1 00:00:02,800 --> 00:00:20,239 Speaker 1: Zone Media. Try spinning. That's a good trick. I'm at Zetron. 2 00:00:20,320 --> 00:00:22,600 Speaker 1: This is Better Offline and this is the second episode 3 00:00:22,600 --> 00:00:25,119 Speaker 1: of my two part series where I explain how open 4 00:00:25,120 --> 00:00:27,600 Speaker 1: ai has become a systemic risk to the tech industry, 5 00:00:27,880 --> 00:00:30,440 Speaker 1: even with its massive forty billion dollar funding round and 6 00:00:30,520 --> 00:00:33,239 Speaker 1: bird brain benefactor in the form of soft Bank, the 7 00:00:33,240 --> 00:00:36,959 Speaker 1: world's foremost authority in losing money. Now, before I continue, 8 00:00:37,040 --> 00:00:39,920 Speaker 1: shameless request, Better Offline has been nominated for a Webbie 9 00:00:39,920 --> 00:00:41,560 Speaker 1: and I want to win this thing. I've linked to 10 00:00:41,640 --> 00:00:43,360 Speaker 1: it in my Twitter and my Blue Sky and if 11 00:00:43,360 --> 00:00:44,519 Speaker 1: you could vote for me, well, it will be in 12 00:00:44,560 --> 00:00:48,879 Speaker 1: the episode notes two. Moving on back at it. Okay, 13 00:00:49,360 --> 00:00:53,559 Speaker 1: all right, open ai now has forty billion dollars somehow, 14 00:00:53,920 --> 00:00:57,120 Speaker 1: right great? Right, Well, Dolgier horses. As part of its 15 00:00:57,160 --> 00:01:00,000 Speaker 1: deal with soft Bank, open Ai must also convert its 16 00:01:00,040 --> 00:01:03,000 Speaker 1: bizarre nonprofit structure into a for profit entity by December 17 00:01:03,000 --> 00:01:05,839 Speaker 1: twenty twenty five, or it will lose ten billion dollars 18 00:01:05,840 --> 00:01:08,600 Speaker 1: from that forty billion dollars and of funding. And just 19 00:01:08,640 --> 00:01:10,600 Speaker 1: to be clear, by the way, they've only really got 20 00:01:10,640 --> 00:01:12,800 Speaker 1: ten billion dollars of that so far. The rest is 21 00:01:12,840 --> 00:01:15,600 Speaker 1: at the end of the year. Furthermore, and the event 22 00:01:15,640 --> 00:01:17,919 Speaker 1: that open ai fails to convert into a for profit 23 00:01:18,000 --> 00:01:21,160 Speaker 1: company by October twenty twenty six, investors in its previous 24 00:01:21,200 --> 00:01:23,560 Speaker 1: six point six billion dollar funding route can claw back 25 00:01:23,600 --> 00:01:25,840 Speaker 1: their investment with a converting into a loan with an 26 00:01:25,840 --> 00:01:29,440 Speaker 1: attached interest rate. Naturally, this represents a night mess scenario 27 00:01:29,480 --> 00:01:32,320 Speaker 1: for the company, as it will increase both its costs 28 00:01:32,360 --> 00:01:35,920 Speaker 1: and its outgoings. This is a complex situation that almost 29 00:01:35,959 --> 00:01:38,640 Speaker 1: warrants its own podcast, But the long and short of 30 00:01:38,680 --> 00:01:41,520 Speaker 1: it is that open ai would have to effectively dissolve itself, 31 00:01:41,720 --> 00:01:44,880 Speaker 1: start the process of reforming an entirely new entity, and 32 00:01:44,959 --> 00:01:48,240 Speaker 1: distribute its assets to other nonprofits or seller license them 33 00:01:48,240 --> 00:01:50,760 Speaker 1: to a for profit company at fair market rates which 34 00:01:50,800 --> 00:01:54,280 Speaker 1: they would not set. It would also require valuing open 35 00:01:54,280 --> 00:01:56,360 Speaker 1: ai as assets, which in and of itself would be 36 00:01:56,400 --> 00:01:58,560 Speaker 1: a difficult task, as well as getting past the necessary 37 00:01:58,640 --> 00:02:02,400 Speaker 1: state regulators, the IRA state revenue agencies, and the upcoming 38 00:02:02,400 --> 00:02:05,800 Speaker 1: trial with Elon Musk Well, that only adds further problems. 39 00:02:06,400 --> 00:02:09,000 Speaker 1: I've simplified things here, and that's because, as I've said, 40 00:02:09,040 --> 00:02:12,440 Speaker 1: this stuff is a little complex and pretty boring. Suffice 41 00:02:12,480 --> 00:02:14,680 Speaker 1: to say, this isn't as simple as liquidating a company 42 00:02:14,680 --> 00:02:17,399 Speaker 1: and starting a fresh or submitting a couple of legal filings. 43 00:02:17,480 --> 00:02:19,400 Speaker 1: It's a long, fraught process and one that will be 44 00:02:19,480 --> 00:02:22,799 Speaker 1: as as has been subject to legal challenges, both from 45 00:02:22,800 --> 00:02:25,560 Speaker 1: open AI's business rivals as well as from civil society 46 00:02:25,680 --> 00:02:28,519 Speaker 1: organizations in California. You may have heard the lost monologue. 47 00:02:29,200 --> 00:02:31,240 Speaker 1: Based on discussions with experts in the field of my 48 00:02:31,280 --> 00:02:33,480 Speaker 1: own research, I simply do not know how open ai 49 00:02:33,639 --> 00:02:37,320 Speaker 1: pulls off this by October twenty twenty six, and honestly, 50 00:02:37,360 --> 00:02:38,880 Speaker 1: I'm not sure how they do it by the end 51 00:02:38,880 --> 00:02:43,200 Speaker 1: of this year. It's insane. It's a it's a really 52 00:02:44,480 --> 00:02:47,400 Speaker 1: I just every time I read this stuff and I 53 00:02:47,440 --> 00:02:49,840 Speaker 1: write out, I'm like, how is nobody else reading this 54 00:02:49,960 --> 00:02:52,040 Speaker 1: and going, what the fuck is going on? You See, 55 00:02:52,040 --> 00:02:55,640 Speaker 1: this is a big problem, this nonprofit thing, because open 56 00:02:55,680 --> 00:02:58,480 Speaker 1: eye really has become a systemic risk to the tech industry, 57 00:02:58,480 --> 00:03:01,519 Speaker 1: and anything that increases that is bad news for everybody. 58 00:03:02,080 --> 00:03:05,519 Speaker 1: Open Ai, they've become a kind of load bearing company 59 00:03:05,520 --> 00:03:07,880 Speaker 1: for this industry, both as a narrative, as I've discussed 60 00:03:07,919 --> 00:03:10,680 Speaker 1: multiple times as chat GPT is the only large language 61 00:03:10,800 --> 00:03:13,560 Speaker 1: model company with any meaningful use base, and also as 62 00:03:13,600 --> 00:03:16,920 Speaker 1: a financial entity. Its ability to meet its obligations and 63 00:03:16,960 --> 00:03:19,800 Speaker 1: its future expansion plans are critical to the future health 64 00:03:19,880 --> 00:03:22,800 Speaker 1: or in some cases survival of multiple large companies. And 65 00:03:22,800 --> 00:03:25,320 Speaker 1: that's before the after effects that will affect its customers 66 00:03:25,320 --> 00:03:28,320 Speaker 1: as a result of any kind of financial collapse. The 67 00:03:28,360 --> 00:03:30,560 Speaker 1: parallels to the two thousand and seven the two thousand 68 00:03:30,560 --> 00:03:34,080 Speaker 1: and eight financial crisis are starting to become a little worrying. 69 00:03:34,320 --> 00:03:37,000 Speaker 1: Layman Brothers wasn't the largest investment bank in the world, 70 00:03:37,120 --> 00:03:39,720 Speaker 1: although it was pretty big, just like open Ai isn't 71 00:03:39,760 --> 00:03:42,600 Speaker 1: the largest tech company, though again it's certainly large in 72 00:03:42,680 --> 00:03:47,360 Speaker 1: terms of alleged valuation and expenditures. Layman Brothers collapse sparked 73 00:03:47,360 --> 00:03:50,080 Speaker 1: a contagient that would later spread throughout the entire global 74 00:03:50,120 --> 00:03:54,120 Speaker 1: financial services industry and consequently the global economy. Now I 75 00:03:54,160 --> 00:03:57,680 Speaker 1: can see open AI's failure not having as big an effect, 76 00:03:57,760 --> 00:04:02,400 Speaker 1: but I can imagine a systemic effect. Still, you have 77 00:04:02,480 --> 00:04:06,280 Speaker 1: to realize that the whole AI trade, the narrative, the bubble, 78 00:04:06,320 --> 00:04:09,240 Speaker 1: it's holding up the economy. I think like thirty thirty 79 00:04:09,240 --> 00:04:11,680 Speaker 1: five percent of the US stock market is in the 80 00:04:11,680 --> 00:04:14,760 Speaker 1: Magnificent seven, and all of their bullshit numbers right now 81 00:04:14,800 --> 00:04:18,240 Speaker 1: are held up by this nonsense. And like the financial crisis, 82 00:04:18,720 --> 00:04:20,640 Speaker 1: the impact in this case won't be limited to just 83 00:04:20,680 --> 00:04:24,440 Speaker 1: bankers and insurers. It will bleed into everything else. This 84 00:04:24,520 --> 00:04:26,680 Speaker 1: episode is going to be a bit grim. I'm not 85 00:04:26,720 --> 00:04:28,680 Speaker 1: going to lie. I want to lay out the direct 86 00:04:28,720 --> 00:04:31,280 Speaker 1: result of any kind of financial crisis at open Ai, 87 00:04:31,320 --> 00:04:34,640 Speaker 1: because I don't think anybody is taking this seriously. Let's 88 00:04:34,680 --> 00:04:36,920 Speaker 1: start with Oracle, who will lose at least a billion 89 00:04:36,960 --> 00:04:42,080 Speaker 1: dollars if open ai doesn't fulfill its obligations per the Information. Oracle, 90 00:04:42,120 --> 00:04:44,640 Speaker 1: which has taken responsibility for organizing the construction of the 91 00:04:44,640 --> 00:04:47,960 Speaker 1: stargate data centers with unproven data center build a Cruso 92 00:04:48,240 --> 00:04:50,240 Speaker 1: and I quote the Information here, may need to raise 93 00:04:50,240 --> 00:04:53,240 Speaker 1: more capital to fund its data center ambitions. Oracle has 94 00:04:53,279 --> 00:04:55,640 Speaker 1: signed a fifteen year lease with Crusoe, and to quote 95 00:04:55,640 --> 00:04:57,920 Speaker 1: the Information is on the hook for one billion dollars 96 00:04:57,960 --> 00:05:01,200 Speaker 1: in payments to that firm. To further quote information. While 97 00:05:01,240 --> 00:05:03,440 Speaker 1: that's the standard deal length, the unprecedented size of the 98 00:05:03,440 --> 00:05:06,280 Speaker 1: facility Oracle is building for just one customerment makes its 99 00:05:06,400 --> 00:05:08,800 Speaker 1: riskier than the standard cloud data center used by lots 100 00:05:08,800 --> 00:05:12,200 Speaker 1: of interchangeable customers with much more predictable needs. According to 101 00:05:12,200 --> 00:05:14,400 Speaker 1: half a dozen people familiar with these types of deals. 