 --> 0:18:18.640 amount of debt in the past five hundred and fifty 0:18:18.680 --> 0:18:20.520 million dollars in the convertible note towards the end of 0:18:20.600 --> 0:18:22.800 last year, this would be more debt than they've ever 0:18:22.880 --> 0:18:26.280 taken on. It Also, as with Crusoe, does not appear 0:18:26.320 --> 0:18:28.840 to have any experienced building AI data centers, a point 0:18:28.880 --> 0:18:31.040 I keep repeating because it's very important. These are other 0:18:31.080 --> 0:18:34.480 companies behind the growth for open AI, except unlike Crusoe, 0:18:34.680 --> 0:18:38.240 Core Scientific is a barely functioning, recently bankrupted bitcoin minor 0:18:38.280 --> 0:18:41.119 pretending to be a data center company. Crusoe, on the 0:18:41.200 --> 0:18:43.520 other hand, is possibly also doing the same thing, but 0:18:43.720 --> 0:18:47.240 less egregious about it. Now, how important do you think 0:18:47.320 --> 0:18:50.200 core Weave is to open ai? Exactly? Well, that's our semaphore. 0:18:51.000 --> 0:18:53.240 Core Weaver has been one of our earliest and largest 0:18:53.280 --> 0:18:56.400 compute partners, open ai chief Sam Allman said in Corwave's 0:18:56.480 --> 0:18:59.440 roadshow video, adding that Corwave's compute power led to the 0:18:59.480 --> 0:19:01.240 creation of some of the models that were best known 0:19:01.320 --> 0:19:04.160 for Core. We figured out how to innovate on hardware, 0:19:04.200 --> 0:19:06.840 to innovate on data center construction, and to delve for results, 0:19:07.000 --> 0:19:10.840 very very quickly did it. But even if it did, 0:19:11.359 --> 0:19:14.200 will it survive long term? Going back to the point 0:19:14.240 --> 0:19:17.000 of the contagion. If open ai fails and core we fails, 0:19:17.200 --> 0:19:20.359 so too does course Scientific And I don't really fancy 0:19:20.440 --> 0:19:23.920 Crusoe's chances either. But let's take a step back for 0:19:24.040 --> 0:19:26.840 a moment. We've been going so hard, haven't we. I've 0:19:26.880 --> 0:19:30.280 got a genuine question, just for the fact finders out there. 0:19:31.160 --> 0:19:34.520 There's a Microsoft book. Open AI's computer's revenue now. Up 0:19:34.600 --> 0:19:37.879 until fairly recently, Microsoft has been the entire infrastructure backing 0:19:38.000 --> 0:19:40.960 open ai, but recently to free open ai up to 0:19:41.119 --> 0:19:43.480 work with Oracle and see other people, released it from 0:19:43.480 --> 0:19:46.920 its exclusive cloud compute deal. Nevertheless, put the information, open 0:19:47.000 --> 0:19:49.639 ai still intends to spend thirteen billion dollars on compute 0:19:49.680 --> 0:19:53.080 on Microsoft as there this year. What's confusing, however, is 0:19:53.119 --> 0:19:55.440 whether any of this is booked as revenue for Microsoft. 0:19:55.560 --> 0:19:57.840 Microsoft claimed earlier in the year that it surpassed thirteen 0:19:57.880 --> 0:20:00.920 billion dollars in annual recurring revenue, by which it means 0:20:01.280 --> 0:20:03.840 it's last month multiplied by twelve by the way, and 0:20:03.920 --> 0:20:06.720 they said it was from ai. Open AI's compute costs 0:20:06.760 --> 0:20:09.240 in twenty twenty four or five billion dollars, and that's 0:20:09.240 --> 0:20:12.119 at are discounted as your rate, which on an anualized 0:20:12.119 --> 0:20:15.040 basis will be about four hundred and sixteen million dollars 0:20:15.119 --> 0:20:18.560