1 00:00:01,880 --> 00:00:05,480 Speaker 1: Al Zone Media, give me that total next shirt, a 2 00:00:05,480 --> 00:00:08,000 Speaker 1: tim pan apple, a German shepherd, a wristbandstand, and a 3 00:00:08,080 --> 00:00:10,680 Speaker 1: lurching red bird and more brave than the Turning Network. 4 00:00:10,760 --> 00:00:13,039 Speaker 1: This is your weekly Better Offline monologue, and I'm your 5 00:00:13,039 --> 00:00:24,000 Speaker 1: host ed zetron. Now, before we go any further, I 6 00:00:24,079 --> 00:00:27,280 Speaker 1: need your help. I look Better Offline is up for 7 00:00:27,320 --> 00:00:28,960 Speaker 1: a webby and I really need you to vote for 8 00:00:29,000 --> 00:00:31,040 Speaker 1: best episode in the business category. It's the man who 9 00:00:31,120 --> 00:00:34,360 Speaker 1: killed Google Search. It's Propagart Ragavan. Let's get him. I 10 00:00:34,440 --> 00:00:36,520 Speaker 1: realize it's a huge pain in the ass to sign 11 00:00:36,600 --> 00:00:38,720 Speaker 1: up for something and vote, but I've never won an 12 00:00:38,760 --> 00:00:41,160 Speaker 1: award in my life and I'd really appreciate it. Link 13 00:00:41,240 --> 00:00:42,960 Speaker 1: is going to be in the episode notes, and while 14 00:00:42,960 --> 00:00:46,000 Speaker 1: you're there, also vote for the wonderful Mollyconger's Weird Little Guys, 15 00:00:46,000 --> 00:00:48,440 Speaker 1: which I'll also have a link to. I know signing 16 00:00:48,520 --> 00:00:51,360 Speaker 1: up to stuff is annoying. I'm asking a lot from you, 17 00:00:51,440 --> 00:00:56,040 Speaker 1: but there you go. I'm doing it anyway. To the monologue, 18 00:00:56,280 --> 00:00:58,720 Speaker 1: I feel like we're approaching a choke point in the 19 00:00:58,720 --> 00:01:01,280 Speaker 1: whole General v Ai bubble, the culmination of over a 20 00:01:01,360 --> 00:01:04,119 Speaker 1: year of different narratives and pressures that I believe will 21 00:01:04,440 --> 00:01:07,839 Speaker 1: lead to an ultimate collapse. Last week, open Ai released 22 00:01:07,840 --> 00:01:10,880 Speaker 1: an image generator with GPT four to zero, which quickly 23 00:01:10,920 --> 00:01:13,440 Speaker 1: gained massive attention for its ability to create images in 24 00:01:13,480 --> 00:01:17,440 Speaker 1: the style of famed Japanese animation company Studio Ghibli. And 25 00:01:17,480 --> 00:01:19,600 Speaker 1: to be clear, I think these images are an abomination 26 00:01:19,680 --> 00:01:21,880 Speaker 1: and everyone involved in launching this tool has committed a 27 00:01:21,920 --> 00:01:27,800 Speaker 1: mortal sin anyway. Nevertheless, creating these disgusting, disgraceful images comes 28 00:01:27,800 --> 00:01:30,679 Speaker 1: at in incredibly high cost, and for the last week, 29 00:01:30,760 --> 00:01:33,479 Speaker 1: open Ai CEO Sam Ortman has been complaining about their 30 00:01:33,480 --> 00:01:36,960 Speaker 1: GPUs melting, leading to open ai having to limit free 31 00:01:37,080 --> 00:01:40,200 Speaker 1: users to only three image generations a day, along with 32 00:01:40,280 --> 00:01:43,400 Speaker 1: longer wait times than capacity issues with video generator Sora. 33 00:01:44,120 --> 00:01:46,960 Speaker 1: To make matters worse, Ortman also announced that and I 34 00:01:47,200 --> 00:01:50,080 Speaker 1: quote by the way, that users should expect new releases 35 00:01:50,120 --> 00:01:52,720 Speaker 1: from open Ai to be delayed, stuff to break, and 36 00:01:52,760 --> 00:01:55,200 Speaker 1: for services to sometimes be slow as we deal with 37 00:01:55,200 --> 00:02:00,800 Speaker 1: capacity challenges. This led me to ask a very simple question. 