102 00:05:14,960 --> 00:05:17,599 Speaker 1: In simpler terms, Oracle is building a giant data center 103 00:05:17,600 --> 00:05:20,599 Speaker 1: for one customer, open Ai, and has taken on the 104 00:05:20,600 --> 00:05:23,800 Speaker 1: financial burden associated with them. If open Ai fails to 105 00:05:23,839 --> 00:05:25,880 Speaker 1: expand or lacks the capital to actually pay for its 106 00:05:25,920 --> 00:05:28,480 Speaker 1: share of the Stargate data center project, Oracle is on 107 00:05:28,520 --> 00:05:30,400 Speaker 1: the hook for at least a billion dollars, and based 108 00:05:30,440 --> 00:05:32,400 Speaker 1: on the Information's reporting, it's also on the hooked by 109 00:05:32,440 --> 00:05:35,320 Speaker 1: the GPUs for the site. This is me quoting them again. 110 00:05:36,960 --> 00:05:39,480 Speaker 1: Even before the Stargate announcement, Oracle and open Ai had 111 00:05:39,480 --> 00:05:41,599 Speaker 1: agreed to expand their Abelene deal from two to eight 112 00:05:41,680 --> 00:05:43,920 Speaker 1: data center buildings, which can hold four hundred thousand and 113 00:05:44,000 --> 00:05:46,760 Speaker 1: VIDIA Blackwell GPUs, adding tens of billions of dollars to 114 00:05:46,800 --> 00:05:50,200 Speaker 1: the cost of the facility. In reality, this development will 115 00:05:50,240 --> 00:05:53,360 Speaker 1: likely cost tens of billions of dollars, nineteen billion dollars 116 00:05:53,360 --> 00:05:55,160 Speaker 1: of which is due from open Ai, which does not 117 00:05:55,279 --> 00:05:57,320 Speaker 1: have the money until it receives its second tranche of 118 00:05:57,320 --> 00:06:00,560 Speaker 1: funding in December twenty twenty five from SoftBank, and this 119 00:06:00,760 --> 00:06:02,919 Speaker 1: is contingent partially on their ability to convert into a 120 00:06:02,920 --> 00:06:06,039 Speaker 1: for profit entity, which has mentioned is extremely difficult and 121 00:06:06,120 --> 00:06:10,039 Speaker 1: extremely unlikely. It's unclear how many of the Blackwell GPUs 122 00:06:10,040 --> 00:06:12,320 Speaker 1: the Oracle has had to purchase in advance, but in 123 00:06:12,360 --> 00:06:15,200 Speaker 1: the event of any kind of financial collapse, open Ai 124 00:06:15,279 --> 00:06:18,360 Speaker 1: Oracle will likely have to toss at least a billion dollars, 125 00:06:18,360 --> 00:06:21,279 Speaker 1: if not several billion dollars, and then we get the 126 00:06:21,360 --> 00:06:25,080 Speaker 1: core with a companies. So a company whose expansion is 127 00:06:25,160 --> 00:06:28,239 Speaker 1: likely driven entirely by open Ai now and cannot survive 128 00:06:28,279 --> 00:06:31,240 Speaker 1: without open Ai flilling its obligations if it doesn't die. 129 00:06:31,240 --> 00:06:34,120 Speaker 1: O Anyway, Now, I've written and spoken a lot about 130 00:06:34,279 --> 00:06:36,880 Speaker 1: publicly traded AI compute firm Core Weave, and it would 131 00:06:36,880 --> 00:06:38,560 Speaker 1: give me the greatest pleasure in my life. Never think 132 00:06:38,680 --> 00:06:41,560 Speaker 1: or talk about them ever again. Nevertheless, I have to. 133 00:06:42,080 --> 00:06:45,559 Speaker 1: This is my curse. This is my curse. Core Weave 134 00:06:45,560 --> 00:06:48,000 Speaker 1: has become my curse every time I think about this. 135 00:06:48,080 --> 00:06:52,320 Speaker 1: Fuck Okay. The Financial Times revealed a few weeks ago 136 00:06:52,360 --> 00:06:54,360 Speaker 1: that core Weave's debt payments could balloon to over two 137 00:06:54,400 --> 00:06:56,120 Speaker 1: point four billion dollars a year by the end of 138 00:06:56,120 --> 00:06:59,120 Speaker 1: twenty twenty five, far outstripping its cash reserves, and the 139 00:06:59,160 --> 00:07:01,799 Speaker 1: Information reported that its cash burn would increase to fifteen 140 00:07:01,839 --> 00:07:05,560 Speaker 1: billion dollars in twenty twenty five, as Bird's IBO filing. 141 00:07:05,600 --> 00:07:08,080 Speaker 1: Sixty two percent of core weaves twenty twenty four revenue 142 00:07:08,120 --> 00:07:10,400 Speaker 1: a little under two billion, with losses amounting to eight 143 00:07:10,520 --> 00:07:13,440 Speaker 1: hundred and sixty three million was Microsoft Compute, and based 144 00:07:13,480 --> 00:07:15,880 Speaker 1: on the conversations I've had with sources, a good amount 145 00:07:15,920 --> 00:07:19,200 Speaker 1: of this was Microsoft running compute for open Ai. Starting 146 00:07:19,200 --> 00:07:21,880 Speaker 1: October twenty twenty five, open Ai will start paying core 147 00:07:21,920 --> 00:07:24,280 Speaker 1: Weave as part of its five year long, twelve billion 148 00:07:24,320 --> 00:07:27,840 Speaker 1: dollar contract, picking up the option that Microsoft declined. This 149 00:07:28,360 --> 00:07:30,920 Speaker 1: is not great timing, or maybe it's perfect timing, because 150 00:07:30,960 --> 00:07:33,160 Speaker 1: this is also when core Weave will have to start 151 00:07:33,160 --> 00:07:36,200 Speaker 1: making payments on their massive, stupid, multi billion dollar d 152 00:07:36,280 --> 00:07:40,360 Speaker 1: DTL two point zero loan mentioned in previous episodes. But really, 153 00:07:40,360 --> 00:07:42,760 Speaker 1: there's a newsletter if you want to you hear me, 154 00:07:42,800 --> 00:07:44,320 Speaker 1: go mad, You want to read me go mad? You 155 00:07:44,360 --> 00:07:46,480 Speaker 1: read my read my core with piece because it really 156 00:07:46,560 --> 00:07:51,679 Speaker 1: drove me insane. Nevertheless, these core Weave payments, the ones 157 00:07:52,000 --> 00:07:54,600 Speaker 1: from open Ai to core Weave that October, they're pretty 158 00:07:54,680 --> 00:07:58,320 Speaker 1: much critical to Corewave's future. This deal also suggests that 159 00:07:58,400 --> 00:08:01,840 Speaker 1: open ai will become core Weave's life just customer. Microsoft 160 00:08:01,840 --> 00:08:04,760 Speaker 1: had previously committed to spending ten billion dollars on core 161 00:08:04,800 --> 00:08:06,920 Speaker 1: Weave services by the end of the decade, but CEO 162 00:08:06,960 --> 00:08:09,400 Speaker 1: satchly Adella added a few months later on a podcast 163 00:08:09,600 --> 00:08:13,440 Speaker 1: that its relationship with core Weave was a one time thing. Man, 164 00:08:14,160 --> 00:08:16,720 Speaker 1: man a really like really fucking around there, such a 165 00:08:16,760 --> 00:08:19,600 Speaker 1: don't love core Weave. Assuming Microsoft keeps spending it its 166 00:08:19,640 --> 00:08:23,720 Speaker 1: previous rate, So about one point like sixty six percent 167 00:08:23,880 --> 00:08:26,840 Speaker 1: of two billion dollars whatever those I have something that 168 00:08:26,880 --> 00:08:28,880 Speaker 1: isn't guaranteed. By the way, it would still only be 169 00:08:29,000 --> 00:08:32,000 Speaker 1: half of open AI's potential revenue to core Weave. Core 170 00:08:32,040 --> 00:08:34,960 Speaker 1: Weave's expansion at this point is entirely driven by open Ai. 171 00:08:35,440 --> 00:08:38,000 Speaker 1: Seventy seven percent of its twenty twenty four revenue came 172 00:08:38,000 --> 00:08:40,360 Speaker 1: from two customers, Microsoft being the largest and yes, I 173 00:08:40,480 --> 00:08:42,800 Speaker 1: just fucked up a number at sixty two percent and 174 00:08:42,880 --> 00:08:46,160 Speaker 1: using core weaves auxiliary compute for open Ai. As a result, 175 00:08:46,200 --> 00:08:49,520 Speaker 1: the future expansion efforts the theoretical one point three gigawatts 176 00:08:49,559 --> 00:08:52,120 Speaker 1: have contracted, and by the way, that means it doesn't exist. 177 00:08:52,120 --> 00:08:54,760 Speaker 1: Compute at core Weaver are largely, if not entirely, for 178 00:08:54,800 --> 00:08:57,160 Speaker 1: the benefit of open Ai. In the event that open 179 00:08:57,200 --> 00:09:00,360 Speaker 1: Ai cannot fulfill its obligations, core weave will collapse. It's 180 00:09:00,360 --> 00:09:03,600 Speaker 1: that fucking simple, and then the shock waves will ripple further. 181 00:09:04,120 --> 00:09:06,560 Speaker 1: In Video relies on corewey for more than six percent 182 00:09:06,600 --> 00:09:09,800 Speaker 1: of its revenue and corwy's future credit worthiness to continue 183 00:09:09,800 --> 00:09:13,440 Speaker 1: receiving said revenue. Well, much of that is dependent on 184 00:09:13,480 --> 00:09:16,640 Speaker 1: open ai continuing to buy services from core Weave now 185 00:09:16,640 --> 00:09:18,319 Speaker 1: and basically this in a comment I received from the 186 00:09:18,400 --> 00:09:21,640 Speaker 1: legendary Gil Luria, Managing director and head of Technology research 187 00:09:21,679 --> 00:09:24,760 Speaker 1: at Analyst DA Davison and Co, I quote him, since 188 00:09:24,880 --> 00:09:27,599 Speaker 1: cor We've bought two hundred thousand GPUs last year, and 189 00:09:27,640 --> 00:09:30,360 Speaker 1: those systems are around forty thousand dollars. We believe cor 190 00:09:30,360 --> 00:09:32,880 Speaker 1: We've spent eight billion dollars on in Vidia last year. 191 00:09:33,040 --> 00:09:36,400 Speaker 1: That represents more than six percent of Nvidia's revenue in 192 00:09:36,440 --> 00:09:38,680 Speaker 1: twenty twenty four, he said last year, but I've just 193 00:09:38,679 --> 00:09:41,760 Speaker 1: wanted to make it sound better. Core We've receives preferential 194 00:09:41,760 --> 00:09:45,320 Speaker 1: access to Nvidia's GPUs, though Nvidia kind of denies that 195 00:09:45,400 --> 00:09:48,600 Speaker 1: and makes up billions of dollars of Nvidia's revenue. Corweave 196 00:09:48,600 --> 00:09:51,600 Speaker 1: then takes those GPUs and then they raise debt using 197 00:09:51,640 --> 00:09:56,000 Speaker 1: the GPU's as collateral as well as customer contracts. Then 198 00:09:56,320 --> 00:09:58,920 Speaker 1: they use the money they've raised to buy more GPUs 199 00:09:58,920 --> 00:10:01,640 Speaker 1: from Nvidia. You may think that doesn't sound right. I 200 00:10:01,679 --> 00:10:04,440 Speaker 1: am being complete, like this is factual information. At this 201 00:10:04,520 --> 00:10:07,480 Speaker 1: point in video was the anchor for Corwave's IPO, and 202 00:10:07,559 --> 00:10:10,160 Speaker 1: CEO Michael and Trader said that the IPO would not 203 00:10:10,320 --> 00:10:13,240 Speaker 1: have closed without Invidia buying two hundred and fifty million 204 00:10:13,280 --> 00:10:16,680 Speaker 1: dollars worth of their shares. Nvidia also invested one hundred 205 00:10:16,679 --> 00:10:18,640 Speaker 1: million dollars in the early days of core weaves, and 206 00:10:18,679 --> 00:10:21,839 Speaker 1: for reasons I cannot understand, also agreed to spend one 207 00:10:21,840 --> 00:10:24,240 Speaker 1: point three billion dollars over four years two and I 208 00:10:24,320 --> 00:10:27,360 Speaker 1: quote the information rent its own chips from core weave 209 00:10:28,040 --> 00:10:30,880 Speaker 1: fum fact. I can't find a single fucking mention of 210 00:10:30,960 --> 00:10:34,480 Speaker 1: core weave in any of Nvidia's filings. Now buried in 211 00:10:34,559 --> 00:10:37,320 Speaker 1: that core weaves s one the document every company publishes 212 00:10:37,360 --> 00:10:40,440 Speaker 1: before going public was a warning about counterparty credit risk, 213 00:10:40,559 --> 00:10:42,800 Speaker 1: which is when one party provides services or goods to 214 00:10:42,840 --> 00:10:46,320 Speaker 1: another with specific repayment terms and the other party doesn't 215 00:10:46,320 --> 00:10:49,160 Speaker 1: meet their side of the deal. While this was written 216 00:10:49,160 --> 00:10:52,600 Speaker 1: as a theoretical as it could, in theoretically speaking, come 217 00:10:52,679 --> 00:10:55,040 Speaker 1: from any company to which core Weave acts as a creditor. 218 00:10:55,480 --> 00:11:00,280 Speaker 1: It only named one open Ai now has discussed previously. 219 00:11:00,360 --> 00:11:03,760 Speaker 1: Core Weaver is saying that should a customer, any customer, 220 00:11:03,800 --> 00:11:06,320 Speaker 1: but really they mean open Ai failed to pay its 221 00:11:06,360 --> 00:11:09,280 Speaker 1: bills for infrastructure built on their behalf or services rendered, 222 00:11:09,559 --> 00:11:11,880 Speaker 1: it can have a material risk of the company. Now. 223 00:11:12,000 --> 00:11:14,520 Speaker 1: As an aside, the information reported that Google and someone's 224 00:11:14,520 --> 00:11:15,839 Speaker 1: going to email me there, so I just want to 225 00:11:15,880 --> 00:11:18,000 Speaker 1: get ahead of it. The core Weave is apparently in 226 00:11:18,120 --> 00:11:21,400 Speaker 1: advanced talks with Google to nd GPUs it Also, it 227 00:11:21,480 --> 00:11:23,439 Speaker 1: also added another thing in this story, just so that 228 00:11:23,520 --> 00:11:25,280 Speaker 1: I don't have to hear from any of you. The 229 00:11:25,360 --> 00:11:29,400 Speaker 1: Google's potential deal with core Weaves significantly smaller than their 230 00:11:29,400 --> 00:11:31,800 Speaker 1: commitments with Microsoft as according to one of the people 231 00:11:31,880 --> 00:11:34,440 Speaker 1: briefed on it, but could potentially expand in future years. 232 00:11:34,760 --> 00:11:36,720 Speaker 1: Do not come to me and claim that Google's going 233 00:11:36,760 --> 00:11:39,800 Speaker 1: to save core Weeve. I'll be so mad anyway. Even 234 00:11:39,840 --> 00:11:42,679 Speaker 1: with Google and open AI's money, Carew's continued the ability 235 00:11:42,720 --> 00:11:44,959 Speaker 1: to do business hinges heavily on its ability to raise 236 00:11:45,000 --> 00:11:47,599 Speaker 1: further debt, which I have previously called into question a 237 00:11:47,679 --> 00:11:50,320 Speaker 1: newsletter that gave me madness, and its ability to raise 238 00:11:50,640 --> 00:11:54,160 Speaker 1: future debt is to quote the financial times secured against 239 00:11:54,160 --> 00:11:56,880 Speaker 1: it's more than two hundred and fifty thousand n vida 240 00:11:56,960 --> 00:12:02,319 Speaker 1: GPUs and its contracts with customers such as Microsoft. Now, 241 00:12:02,720 --> 00:12:05,280 Speaker 1: any future debt that core Weave raises will be based 242 00:12:05,280 --> 00:12:07,079 Speaker 1: off of its contract with open Ai. You know, the 243 00:12:07,160 --> 00:12:10,800 Speaker 1: counterparty credit risk threat that represents a disproportionate sharef it's 244 00:12:10,840 --> 00:12:13,840 Speaker 1: revenue I just mentioned, and also whatever GPUs they still 245 00:12:13,920 --> 00:12:16,319 Speaker 1: have left that they can get debt on. As a result, 246 00:12:16,360 --> 00:12:18,240 Speaker 1: a chunk of a video's future revenue is dependent on 247 00:12:18,320 --> 00:12:20,640 Speaker 1: open AI's ability to fulfill its obligations the core Weave 248 00:12:20,679 --> 00:12:22,320 Speaker 1: both in its ability to pay them, and they're timing 249 00:12:22,360 --> 00:12:24,760 Speaker 1: less in doing so. If open ai fails, then core 250 00:12:24,840 --> 00:12:27,719 Speaker 1: Weave fails, then that hurts and video Jensen's going to 251 00:12:27,760 --> 00:12:29,360 Speaker 1: have to go. He's going to have to go to 252 00:12:29,559 --> 00:12:34,680 Speaker 1: a cheaper leather jacketarium and it gets worse. Open AI's 253 00:12:34,720 --> 00:12:37,520 Speaker 1: expansion is dependent on two unproven startups, one of them 254 00:12:37,559 --> 00:12:40,920 Speaker 1: I just mentioned, who are also dependent on open ai 255 00:12:41,040 --> 00:12:44,040 Speaker 1: to live with Microsoft's data center pullback and open AI's 256 00:12:44,040 --> 00:12:46,920 Speaker 1: intend to become independent from redmen. Future data center expansion 257 00:12:47,000 --> 00:12:49,400 Speaker 1: is based on two partners supporting coll Weave I know 258 00:12:50,520 --> 00:12:53,560 Speaker 1: will get there and Oracle. Now I'm referring, of course 259 00:12:53,600 --> 00:12:56,920 Speaker 1: to Core Scientific, which is the data center developer for 260 00:12:57,400 --> 00:13:00,280 Speaker 1: core Weave, and of course Crusos, the data center develer 261 00:13:00,400 --> 00:13:03,040 Speaker 1: four Oracle. Now, if you were wondering, I can't hint 262 00:13:03,120 --> 00:13:06,280 Speaker 1: it about this earlier how many data center how many 263 00:13:06,400 --> 00:13:09,599 Speaker 1: DIIDENTA centers do you think Cruso's ever built, and the 264 00:13:09,679 --> 00:13:12,559 Speaker 1: answer is none, And of course Scientific, how many do 265 00:13:12,559 --> 00:13:14,520 Speaker 1: you think they've built and the answer is also none. 266 00:13:14,880 --> 00:13:19,280 Speaker 1: These are the fucking companies underpending the AI boom. I 267 00:13:19,360 --> 00:13:21,920 Speaker 1: also really must explain how difficult it is to build 268 00:13:21,920 --> 00:13:24,319 Speaker 1: a data center, and how said difficult he increases when 269 00:13:24,320 --> 00:13:27,120 Speaker 1: you're building an AI focused one. For example, in Video 270 00:13:27,160 --> 00:13:29,240 Speaker 1: had to delay the launch of its Blackwell GPUs because 271 00:13:29,240 --> 00:13:32,079 Speaker 1: of how finnicky the associated infrastructure, so the servers and 272 00:13:32,120 --> 00:13:35,400 Speaker 1: the cooling and such is for customers. This was for 273 00:13:35,480 --> 00:13:39,000 Speaker 1: customers that had already been using GPUs and therefore likely 274 00:13:39,120 --> 00:13:41,920 Speaker 1: knew how to manage the temperatures created by them. Also 275 00:13:42,080 --> 00:13:44,400 Speaker 1: is another reminder open eyes on the hook for nineteen 276 00:13:44,400 --> 00:13:47,520 Speaker 1: billion dollars of funding behind Stargate, and neither of them 277 00:13:47,640 --> 00:13:49,760 Speaker 1: have that money. I just want to remind you of that, 278 00:13:49,880 --> 00:13:52,320 Speaker 1: because it costs so much money to build a fucking 279 00:13:52,440 --> 00:13:55,760 Speaker 1: data center. And imagine if you didn't have any experience 280 00:13:55,920 --> 00:13:58,079 Speaker 1: and effectively had to learn from scratch, how do you 281 00:13:58,320 --> 00:14:03,640 Speaker 1: think it would be building these data centers? Let's find out. 282 00:14:04,120 --> 00:14:06,959 Speaker 1: So let's start in Abilene, Texas with Crusoe and the 283 00:14:07,040 --> 00:14:10,360 Speaker 1: Stargate Data Center project. Now. Cruso is a former cryptocurrency 284 00:14:10,400 --> 00:14:12,439 Speaker 1: mining company that has now raised hundreds of millions of 285 00:14:12,520 --> 00:14:15,400 Speaker 1: dollars to build data centers for AI companies, starting with 286 00:14:15,440 --> 00:14:17,920 Speaker 1: a three point four billion dollar data center financing deal 287 00:14:17,960 --> 00:14:20,960 Speaker 1: with asset manager Blue Owl Capital. This yet to be 288 00:14:21,120 --> 00:14:23,640 Speaker 1: completed data center has now been leased by Oracle, which 289 00:14:23,680 --> 00:14:26,160 Speaker 1: will allegedly fill it full of GPUs for open AI. 290 00:14:26,720 --> 00:14:29,640 Speaker 1: Despite calling itself and I quote the industry's first vertically 291 00:14:29,720 --> 00:14:33,360 Speaker 1: integrated AI infrastructure provider, with the company using flared gas 292 00:14:33,400 --> 00:14:35,560 Speaker 1: as a waste by product of oil production to power 293 00:14:35,640 --> 00:14:38,240 Speaker 1: I infrastructure, Cruso does not appear to have built a 294 00:14:38,320 --> 00:14:40,520 Speaker 1: single AI data center and is now being tasked with 295 00:14:40,600 --> 00:14:45,360 Speaker 1: building one point two Gigawatt's had a data center capacity 296 00:14:45,440 --> 00:14:49,120 Speaker 1: for open Ai. It's just so fucking Cruso is the 297 00:14:49,240 --> 00:14:52,080 Speaker 1: sole developer and operator of the Abilene site, meaning, according 298 00:14:52,080 --> 00:14:54,360 Speaker 1: to the information that it is in charge of contracting 299 00:14:54,440 --> 00:14:57,560 Speaker 1: with construction contractors and data center customers as well as 300 00:14:57,640 --> 00:15:01,280 Speaker 1: running the data center after it is built. Oracle, it seems, 301 00:15:01,280 --> 00:15:03,720 Speaker 1: will be responsible for filling said data center with GPUs 302 00:15:03,720 --> 00:15:08,080 Speaker 1: as mentioned. Nevertheless, the project also appears to be behind schedule. 