38 00:02:00,960 --> 00:02:03,480 Speaker 1: I think everybody in the tech media really should be asking, 39 00:02:04,040 --> 00:02:08,600 Speaker 1: why can't Sam Waltman ask Microsoft for more GPUs. The 40 00:02:08,639 --> 00:02:10,720 Speaker 1: answer is, as you may have guessed from my last 41 00:02:10,760 --> 00:02:13,359 Speaker 1: monologue is that there may not actually be capacity for 42 00:02:13,440 --> 00:02:16,920 Speaker 1: them to do so. Open AI's relationship with Redmond has 43 00:02:16,960 --> 00:02:20,160 Speaker 1: grown kind of chilly over the past year. I'd speculate 44 00:02:20,160 --> 00:02:23,120 Speaker 1: that Microsoft has refused to provide additional immediate capacity or 45 00:02:23,120 --> 00:02:25,400 Speaker 1: has refused to provide capacity on the chummy terms that 46 00:02:25,400 --> 00:02:28,920 Speaker 1: open Ai previously enjoyed, receiving a significant discount on the 47 00:02:28,960 --> 00:02:32,240 Speaker 1: usual ticket prices in the past. We know that Microsoft 48 00:02:32,240 --> 00:02:34,680 Speaker 1: has both walked away from two gigawats of future compute 49 00:02:34,680 --> 00:02:37,560 Speaker 1: capacity and declined the option to spend another twelve billion 50 00:02:37,600 --> 00:02:39,960 Speaker 1: dollars on core Weave's compute and core Weave if you 51 00:02:40,000 --> 00:02:44,040 Speaker 1: don't remember there that the publicly traded data centered AI 52 00:02:44,200 --> 00:02:48,880 Speaker 1: company a whole dog's dinner onto itself, and analyst house 53 00:02:48,919 --> 00:02:51,560 Speaker 1: TD Cohen suggested in that this is a sign that 54 00:02:51,639 --> 00:02:54,799 Speaker 1: Microsoft is no longer willing to shoulder the immense financial 55 00:02:54,800 --> 00:02:58,440 Speaker 1: burden of supporting open Ai, even though open ai picked 56 00:02:58,440 --> 00:03:00,600 Speaker 1: that option up, which by which I mean they took 57 00:03:00,639 --> 00:03:03,239 Speaker 1: the twelve billion dollars of compute. It isn't clear if 58 00:03:03,400 --> 00:03:06,840 Speaker 1: Corwave can actually build the capacity they need, and definitely 59 00:03:06,960 --> 00:03:08,200 Speaker 1: don't think they're going to be able to do it 60 00:03:08,240 --> 00:03:11,520 Speaker 1: in the time they need it. Microsoft allegedly walked away 61 00:03:11,520 --> 00:03:14,440 Speaker 1: from Corewave due to its failure to deliver and that 62 00:03:14,760 --> 00:03:17,120 Speaker 1: deliver the services they asked for, and indeed probably the 63 00:03:17,120 --> 00:03:19,880 Speaker 1: compute as well. If that's true, it's unclear what has 64 00:03:20,000 --> 00:03:22,800 Speaker 1: changed to make core Weave magically able to support open Ai, 65 00:03:23,440 --> 00:03:25,800 Speaker 1: or even how a company that's drowning in high interest 66 00:03:25,840 --> 00:03:28,720 Speaker 1: debt can finance the creation of several billion dollars worth 67 00:03:28,760 --> 00:03:32,239 Speaker 1: of new data centers. Also, it's not quite as simple 68 00:03:32,280 --> 00:03:34,480 Speaker 1: as open ai calling up a data center company with 69 00:03:34,520 --> 00:03:37,120 Speaker 1: a bunch of GPUs and running chat, GPT, dot ex. 70 00:03:37,920 --> 00:03:41,000 Speaker 1: Open Ai likely has reams of different requirements, and the 71 00:03:41,040 --> 00:03:43,720 Speaker 1: amount of GPUs they will need will likely vary based 72 00:03:43,760 --> 00:03:46,520 Speaker 1: on demand, putting them in a problematic situation where they 73 00:03:46,520 --> 00:03:48,320 Speaker 1: could be commuting to a bunch of compute that they 74 00:03:48,400 --> 00:03:51,760 Speaker 1: don't need if demand slows down. I've heard that companies 75 00:03:51,800 --> 00:03:55,000 Speaker 1: generally want a six to twelve month commitment for GPUs 76 