303 00:15:08,440 --> 00:15:11,240 Speaker 1: The Information reported in October twenty twenty four that Abilene 304 00:15:11,280 --> 00:15:13,720 Speaker 1: was meant to have fifty thousand of Nvidia's Blackwell AI 305 00:15:13,920 --> 00:15:15,960 Speaker 1: chips in the first quarter of twenty twenty five, and 306 00:15:16,000 --> 00:15:18,320 Speaker 1: also suggested that the site was projected to have a 307 00:15:18,440 --> 00:15:20,400 Speaker 1: whopping one hundred thousand of them by the end of 308 00:15:20,440 --> 00:15:23,360 Speaker 1: twenty twenty five. Now you can join me back here 309 00:15:23,440 --> 00:15:26,000 Speaker 1: in reality, because a report from Bloomberg in March twenty 310 00:15:26,080 --> 00:15:29,040 Speaker 1: twenty five said the Open AI and Oracle we're expected 311 00:15:29,040 --> 00:15:32,080 Speaker 1: to have sixteen thousand available by the summer of twenty 312 00:15:32,120 --> 00:15:35,000 Speaker 1: twenty five, with and a quote open Ai and Oracle 313 00:15:35,080 --> 00:15:38,000 Speaker 1: expecting to deploy sixty four thousand then video GB two 314 00:15:38,160 --> 00:15:41,200 Speaker 1: hundreds at the Stargate Data center by the end of 315 00:15:41,280 --> 00:15:48,240 Speaker 1: twenty twenty six. That's that's very delayed. That's really delayed. Again. 316 00:15:48,440 --> 00:15:50,520 Speaker 1: How I run a PR firm in that I record 317 00:15:50,520 --> 00:15:53,040 Speaker 1: a podcast, I've write a newsletter, I have a book. 318 00:15:53,040 --> 00:15:54,760 Speaker 1: I'm right in. I got all this shit on and 319 00:15:54,800 --> 00:15:58,160 Speaker 1: I'm the asshole who notices this anyway, has discussed previously. 320 00:15:58,240 --> 00:16:01,680 Speaker 1: Open ai needs this capacity very bad. According to the information, 321 00:16:02,000 --> 00:16:04,640 Speaker 1: open ai expect stargate to handle three quarters of its 322 00:16:04,680 --> 00:16:07,120 Speaker 1: compute by twenty thirty and these delays call into question, 323 00:16:07,160 --> 00:16:09,320 Speaker 1: at the very least whether this schedule is reasonable, or 324 00:16:09,400 --> 00:16:13,680 Speaker 1: logical or even possible. And I actually really question whether 325 00:16:13,800 --> 00:16:16,800 Speaker 1: Stargate itself is possible at this point. But it can 326 00:16:16,880 --> 00:16:19,480 Speaker 1: get dumber because we're about to talk about Core Scientific, 327 00:16:19,880 --> 00:16:22,960 Speaker 1: and they are core Weave's friends. They're the people building 328 00:16:23,040 --> 00:16:36,280 Speaker 1: data centers for core Wave in Denton, Texas. Now, as 329 00:16:36,360 --> 00:16:38,520 Speaker 1: you can probably tell, I've written a great deal about 330 00:16:38,560 --> 00:16:40,200 Speaker 1: core Wave in the past. That got a monologue, got 331 00:16:40,240 --> 00:16:42,960 Speaker 1: a newsletter, and I got a therapy bill for it. 332 00:16:43,080 --> 00:16:46,560 Speaker 1: And specifically I've written about their build out partner, Core Scientific, 333 00:16:47,000 --> 00:16:51,040 Speaker 1: a cryptocurrency mining company, yes, another one that has exactly 334 00:16:51,120 --> 00:16:53,360 Speaker 1: one customer for its AI data centers, and you'll never 335 00:16:53,440 --> 00:16:56,160 Speaker 1: guess who it is. It's Core Wave. Now here's a 336 00:16:56,200 --> 00:16:59,800 Speaker 1: few fun facts about Core Scientific. Core Scientific was bankrupt 337 00:16:59,840 --> 00:17:02,720 Speaker 1: lotat year. Course Scientific has never built an AI data center, 338 00:17:02,760 --> 00:17:05,640 Speaker 1: and its cryptocurrency mining operations were built around a six 339 00:17:06,040 --> 00:17:09,400 Speaker 1: specialist computers for mining bitcoin, which led to an analyst 340 00:17:09,480 --> 00:17:12,520 Speaker 1: to tell CNBC that said data centers would and I 341 00:17:12,600 --> 00:17:14,440 Speaker 1: quote need to be buildozed and built from the ground 342 00:17:14,520 --> 00:17:18,520 Speaker 1: up to accommodate AI compute. That's the stuff. Course Scientific 343 00:17:18,560 --> 00:17:21,080 Speaker 1: also does not appear to have any meaningful AI compute 344 00:17:21,080 --> 00:17:24,000 Speaker 1: of any kind. It's AI slash HPC, which is high 345 00:17:24,040 --> 00:17:28,639 Speaker 1: performance computing. Revenue represents a teeny tiny, teeny little percentage 346 00:17:28,680 --> 00:17:31,840 Speaker 1: of overall revenue, which mostly comes from mining crypto, both 347 00:17:31,840 --> 00:17:35,879 Speaker 1: for itself and other parties. Now, hearing all of this, 348 00:17:36,920 --> 00:17:41,639 Speaker 1: would you give this company your your compute? Would you 349 00:17:41,720 --> 00:17:43,960 Speaker 1: think these are the people that I am going to 350 00:17:44,080 --> 00:17:47,680 Speaker 1: call to build my data centers. If you said no, 351 00:17:48,080 --> 00:17:50,520 Speaker 1: you are smarter than Corewave, who has given their entire 352 00:17:50,560 --> 00:17:55,160 Speaker 1: one point three gigabat build out to Core Scientific. Course Scientific. Also, 353 00:17:55,280 --> 00:17:58,159 Speaker 1: it seems they seem to be taking on like one 354 00:17:58,200 --> 00:18:00,880 Speaker 1: point one four billion dollars of capital expenditures to build 355 00:18:00,920 --> 00:18:03,040 Speaker 1: these data centers, which, by the way, is not enough money. 356 00:18:03,080 --> 00:18:05,600 Speaker 1: But nevertheless, Corwave has promised to reimburse them at eight 357 00:18:05,720 --> 00:18:08,760 Speaker 1: hundred and ninety nine point three million of these costs. 358 00:18:09,160 --> 00:18:11,320 Speaker 1: This is all from public filings. By the way, it's 359 00:18:11,359 --> 00:18:14,240 Speaker 1: as one clear house course Scientific actually intends to do 360 00:18:14,320 --> 00:18:16,640 Speaker 1: any of this shit. While they've taken on a good 361 00:18:16,680 --> 00:18:18,640 Speaker 1: amount of debt in the past five hundred and fifty 362 00:18:18,680 --> 00:18:20,520 Speaker 1: million dollars in the convertible note towards the end of 363 00:18:20,600 --> 00:18:22,800 Speaker 1: last year, this would be more debt than they've ever 364 00:18:22,880 --> 00:18:26,280 Speaker 1: taken on. It Also, as with Crusoe, does not appear 365 00:18:26,320 --> 00:18:28,840 Speaker 1: to have any experienced building AI data centers, a point 366 00:18:28,880 --> 00:18:31,040 Speaker 1: I keep repeating because it's very important. These are other 367 00:18:31,080 --> 00:18:34,480 Speaker 1: companies behind the growth for open AI, except unlike Crusoe, 368 00:18:34,680 --> 00:18:38,240 Speaker 1: Core Scientific is a barely functioning, recently bankrupted bitcoin minor 369 00:18:38,280 --> 00:18:41,119 Speaker 1: pretending to be a data center company. Crusoe, on the 370 00:18:41,200 --> 00:18:43,520 Speaker 1: other hand, is possibly also doing the same thing, but 371 00:18:43,720 --> 00:18:47,240 Speaker 1: less egregious about it. Now, how important do you think 372 00:18:47,320 --> 00:18:50,200 Speaker 1: core Weave is to open ai? Exactly? Well, that's our semaphore. 373 00:18:51,000 --> 00:18:53,240 Speaker 1: Core Weaver has been one of our earliest and largest 374 00:18:53,280 --> 00:18:56,400 Speaker 1: compute partners, open ai chief Sam Allman said in Corwave's 375 00:18:56,480 --> 00:18:59,440 Speaker 1: roadshow video, adding that Corwave's compute power led to the 376 00:18:59,480 --> 00:19:01,240 Speaker 1: creation of some of the models that were best known 377 00:19:01,320 --> 00:19:04,160 Speaker 1: for Core. We figured out how to innovate on hardware, 378 00:19:04,200 --> 00:19:06,840 Speaker 1: to innovate on data center construction, and to delve for results, 379 00:19:07,000 --> 00:19:10,840 Speaker 1: very very quickly did it. But even if it did, 380 00:19:11,359 --> 00:19:14,200 Speaker 1: will it survive long term? Going back to the point 381 00:19:14,240 --> 00:19:17,000 Speaker 1: of the contagion. If open ai fails and core we fails, 382 00:19:17,200 --> 00:19:20,359 Speaker 1: so too does course Scientific And I don't really fancy 383 00:19:20,440 --> 00:19:23,920 Speaker 1: Crusoe's chances either. But let's take a step back for 384 00:19:24,040 --> 00:19:26,840 Speaker 1: a moment. We've been going so hard, haven't we. I've 385 00:19:26,880 --> 00:19:30,280 Speaker 1: got a genuine question, just for the fact finders out there. 386 00:19:31,160 --> 00:19:34,520 Speaker 1: There's a Microsoft book. Open AI's computer's revenue now. Up 387 00:19:34,600 --> 00:19:37,879 Speaker 1: until fairly recently, Microsoft has been the entire infrastructure backing 388 00:19:38,000 --> 00:19:40,960 Speaker 1: open ai, but recently to free open ai up to 389 00:19:41,119 --> 00:19:43,480 Speaker 1: work with Oracle and see other people, released it from 390 00:19:43,480 --> 00:19:46,920 Speaker 1: its exclusive cloud compute deal. Nevertheless, put the information, open 391 00:19:47,000 --> 00:19:49,639 Speaker 1: ai still intends to spend thirteen billion dollars on compute 392 00:19:49,680 --> 00:19:53,080 Speaker 1: on Microsoft as there this year. What's confusing, however, is 393 00:19:53,119 --> 00:19:55,440 Speaker 1: whether any of this is booked as revenue for Microsoft. 394 00:19:55,560 --> 00:19:57,840 Speaker 1: Microsoft claimed earlier in the year that it surpassed thirteen 395 00:19:57,880 --> 00:20:00,920 Speaker 1: billion dollars in annual recurring revenue, by which it means 396 00:20:01,280 --> 00:20:03,840 Speaker 1: it's last month multiplied by twelve by the way, and 397 00:20:03,920 --> 00:20:06,720 Speaker 1: they said it was from ai. Open AI's compute costs 398 00:20:06,760 --> 00:20:09,240 Speaker 1: in twenty twenty four or five billion dollars, and that's 399 00:20:09,240 --> 00:20:12,119 Speaker 1: at are discounted as your rate, which on an anualized 400 00:20:12,119 --> 00:20:15,040 Speaker 1: basis will be about four hundred and sixteen million dollars 401 00:20:15,119 --> 00:20:18,560 Speaker 1: in revenue a month for Microsoft. It isn't, however, clear 402 00:20:18,600 --> 00:20:21,280 Speaker 1: whether Microsoft counts open ai is computer's money, which is 403 00:20:22,200 --> 00:20:25,120 Speaker 1: really fucking weird. You'd think with all this money they're 404 00:20:25,160 --> 00:20:28,440 Speaker 1: making from this company, they'd be saying there was money 405 00:20:28,520 --> 00:20:32,919 Speaker 1: coming in. It's peculiar. I've yet to find a real answer. Now. 406 00:20:32,960 --> 00:20:36,119 Speaker 1: Microsoft sernings do not include an artificial intelligence section. No, 407 00:20:36,200 --> 00:20:39,480 Speaker 1: They're made up of three separate segments, Productivity and Business Processes, 408 00:20:39,520 --> 00:20:42,520 Speaker 1: which includes things like LinkedIn, Microsoft three sixty five and 409 00:20:42,600 --> 00:20:46,080 Speaker 1: so on. More Personal Computing, which includes Windows and gaming products, 410 00:20:46,440 --> 00:20:49,520 Speaker 1: and then Intelligent Cloud including server products and cloud services 411 00:20:49,560 --> 00:20:52,600 Speaker 1: like A zero, which is likely where open AI's computers included, 412 00:20:52,680 --> 00:20:56,000 Speaker 1: and where Microsoft book the revenue from selling access to 413 00:20:56,119 --> 00:21:00,959 Speaker 1: open AI's models but not open AI's compute question. As 414 00:21:00,960 --> 00:21:03,800 Speaker 1: a result, it's hard to say specifically where open AI's 415 00:21:03,840 --> 00:21:07,040 Speaker 1: revenue might sit. Even guessing intelligent Cloud might not be right, 416 00:21:07,359 --> 00:21:10,479 Speaker 1: but based on an analysis of Microsoft's Intelligent Cloud segment 417 00:21:10,560 --> 00:21:13,040 Speaker 1: from financial year twenty twenty three Q one through its 418 00:21:13,080 --> 00:21:16,120 Speaker 1: most recent earnings, and there was a spike in revenue 419 00:21:16,240 --> 00:21:19,600 Speaker 1: from twenty three Q one to twenty four Q one. 420 00:21:20,400 --> 00:21:23,280 Speaker 1: In financial year Q one, which ended on September thirtieth, 421 00:21:23,320 --> 00:21:25,680 Speaker 1: twenty twenty two, a month before Chat GPT's launch, the 422 00:21:25,680 --> 00:21:29,160 Speaker 1: segment made twenty point three billion dollars. The following year, 423 00:21:29,640 --> 00:21:32,440 Speaker 1: in FY twenty four Q one, it made twenty four 424 00:21:32,480 --> 00:21:36,040 Speaker 1: point three billion dollars, a nineteen point seven percent year 425 00:21:36,119 --> 00:21:38,960 Speaker 1: of a year growth, or roughly four billion dollars. This 426 00:21:39,080 --> 00:21:41,840 Speaker 1: could represent the massive increase in training and inference costs 427 00:21:41,840 --> 00:21:45,160 Speaker 1: associate with hosting Chat GPT, and they peaked at twenty 428 00:21:45,240 --> 00:21:47,720 Speaker 1: eight point five billion dollars in revenue in the financial 429 00:21:47,800 --> 00:21:51,400 Speaker 1: year twenty four Q four, before dropping dramatically to twenty 430 00:21:51,440 --> 00:21:54,159 Speaker 1: four point one billion dollars in financial year twenty five 431 00:21:54,280 --> 00:21:56,640 Speaker 1: Q one, and raising a little twenty five point five 432 00:21:56,720 --> 00:22:00,280 Speaker 1: billion dollars in financial year twenty five Q two. I'm 433 00:22:00,320 --> 00:22:03,320 Speaker 1: so sorry none of this is easy to read. This 434 00:22:03,520 --> 00:22:06,640 Speaker 1: is a plausible explanation. Open ai spent twenty twenty three 435 00:22:06,720 --> 00:22:09,280 Speaker 1: training its GPT four to Roho model before transitioning to 436 00:22:09,359 --> 00:22:12,280 Speaker 1: its massive, expensive iryme model, which would eventually become GPT 437 00:22:12,400 --> 00:22:16,159 Speaker 1: four point five, as well as its video generating model sourer. 438 00:22:16,520 --> 00:22:19,240 Speaker 1: According to the Wall Street Journal, training GPT four point 439 00:22:19,280 --> 00:22:22,120 Speaker 1: five involved at least one training run, costing around half 440 00:22:22,160 --> 00:22:25,520 Speaker 1: a billion dollars in compute costs alone. These are huge sums, 441 00:22:25,600 --> 00:22:28,080 Speaker 1: but it's worth noting a couple of things. First, Microsoft 442 00:22:28,160 --> 00:22:30,960 Speaker 1: licenses open AI's models the third parties, so some of 443 00:22:31,040 --> 00:22:34,199 Speaker 1: this revenue could be from other companies using GPT on Azure. 444 00:22:34,640 --> 00:22:37,480 Speaker 1: We've seen lots of companies launch Ai products, and not 445 00:22:37,720 --> 00:22:41,440 Speaker 1: all of them are based on l lambs mrieing things. Further, 446 00:22:41,560 --> 00:22:44,439 Speaker 1: Microsoft provides open ai access to a zerr cloud services 447 00:22:44,520 --> 00:22:46,600 Speaker 1: at a discounted rate, as I've mentioned in the past, 448 00:22:47,680 --> 00:22:50,280 Speaker 1: and so there's a giant question mark over open AI's 449 00:22:50,320 --> 00:22:53,440 Speaker 1: actual contribution to the various spikes in revenue for Microsoft's 450 00:22:53,440 --> 00:22:56,080 Speaker 1: Intelligent cloud segment, or whether other third parties played a 451 00:22:56,119 --> 00:22:59,880 Speaker 1: significant role. Furthermore, Microsoft's investment in open Ai isn't entirely 452 00:23:00,160 --> 00:23:03,040 Speaker 1: cold hard cash. Rather, it's provided the company with credits 453 00:23:03,080 --> 00:23:05,000 Speaker 1: to be redeemed on in see your services, kind of 454 00:23:05,080 --> 00:23:08,760 Speaker 1: like Chucky cheese tokens. I'm not entirely sure how this 455 00:23:08,840 --> 00:23:11,080 Speaker 1: would be represented in accounting terms, and if anyone can 456 00:23:11,119 --> 00:23:13,760 Speaker 1: shed any light on this, please get in touch. Would 457 00:23:13,760 --> 00:23:16,560 Speaker 1: it be noted as revenue or something else? Open Ai 458 00:23:16,720 --> 00:23:20,040 Speaker 1: isn't paying Microsoft or are they? Are they doing the 459 00:23:20,200 --> 00:23:22,840 Speaker 1: tech equivalent of redeeming air miles or have they spent 460 00:23:22,920 --> 00:23:25,280 Speaker 1: a gift card of us? You're it really isn't obvious, 461 00:23:25,359 --> 00:23:28,440 Speaker 1: and Microsoft is doing some accounting bullshit here and not 462 00:23:28,600 --> 00:23:32,800 Speaker 1: suggesting impropriety, not suggesting anything illegal. I'm just saying it's 463 00:23:32,880 --> 00:23:36,320 Speaker 1: insane that they have this company spending billions of dollars 464 00:23:36,720 --> 00:23:41,960 Speaker 1: theoretically on their services and it's just nowhere. Additionally, while 465 00:23:42,000 --> 00:23:44,639 Speaker 1: equity is often treated as income for tax purposes, as 466 00:23:44,760 --> 00:23:47,560 Speaker 1: is the case when an employee receives RSUs as part 467 00:23:47,600 --> 00:23:50,840 Speaker 1: of their compensation package, under the existing open Ai structure, 468 00:23:50,920 --> 00:23:53,399 Speaker 1: Microsoft isn't actually a shareholder, but rather the owner of 469 00:23:53,480 --> 00:23:57,360 Speaker 1: profit sharing units. This is a distinction worth noting. These 470 00:23:57,440 --> 00:24:00,480 Speaker 1: profit sharing units are treated as analogous to equity, or 471 00:24:00,480 --> 00:24:03,359 Speaker 1: at least in terms of open AI's ability to raise capital, 472 00:24:03,520 --> 00:24:05,640 Speaker 1: but in practice they aren't the same thing. They don't 473 00:24:05,640 --> 00:24:08,200 Speaker 1: represent ownership in a company as directly as, for example, 474 00:24:08,320 --> 00:24:10,920 Speaker 1: a normal share would. They lack the liquidity of a 475 00:24:10,960 --> 00:24:14,359 Speaker 1: share in the upside they provide, namely dividends is purely theoretical. 476 00:24:14,760 --> 00:24:18,000 Speaker 1: Another key difference. When a company goes bankrupt and enters liquidation, 477 00:24:18,119 --> 00:24:22,000 Speaker 1: shareholders can potentially receive a share of the proceeds after creditors, employees, 478 00:24:22,040 --> 00:24:24,480 Speaker 1: and so on are paid. Well, that often doesn't happen 479 00:24:24,560 --> 00:24:28,520 Speaker 1: as is, as in the liabilities generally, they can exceed 480 00:24:28,600 --> 00:24:30,720 Speaker 1: the assets of the company. In many cases it's at 481 00:24:30,840 --> 00:24:35,119 Speaker 1: least theoretically possible, given that profit sharing units aren't actual 482 00:24:35,200 --> 00:24:39,520 Speaker 1: salaries or shares. Where does that leave Microsoft? This stuff 483 00:24:39,600 --> 00:24:41,200 Speaker 1: is confusing, and I'm not ashamed to say that I 484 00:24:41,440 --> 00:24:44,560 Speaker 1: just fucked up a word and that complicated accounting questions 485 00:24:44,640 --> 00:24:46,840 Speaker 1: like these are far beyond my understanding. If anyone can 486 00:24:46,920 --> 00:24:48,879 Speaker 1: shed some light, drop me an email, buzz me on 487 00:24:48,920 --> 00:24:50,920 Speaker 1: Twitter or blue sky, hit me up on clerk or 488 00:24:51,000 --> 00:24:54,639 Speaker 1: gorp or post on the better offline subware. Someone might 489 00:24:54,720 --> 00:24:57,480 Speaker 1: take your wallet though. Anyway, back on track, I think 490 00:24:57,480 --> 00:25:00,680 Speaker 1: it's worth understanding the scale of the open air vortex 491 00:25:00,760 --> 00:25:02,960 Speaker 1: and how it's distorting the tech investment market and why, 492 00:25:03,520 --> 00:25:07,320 Speaker 1: even without having failed, it represents the systemic risk. Without 493 00:25:07,359 --> 00:25:10,160 Speaker 1: open ai, and American startup investment is flat, and even 494 00:25:10,240 --> 00:25:13,640 Speaker 1: with it, less startups are receiving investment. Crunch based News 495 00:25:13,680 --> 00:25:16,480 Speaker 1: reported in early April the North American startup investments spiked 496 00:25:16,520 --> 00:25:18,640 Speaker 1: in Q one due to open Ai, hitting eighty two 497 00:25:18,680 --> 00:25:22,959 Speaker 1: billion dollars. Great, right, sounds great? This statement sadly has 498 00:25:23,000 --> 00:25:26,240 Speaker 1: a darker undertone. American star up investment was actually like 499 00:25:26,359 --> 00:25:28,520 Speaker 1: forty two billion in Q one twenty twenty five when 500 00:25:28,520 --> 00:25:30,680 Speaker 1: you remove the deal, which is appropriate because none of 501 00:25:30,680 --> 00:25:33,159 Speaker 1: the money is actually received by open Ai yet and 502 00:25:33,240 --> 00:25:34,960 Speaker 1: at best only ten billion dollars if it will be 503 00:25:35,040 --> 00:25:38,600 Speaker 1: received before Sember twenty twenty five. This quarter also included 504 00:25:38,640 --> 00:25:42,480 Speaker 1: a three point five billion dollar investment in Wormlight competitors 505 00:25:42,520 --> 00:25:45,720 Speaker 1: Aroundthropic run by Warrio Amma Day, making the appropriate number 506 00:25:45,760 --> 00:25:49,600 Speaker 1: of paltry thirty nine point five billion dollars. Now, this 507 00:25:49,720 --> 00:25:52,200 Speaker 1: is still an improvement, though a marginal one over the 508 00:25:52,240 --> 00:25:54,359 Speaker 1: thirty seven point five billion dollars raised in Q one 509 00:25:54,400 --> 00:25:58,359 Speaker 1: twenty twenty four. Nevertheless, crunch based news also has a far, 510 00:25:58,480 --> 00:26:01,879 Speaker 1: far darker story. Your volume women in American startups has 511 00:26:01,920 --> 00:26:05,080 Speaker 1: begun to collapse, trending downward almost every quarter or deal 512 00:26:05,200 --> 00:26:08,479 Speaker 1: volume isn't the direct result of open AI's financial condition. 513 00:26:08,640 --> 00:26:10,959 Speaker 1: The so called revolution created by open ai and other 514 00:26:11,000 --> 00:26:13,960 Speaker 1: generative AI companies technology appears to be petering out, and 515 00:26:14,040 --> 00:26:16,480 Speaker 1: the contagion is starting to impact the wider tech sector. 516 00:26:17,119 --> 00:26:19,600 Speaker 1: It's important to understand how bleak things are the future 517 00:26:19,640 --> 00:26:22,280 Speaker 1: of generative AI, wrest and open ai, and open AI's 518 00:26:22,320 --> 00:26:25,840 Speaker 1: future rests on near and possible financial requirements. I've done 519 00:26:25,880 --> 00:26:28,440 Speaker 1: my best to make this argument is in as objective 520 00:26:28,440 --> 00:26:30,680 Speaker 1: a tone as possible, regardless of my feelings about the 521 00:26:30,720 --> 00:26:35,080 Speaker 1: bubble and its associated boosters open Ai. As I've said 522 00:26:35,160 --> 00:26:39,800 Speaker 1: before and argued countless times in interviews and podcasts and newsletters, 523 00:26:40,560 --> 00:26:43,960 Speaker 1: it's effectively the entire generative AI industry, with its nearest 524 00:26:44,000 --> 00:26:46,600 Speaker 1: competitor being less than five percent of its five hundred 525 00:26:46,720 --> 00:26:52,000 Speaker 1: million weekly active users Anthropic, Google, Microsoft XAI. They're all 526 00:26:52,119 --> 00:26:55,000 Speaker 1: rounding errors in the grand scheme of things. But open 527 00:26:55,040 --> 00:26:58,400 Speaker 1: AI's future is dependent and this is not an opinion. 528 00:26:58,480 --> 00:27:02,280 Speaker 1: This is an objective fact on effectively infinite resources in 529 00:27:02,359 --> 00:27:06,160 Speaker 1: many forms. Let's start with the financial resources. If open 530 00:27:06,240 --> 00:27:09,360 Speaker 1: ai required forty billion dollars to continue operations this year, 531 00:27:09,520 --> 00:27:12,040 Speaker 1: it's reasonable to believe it will need at least another 532 00:27:12,119 --> 00:27:15,000 Speaker 1: forty billion dollars next year, and based on its internal projections, 533 00:27:15,040 --> 00:27:18,119 Speaker 1: will need at least forty billion dollars every single year 534 00:27:18,200 --> 00:27:21,000 Speaker 1: until twenty thirty, when it claims somehow it will be profitable. 535 00:27:21,080 --> 00:27:23,399 Speaker 1: And I quote the information with the completion of the 536 00:27:23,440 --> 00:27:28,760 Speaker 1: Stargate Data Center project, you may be wondering, how's that possible? Ed? 537 00:27:29,720 --> 00:27:32,359 Speaker 1: How you think the information wrote that down? Fuck no, 538 00:27:32,760 --> 00:27:35,840 Speaker 1: just go Lesson's too busy humiliating people. She let go 539 00:27:36,400 --> 00:27:39,200 Speaker 1: by name on Twitter, Ess Golesson. I like the information. 540 00:27:39,280 --> 00:27:41,160 Speaker 1: I think you're a fucking asshole for how you treated 541 00:27:41,240 --> 00:27:43,280 Speaker 1: your people. Say it on my podcast, I say it 542 00:27:43,320 --> 00:27:46,280 Speaker 1: on Twitter. Anyway. Let's keep talking about some of these 543 00:27:46,320 --> 00:27:50,080 Speaker 1: resources that open ai is dealing with, specifically the compute 544 00:27:50,119 --> 00:27:54,040 Speaker 1: resources and expansion. Open ai requires more compute resources than 545 00:27:54,080 --> 00:27:56,119 Speaker 1: anyone has ever needed, and will continue to do so 546 00:27:56,240 --> 00:27:59,879 Speaker 1: in perpetuity. Building these resources is now dependent on two partners, 547 00:28:00,320 --> 00:28:03,119 Speaker 1: Course Scientific and Crusoe. Though I've never built a data center, 548 00:28:03,280 --> 00:28:06,080 Speaker 1: as Microsoft has materially pulled back on data center development 549 00:28:06,200 --> 00:28:10,200 Speaker 1: and has as aforementioned, pulled back on two gigawads of 550 00:28:10,359 --> 00:28:13,159 Speaker 1: data centers, slowed or paused. Of course, some of its 551 00:28:13,200 --> 00:28:17,400 Speaker 1: early stage center products too, with TD Cohen's recent analyst 552 00:28:17,520 --> 00:28:20,000 Speaker 1: reports saying that data center pullbacks were and I quote 553 00:28:20,000 --> 00:28:23,040 Speaker 1: them March twenty six, twenty twenty five, data center channel 554 00:28:23,119 --> 00:28:26,400 Speaker 1: checks letter because it's so good, driven by the decision 555 00:28:26,480 --> 00:28:30,320 Speaker 1: to not support incremental open Ai training workloads, that's the stuff. 556 00:28:31,080 --> 00:28:33,480 Speaker 1: In simpler terms, open ai needs more computer at a 557 00:28:33,480 --> 00:28:35,640 Speaker 1: time when it's lead backer, which has the most GPUs 558 00:28:35,680 --> 00:28:38,160 Speaker 1: in the world, has specifically walked away from building it. 559 00:28:39,120 --> 00:28:41,640 Speaker 1: Even in my most optimistic frame of mind, it isn't 560 00:28:41,680 --> 00:28:44,240 Speaker 1: realistic to believe that Cruso or Core Scientific can build 561 00:28:44,280 --> 00:28:47,800 Speaker 1: the data centers necessary for open AI's expansion, even if 562 00:28:47,840 --> 00:28:49,920 Speaker 1: soft Bank and open Ai had the money to invest 563 00:28:49,960 --> 00:28:52,880 Speaker 1: in Stargate today, which they do not. Dollars do not 564 00:28:53,080 --> 00:28:56,200 Speaker 1: change the fabric of reality. Data Centers take time to build, 565 00:28:56,280 --> 00:28:59,480 Speaker 1: requiring concrete would steal in other materials to be manufactured 566 00:28:59,520 --> 00:29:01,840 Speaker 1: and placed, and that's after the permitting required to get 567 00:29:01,880 --> 00:29:04,840 Speaker 1: these deals done. Even if that succeeds, getting the power 568 00:29:04,920 --> 00:29:07,280 Speaker 1: necessary is a challenge unto itself, to the point that 569 00:29:07,360 --> 00:29:10,840 Speaker 1: even Oracle, an established and storied cloud compute company run 570 00:29:10,920 --> 00:29:12,960 Speaker 1: by a very evil man at one point to quote 571 00:29:13,000 --> 00:29:16,200 Speaker 1: the Information, has less experience than its larger arrivals in 572 00:29:16,280 --> 00:29:18,480 Speaker 1: dealing with the utilities, to secure power and working with 573 00:29:18,520 --> 00:29:22,080 Speaker 1: powerful and demanding cloud customers whose plans change frequently. A 574 00:29:22,160 --> 00:29:25,440 Speaker 1: partner like Cruso will Core Scientific simply doesn't have the muscle, memory, 575 00:29:25,480 --> 00:29:27,640 Speaker 1: or domainer expertise that Microsoft has when it comes to 576 00:29:27,680 --> 00:29:30,400 Speaker 1: building and operating data centers. As a result, it's hard 577 00:29:30,440 --> 00:29:33,320 Speaker 1: to imagine, even in the best case scenario, that they 578 00:29:33,360 --> 00:29:35,280 Speaker 1: are able to match the hunger for compute the open 579 00:29:35,360 --> 00:29:38,040 Speaker 1: Ai has now. I want to be clear, I believe 580 00:29:38,080 --> 00:29:40,800 Speaker 1: open ai will still continue to use Microsoft's compute and 581 00:29:40,880 --> 00:29:44,920 Speaker 1: even expand further into whatever remaining compute Microsoft may have. However, 582 00:29:45,000 --> 00:29:46,800 Speaker 1: there is now a hard limit on how much of 583 00:29:46,880 --> 00:29:50,040 Speaker 1: that there's going to be, both literally and what's physically available, 584 00:29:50,240 --> 00:29:52,960 Speaker 1: and in what Microsoft itself will actually allow open ai 585 00:29:53,120 --> 00:29:56,440 Speaker 1: to use, especially given how unprofitable GPU compute seems to 586 00:29:56,520 --> 00:29:59,560 Speaker 1: be based on how every single company that isn't in 587 00:29:59,680 --> 00:30:17,480 Speaker 1: vidio lose his money running them. But really, and we're 588 00:30:17,520 --> 00:30:20,520 Speaker 1: coming to the end of this, which leads to a question, 589 00:30:21,960 --> 00:30:26,120 Speaker 1: how does all of this end? Last week, a truly 590 00:30:26,160 --> 00:30:29,040 Speaker 1: offensive piece of fan fiction framed as a report called 591 00:30:29,080 --> 00:30:32,280 Speaker 1: AI twenty twenty seven went viral, garnering press with the 592 00:30:32,360 --> 00:30:36,440 Speaker 1: Duiskesh podcast and gormles childlike wonder from Dope New York 593 00:30:36,520 --> 00:30:39,640 Speaker 1: Times reporter Kevin Rus and reporter I think is a 594 00:30:39,720 --> 00:30:44,400 Speaker 1: fucking stretch. Its predictions vaguely suggest a theoretical company called 595 00:30:44,480 --> 00:30:47,720 Speaker 1: open Brain will invent a self teaching agent of some sort. 596 00:30:48,080 --> 00:30:50,520 Speaker 1: It's total bullshit, but it captured the handsome minds of 597 00:30:50,560 --> 00:30:53,400 Speaker 1: AI boosters and other people without object permanence because it 598 00:30:53,480 --> 00:30:56,920 Speaker 1: vaguely suggests that somehow, large language models and their associated 599 00:30:56,960 --> 00:31:01,480 Speaker 1: technology will become something entirely different. Like making predictions like 600 00:31:01,600 --> 00:31:04,440 Speaker 1: these because the future, especially in our current political climate, 601 00:31:04,640 --> 00:31:07,320 Speaker 1: is utter chaos. But I will say that I do 602 00:31:07,480 --> 00:31:10,040 Speaker 1: not see, and I say this with complete objectivity, how 603 00:31:10,160 --> 00:31:13,960 Speaker 1: any of this bullshit continues. I want to be extremely 604 00:31:14,160 --> 00:31:16,280 Speaker 1: blunt with the following points, as I feel like both 605 00:31:16,320 --> 00:31:19,560 Speaker 1: members of the media and tech analysts have categorically failed 606 00:31:19,600 --> 00:31:22,640 Speaker 1: to express how ridiculous things have become. I will be 607 00:31:22,800 --> 00:31:25,960 Speaker 1: repeating myself, but it's fucking necessary, as I need you 608 00:31:26,080 --> 00:31:29,880 Speaker 1: to understand how untenable things are. Soft Bank is putting 609 00:31:29,920 --> 00:31:33,120 Speaker 1: itself in dire straits simply to fund open Ai once 610 00:31:33,520 --> 00:31:36,200 Speaker 1: this deal threatens its credit rating, with soft Bank having 611 00:31:36,280 --> 00:31:38,200 Speaker 1: to take on what will be multiple loans to fund 612 00:31:38,320 --> 00:31:40,800 Speaker 1: this forty billion dollar round, and open Ai will need 613 00:31:40,840 --> 00:31:44,080 Speaker 1: at least another forty billion dollars a year later. This 614 00:31:44,320 --> 00:31:46,880 Speaker 1: is before you consider the other nineteen billion dollars that 615 00:31:46,960 --> 00:31:48,960 Speaker 1: soft Bank has agreed to contribute to the data center 616 00:31:49,000 --> 00:31:53,440 Speaker 1: project with Stargate money it does not currently have available. Now. 617 00:31:53,520 --> 00:31:56,160 Speaker 1: Open Ai has promised nineteen billion dollars to the Stargate 618 00:31:56,280 --> 00:31:59,320 Speaker 1: Data Center project two and again they do not have it, 619 00:31:59,440 --> 00:32:01,320 Speaker 1: and they need soft Bank to give it to them. 620 00:32:02,120 --> 00:32:05,040 Speaker 1: And again I've said it, and I'll say it again. 621 00:32:05,560 --> 00:32:09,680 Speaker 1: Neither of these companies have the money. The money is 622 00:32:09,880 --> 00:32:13,600 Speaker 1: not there, and open Ai needs Stargate to get built 623 00:32:13,640 --> 00:32:17,320 Speaker 1: to grow much further. I see no way in which 624 00:32:17,360 --> 00:32:19,760 Speaker 1: open ai can continue to raise money at this rate, 625 00:32:19,920 --> 00:32:23,640 Speaker 1: even if open Ai somehow actually receives the forty billion 626 00:32:23,720 --> 00:32:26,560 Speaker 1: dollars it's been promised, which will require it to become 627 00:32:26,600 --> 00:32:28,520 Speaker 1: a for profit entity, which I don't think it can 628 00:32:28,640 --> 00:32:31,960 Speaker 1: fucking do. While it could theoretically stretch that forty billion 629 00:32:32,000 --> 00:32:34,400 Speaker 1: dollars to the last multiple years, projections say it will 630 00:32:34,400 --> 00:32:36,600 Speaker 1: burn three hundred and twenty billion dollars in the next 631 00:32:36,760 --> 00:32:40,240 Speaker 1: five years, or more likely, I can't see a realistic 632 00:32:40,280 --> 00:32:42,520 Speaker 1: way in which open ai gets the resources it needs 633 00:32:42,600 --> 00:32:46,280 Speaker 1: to survive. It will need this insane streak of good fortune, 634 00:32:46,320 --> 00:32:47,800 Speaker 1: the kind of which you only really hear about in 635 00:32:48,080 --> 00:32:51,320 Speaker 1: Greek poems or JoJo's Bizarre Adventure. You know, the more 636 00:32:51,400 --> 00:32:54,600 Speaker 1: cultured choice, but let's go through them. Somehow soft Bank 637 00:32:54,600 --> 00:32:57,840 Speaker 1: gets the resources and loses the constraints required to bankroll 638 00:32:57,880 --> 00:33:01,320 Speaker 1: this company forever. The world's wealthiest entities, those sovereign wealth 639 00:33:01,360 --> 00:33:03,440 Speaker 1: funds mentioned in the last episode, sounds and so on, 640 00:33:03,680 --> 00:33:06,239 Speaker 1: they pick up the slack until open Ai receive they 641 00:33:06,960 --> 00:33:12,400 Speaker 1: reach profitability, which is a huge assumption. It's also assuming 642 00:33:12,440 --> 00:33:15,240 Speaker 1: that open ai will have enough of these megawealthy benefactors 643 00:33:15,280 --> 00:33:16,880 Speaker 1: to provide it with the three hundred and twenty billion 644 00:33:16,880 --> 00:33:19,560 Speaker 1: dollars they need to reach profitability, which it won't. There'll 645 00:33:19,600 --> 00:33:22,520 Speaker 1: also need Cruso and Core Scientific to turn out to 646 00:33:22,560 --> 00:33:25,440 Speaker 1: be really good at building AI infrastructure, which they've never 647 00:33:25,760 --> 00:33:30,280 Speaker 1: done before, which is that's very possible, I'm sure. And 648 00:33:30,360 --> 00:33:33,240 Speaker 1: then Microsoft will then walk back its walk back on 649 00:33:33,400 --> 00:33:36,840 Speaker 1: building UAI infrastructure and recommit to tens of billions of 650 00:33:36,840 --> 00:33:39,920 Speaker 1: dollars of CAPEX, specifically on AI data centers, and also 651 00:33:40,000 --> 00:33:42,120 Speaker 1: will give it to open Ai. And then, of course 652 00:33:42,160 --> 00:33:45,120 Speaker 1: Stargate's construction happens faster than expected and there are no 653 00:33:45,240 --> 00:33:48,520 Speaker 1: supply chain issues in terms of labor, building materials, GPUs 654 00:33:48,560 --> 00:33:51,640 Speaker 1: and so on. Now I don't know, I haven't checked 655 00:33:51,640 --> 00:33:53,400 Speaker 1: the news in the last three weeks, but is there 656 00:33:53,440 --> 00:33:56,360 Speaker 1: anything going on that might increase the costs of materials? 657 00:33:58,280 --> 00:34:02,000 Speaker 1: Probably not. Anyway, if those things happen, I'll eat quo. 658 00:34:02,840 --> 00:34:07,960 Speaker 1: I'm not particularly worried. In the present conditions. Open ai 659 00:34:08,120 --> 00:34:09,960 Speaker 1: is on course to run out of money or run 660 00:34:10,040 --> 00:34:13,439 Speaker 1: out of compute capacity, and it's unclear which will happen first. 661 00:34:13,960 --> 00:34:16,040 Speaker 1: But what is clear is it's time to wake up. 662 00:34:17,040 --> 00:34:19,840 Speaker 1: Even in a hysterical bubble where everybody is agreeing that 663 00:34:19,920 --> 00:34:23,680 Speaker 1: this is the future, open ai is currently requiring more 664 00:34:23,719 --> 00:34:30,880 Speaker 1: money in more compute than is reasonable to acquire. Nobody nowhere, ever, anywhere, 665 00:34:31,120 --> 00:34:33,719 Speaker 1: has ever raised as much money as open ai needs to. 666 00:34:34,040 --> 00:34:36,680 Speaker 1: And based on the sheer amount of difficulty that soft 667 00:34:36,760 --> 00:34:39,840 Speaker 1: Bank is having raising the funds to meet the lower tranche, 668 00:34:39,960 --> 00:34:42,480 Speaker 1: the ten billion dollar one of its commitment, it may 669 00:34:42,560 --> 00:34:45,960 Speaker 1: not actually be possible for this company to continue, even 670 00:34:46,040 --> 00:34:50,320 Speaker 1: with the extremely preferential payment terms months long deferred payments 671 00:34:50,320 --> 00:34:53,319 Speaker 1: for example that open ai probably has. At some point 672 00:34:53,440 --> 00:34:56,640 Speaker 1: someone will need a dollar. I'll give Sam Ortman some 673 00:34:56,719 --> 00:34:59,560 Speaker 1: fucking credit. He's found many partners the shoulder the burden 674 00:34:59,600 --> 00:35:02,720 Speaker 1: of the rock economics of open Ai. With Microsoft, Oracle, 675 00:35:02,760 --> 00:35:05,320 Speaker 1: Crusoe and Core We've handling the upfront costs of building 676 00:35:05,360 --> 00:35:08,759 Speaker 1: the infrastructure, and SoftBank finding the investors for its monstrous 677 00:35:08,840 --> 00:35:12,000 Speaker 1: stupid round, and the tech media mostly handling marketing for him, 678 00:35:12,000 --> 00:35:16,359 Speaker 1: which is really nice, great job everybody. He is. However, overleveraged. 679 00:35:16,800 --> 00:35:19,359 Speaker 1: Open Ai has never been forced to stand on its 680 00:35:19,400 --> 00:35:22,360 Speaker 1: own two feet or focus on efficiency, and I believe 681 00:35:22,400 --> 00:35:25,960 Speaker 1: the constant enabling of this ugly nonsensical burn rate has 682 00:35:26,000 --> 00:35:29,240 Speaker 1: doomed this company. Open Ai has acted like it'll always 683 00:35:29,280 --> 00:35:33,600 Speaker 1: have more money. In compute, And that's kind of because 684 00:35:33,880 --> 00:35:36,080 Speaker 1: everyone's acted as that would be the case. No one's 685 00:35:36,120 --> 00:35:38,600 Speaker 1: really called sam Altman out on his bullshit. There are 686 00:35:38,680 --> 00:35:41,200 Speaker 1: some people, but really no one in the mainstream media 687 00:35:41,239 --> 00:35:45,759 Speaker 1: has bothered. Really, Sam Altman has been enabled. Open Ai, 688 00:35:45,920 --> 00:35:48,120 Speaker 1: by the way, cannot just make things cheaper at this point, 689 00:35:48,160 --> 00:35:50,200 Speaker 1: because the money has always been there to make things 690 00:35:50,239 --> 00:35:52,960 Speaker 1: more expensive, as has the compute to make larger and 691 00:35:53,120 --> 00:35:55,680 Speaker 1: larger language models that burn billions of dollars a year. 692 00:35:56,840 --> 00:35:59,200 Speaker 1: This company is not built to reduce its footprint in 693 00:35:59,320 --> 00:36:01,400 Speaker 1: any way, nor is it built for a future in 694 00:36:01,480 --> 00:36:05,000 Speaker 1: which it wouldn't have access to infinite resources. Worse still, 695 00:36:05,080 --> 00:36:07,120 Speaker 1: investors in the media have run cover for the fact 696 00:36:07,160 --> 00:36:09,120 Speaker 1: that these models don't really do much more than they 697 00:36:09,160 --> 00:36:11,600 Speaker 1: did a year ago, and for the overall diminishing returns 698 00:36:11,640 --> 00:36:14,759 Speaker 1: of large language models writ large. Now, I've had many 699 00:36:14,880 --> 00:36:17,759 Speaker 1: people attack my work about open ai, but none of them, 700 00:36:18,160 --> 00:36:21,960 Speaker 1: not one of them. Nobody has provided me any real 701 00:36:22,120 --> 00:36:25,080 Speaker 1: counterpoint to the underlying economic argument I've made since July 702 00:36:25,160 --> 00:36:28,560 Speaker 1: of last year, the open ai is unsustainable. Now this 703 00:36:28,719 --> 00:36:31,439 Speaker 1: is likely because there really isn't one other than open 704 00:36:31,520 --> 00:36:33,920 Speaker 1: ai will continue to raise more money than anybody has 705 00:36:33,960 --> 00:36:37,360 Speaker 1: ever raised in history imperpetuity and will somehow turn the 706 00:36:37,440 --> 00:36:40,680 Speaker 1: least profitable company of all time into a profitable company. 707 00:36:41,239 --> 00:36:44,160 Speaker 1: This is not a rational argument. It's a religious one. 708 00:36:44,440 --> 00:36:47,719 Speaker 1: It's a call for faith. And it's disgusting to see 709 00:36:48,400 --> 00:36:53,839 Speaker 1: well paid reporters with one hundred and fifty thousand subscribers 710 00:36:53,880 --> 00:36:57,239 Speaker 1: to the newsletters and a really shitty podcast with a 711 00:36:57,320 --> 00:37:00,880 Speaker 1: major news outlet constantly just ignore them share and I 712 00:37:00,960 --> 00:37:03,960 Speaker 1: see no greater payal horse of the apocalypse than Microsoft's 713 00:37:03,960 --> 00:37:07,080 Speaker 1: material pullback on data centers. Well, the argument might be 714 00:37:07,160 --> 00:37:09,760 Speaker 1: that Microsoft wants open ai to have an independent future. 715 00:37:09,800 --> 00:37:13,480 Speaker 1: That's fucking laughable when you consider Microsoft's deeply monopolistic tendencies, 716 00:37:13,680 --> 00:37:16,480 Speaker 1: and for that matter, it owns a massive proportion of 717 00:37:16,600 --> 00:37:20,240 Speaker 1: open AI's pseudoequity. At one point, Microsoft's portion was valued 718 00:37:20,239 --> 00:37:23,160 Speaker 1: at forty nine percent, and while additional fundraising has likely 719 00:37:23,239 --> 00:37:26,560 Speaker 1: diluted Microsoft's steak, it still owns a massive portion of 720 00:37:26,640 --> 00:37:30,359 Speaker 1: what is, at the very least, if you believe any 721 00:37:30,400 --> 00:37:33,560 Speaker 1: of this nonsense, the most valuable private startup of all time. 722 00:37:34,080 --> 00:37:36,880 Speaker 1: And we're supposed to believe that Microsoft's pullback, which limits 723 00:37:36,920 --> 00:37:39,560 Speaker 1: open AI's access to infrastructure it needs to train its 724 00:37:39,640 --> 00:37:42,000 Speaker 1: and run its models, and thus is mentioned represents an 725 00:37:42,040 --> 00:37:44,200 Speaker 1: existential threat to the company. You meant to believe that 726 00:37:44,320 --> 00:37:46,600 Speaker 1: this is because of some sort of paternal desire to 727 00:37:46,680 --> 00:37:50,160 Speaker 1: see open ai leave childhood behind to spread its wings 728 00:37:50,200 --> 00:37:55,040 Speaker 1: and enter the real world. Are you fucking stupid? Sorry? 729 00:37:55,080 --> 00:37:58,960 Speaker 1: I shouldn't be calling people stupid. I shouldn't. I really shouldn't. 730 00:37:59,480 --> 00:38:02,200 Speaker 1: But I am more likely Microsoft got would have needed 731 00:38:02,239 --> 00:38:04,080 Speaker 1: out of open Ai, which has reached the limit of 732 00:38:04,080 --> 00:38:06,680 Speaker 1: the models that can develop, and which Microsoft, by the way, 733 00:38:06,719 --> 00:38:09,120 Speaker 1: already owns the ip of due to their twenty nineteen 734 00:38:09,160 --> 00:38:11,799 Speaker 1: funning round. There's probably no reason for Microsoft to make 735 00:38:11,800 --> 00:38:14,319 Speaker 1: any further significant investments other than just kind of throwing 736 00:38:14,320 --> 00:38:16,440 Speaker 1: a little cash in there, and then I imagine some 737 00:38:16,560 --> 00:38:19,920 Speaker 1: sort of tax dodge. I'm just guessing. It's also important 738 00:38:19,920 --> 00:38:22,680 Speaker 1: to note that absolutely nobody other than Nvidia is making 739 00:38:22,719 --> 00:38:26,000 Speaker 1: any money from generative AI. Core Weave loses billions of dollars, 740 00:38:26,080 --> 00:38:29,120 Speaker 1: open Ai loses billions of dollars, Anthropic loses billions of dollars. 741 00:38:29,120 --> 00:38:31,760 Speaker 1: And I can't find a single fucking company providing generative 742 00:38:31,800 --> 00:38:35,480 Speaker 1: AI powered software that's actually making a profit. The only 743 00:38:35,560 --> 00:38:38,480 Speaker 1: company is even close to doing so Are Consultancy is 744 00:38:38,520 --> 00:38:40,880 Speaker 1: providing services to drain and create data for models like 745 00:38:40,960 --> 00:38:44,839 Speaker 1: Churing and Scale AI, and Scale isn't even fucking profitable now. 746 00:38:44,880 --> 00:38:47,040 Speaker 1: The knock on effects of open AI's collapse will be 747 00:38:47,120 --> 00:38:50,040 Speaker 1: wide ranging. Neither core Weave nor Crusoe will have tenants 748 00:38:50,080 --> 00:38:53,320 Speaker 1: for their massive, unsustainable operations. An oracle will have nobody 749 00:38:53,360 --> 00:38:56,200 Speaker 1: to sell compute to because they've leased that thing for 750 00:38:56,320 --> 00:38:59,400 Speaker 1: fifteen fucking years, or one customer who else is going 751 00:38:59,480 --> 00:39:02,319 Speaker 1: to take that anyway. Cor we will likely collapse under 752 00:39:02,320 --> 00:39:04,319 Speaker 1: the weight of its abominable debt anyway, which will lead 753 00:39:04,360 --> 00:39:06,840 Speaker 1: to a six seven percent or more revenue drop for 754 00:39:06,920 --> 00:39:09,160 Speaker 1: in video or a time when revenue growth has already 755 00:39:09,200 --> 00:39:13,480 Speaker 1: begun to slow. On a philosophical level, too, open AI's 756 00:39:13,520 --> 00:39:16,600 Speaker 1: health is what keeps this industry alive. Open ai has 757 00:39:16,640 --> 00:39:19,600 Speaker 1: truly the only meaningful user base in generative AI, and 758 00:39:19,680 --> 00:39:22,440 Speaker 1: this entire hype cycle has been driven by its success. 759 00:39:22,719 --> 00:39:25,359 Speaker 1: Meaning any deterioration or collapse of open ai will tell 760 00:39:25,440 --> 00:39:27,279 Speaker 1: the market what I've been saying for over a year. 761 00:39:27,560 --> 00:39:29,640 Speaker 1: The generative AI is not the next hype of growth 762 00:39:29,719 --> 00:39:34,359 Speaker 1: market and it's underlying economics do not make sense. But look, 763 00:39:36,160 --> 00:39:38,440 Speaker 1: I'm not saying this to be a hater. I'm not 764 00:39:38,560 --> 00:39:42,520 Speaker 1: saying this to be right. This stuff has driven me insane, 765 00:39:43,120 --> 00:39:45,719 Speaker 1: but I'm not doing it to be a pundit, to 766 00:39:45,800 --> 00:39:48,040 Speaker 1: be a skeptic, to be a cynic, to be someone 767 00:39:48,120 --> 00:39:50,520 Speaker 1: that hates because I want to hate. And I hate 768 00:39:50,560 --> 00:39:53,480 Speaker 1: them not because I think people like me because I 769 00:39:53,520 --> 00:39:56,000 Speaker 1: hate them. I hate them because I have brainworms. I 770 00:39:56,160 --> 00:39:59,959 Speaker 1: have something wrong with me inside my brain that tells 771 00:40:00,360 --> 00:40:02,040 Speaker 1: I have to be like this, and I have to 772 00:40:02,080 --> 00:40:04,800 Speaker 1: look at these things and I have to try and 773 00:40:04,880 --> 00:40:07,640 Speaker 1: find what's going on, otherwise I will be driven mad, 774 00:40:08,520 --> 00:40:11,080 Speaker 1: which is why I'll say if something changes, if I'm 775 00:40:11,120 --> 00:40:15,480 Speaker 1: wrong somehow, I promise you I will tell you exactly how, 776 00:40:15,680 --> 00:40:18,880 Speaker 1: exactly why, and what mistakes I made to come to 777 00:40:18,960 --> 00:40:22,800 Speaker 1: the conclusions I have in this episode and the episodes before. 778 00:40:23,920 --> 00:40:25,680 Speaker 1: But I don't believe that my peers in the media 779 00:40:25,719 --> 00:40:28,520 Speaker 1: will do the same when this collapses. But I promise 780 00:40:28,640 --> 00:40:31,560 Speaker 1: you that they will be held accountable because all of 781 00:40:31,680 --> 00:40:36,400 Speaker 1: this abominable waste could have been avoided. Large language models 782 00:40:36,440 --> 00:40:39,560 Speaker 1: are not on their own the problem. The tools capable 783 00:40:39,600 --> 00:40:42,280 Speaker 1: of some outcomes, doing some things, But the problem, ultimately 784 00:40:42,320 --> 00:40:45,480 Speaker 1: are the extrapolations made about their abilities and the unnecessary 785 00:40:45,600 --> 00:40:48,480 Speaker 1: drive to make them larger, even if said largeness never 786 00:40:48,600 --> 00:40:52,640 Speaker 1: really amounted to much. Everything that I'm describing is the 787 00:40:52,719 --> 00:40:56,080 Speaker 1: result of a tech industry, including media and analysts, that 788 00:40:56,200 --> 00:40:59,400 Speaker 1: refuses to do business with reality, trafficking in the ideas 789 00:40:59,400 --> 00:41:02,760 Speaker 1: and ideolo, celebrating victories that have yet to take place, 790 00:41:02,800 --> 00:41:05,320 Speaker 1: applauding those who have yet to create the things that 791 00:41:05,400 --> 00:41:08,840 Speaker 1: they're talking about, cheering on men lying about what's possible 792 00:41:09,120 --> 00:41:11,240 Speaker 1: so that they can continue to burn billions of dollars 793 00:41:11,239 --> 00:41:14,560 Speaker 1: and increase their wealth and influence for barely any fucking reason. 794 00:41:15,920 --> 00:41:19,399 Speaker 1: I understand why others might not have said what I've said. 795 00:41:20,440 --> 00:41:23,239 Speaker 1: What I am describing is a systemic failure, one at 796 00:41:23,239 --> 00:41:26,200 Speaker 1: a scale here too unseen, one that has involved so 797 00:41:26,360 --> 00:41:29,960 Speaker 1: many rich and powerful and influential people agreeing to ignore reality, 798 00:41:30,760 --> 00:41:33,600 Speaker 1: and that'll have crushing impacts for the wider tech ecosystem 799 00:41:33,640 --> 00:41:45,520 Speaker 1: when it happens. Don't say I didn't warn you. Thank 800 00:41:45,560 --> 00:41:48,280 Speaker 1: you for listening to Better Offline. The editor and composer 801 00:41:48,320 --> 00:41:51,120 Speaker 1: of the Better Offline theme song is Matasowski. You can 802 00:41:51,200 --> 00:41:53,520 Speaker 1: check out more of his music and audio projects at 803 00:41:53,600 --> 00:41:57,040 Speaker 1: Mattasowski dot com m A T. T O S O 804 00:41:57,760 --> 00:42:01,520 Speaker 1: w Ski dot com. You can email me at easy 805 00:42:01,600 --> 00:42:04,200 Speaker 1: at better offline dot com or visit better offline dot 806 00:42:04,239 --> 00:42:06,880 Speaker 1: com to find more podcast links and of course, my newsletter. 807 00:42:07,280 --> 00:42:10,160 Speaker 1: I also really recommend you go to chat dot where'soead 808 00:42:10,200 --> 00:42:12,440 Speaker 1: dot at to visit the discord, and go to our 809 00:42:12,560 --> 00:42:15,800 Speaker 1: slash Better Offline to check out I'll Reddit. Thank you 810 00:42:15,920 --> 00:42:19,279 Speaker 1: so much for listening. Better Offline is a production of 811 00:42:19,360 --> 00:42:22,200 Speaker 1: cool Zone Media. For more from cool Zone Media, visit 812 00:42:22,280 --> 00:42:25,319 Speaker 1: our website cool Zonemedia dot com, or check us out 813 00:42:25,360 --> 00:42:28,359 Speaker 1: on the iHeartRadio app, Apple Podcasts, or wherever you get 814 00:42:28,400 --> 00:42:29,040 Speaker 1: your podcasts.