00:03:55,040 --> 00:03:57,240 Speaker 1: two the cost is fixed no matter how much they 77 00:03:57,280 --> 00:04:00,560 Speaker 1: get used, or at least there's a minimum commitment. But 78 00:04:00,800 --> 00:04:04,120 Speaker 1: let's assume for a second that demand for chat GPT 79 00:04:04,240 --> 00:04:08,200 Speaker 1: continues to rise. How does OpenAI actually get that compute 80 00:04:08,280 --> 00:04:11,440 Speaker 1: if Microsoft isn't handing it over, and the Information reports 81 00:04:11,480 --> 00:04:14,040 Speaker 1: that open ai still projects to spend about thirteen billion 82 00:04:14,080 --> 00:04:16,680 Speaker 1: dollars on as your Cloud in twenty twenty five, there 83 00:04:16,720 --> 00:04:19,200 Speaker 1: aren't really a ton of other options, especially for a 84 00:04:19,200 --> 00:04:23,360 Speaker 1: company with such gigantic requirements, meaning that whatever infrastructure open 85 00:04:23,360 --> 00:04:26,160 Speaker 1: ai is building is a patchwork between smaller players, and 86 00:04:26,279 --> 00:04:31,320 Speaker 1: using so many smaller providers likely creates unavoidable inefficiencies and overhead. 87 00:04:31,839 --> 00:04:35,280 Speaker 1: I'm naming another pale horse of the AI apocalypse by 88 00:04:35,320 --> 00:04:40,039 Speaker 1: the way limits to service and service degradation across chat GPT. 89 00:04:40,720 --> 00:04:43,200 Speaker 1: Open ai is running out of compute capacity. They've talked 90 00:04:43,200 --> 00:04:45,919 Speaker 1: about it since October of last year, and chat GPT's 91 00:04:45,960 --> 00:04:49,120 Speaker 1: new image generation is a significant drain on their resources, 92 00:04:49,360 --> 00:04:52,320 Speaker 1: meaning that to continue providing their services, they're going to 93 00:04:52,400 --> 00:04:56,880 Speaker 1: need to expand capacity or reduce access to services otherwise. 94 00:04:57,600 --> 00:05:01,360 Speaker 1: The problem is that expanding is extremely different. Data centers 95 00:05:01,440 --> 00:05:03,920 Speaker 1: take three to six years to build, and open ai 96 00:05:03,960 --> 00:05:06,800 Speaker 1: has planned Stargate data Center won't have anything ready before 97 00:05:06,800 --> 00:05:09,640 Speaker 1: twenty twenty six at the earliest, which means we're approaching 98 00:05:09,640 --> 00:05:12,120 Speaker 1: a point where there simply might not be enough data 99 00:05:12,160 --> 00:05:16,240 Speaker 1: centers or GPUs to burn, while open ai could theoretically 100 00:05:16,279 --> 00:05:19,000 Speaker 1: go to Google or Amazon. Both of those companies are 101 00:05:19,000 --> 00:05:21,520 Speaker 1: invested in anthropic and have little incentive to align with 102 00:05:21,600 --> 00:05:24,839 Speaker 1: open Ai. Meta is building their own chet GPT competitor, 103 00:05:24,920 --> 00:05:29,160 Speaker 1: and Elon must despises Sam Mortman real shithead versus fuckwad 104 00:05:29,279 --> 00:05:33,000 Speaker 1: situation there. While I can't say for certain, I can't 105 00:05:33,000 --> 00:05:36,279 Speaker 1: work out where open ai will get the capacity to continue, 106 00:05:36,360 --> 00:05:38,240 Speaker 1: And I just don't know how they're going to expand 107 00:05:38,240 --> 00:05:41,600 Speaker 1: their services if Microsoft isn't providing capacity. Yes, there's a 108 00:05:41,600 --> 00:05:44,039 Speaker 1: Oracle which open ai has a partnership with, but they're 109 00:05:44,120 --> 00:05:48,720 Speaker 1: relatively small in this space. Chat GPT's immage generation has 110 00:05:48,760 --> 00:05:51,240 Speaker 1: become this massive burden on the company right at the 111 00:05:51,240 --> 00:05:54,280 Speaker 1: point where it's introducing some of its most expensive models ever, 112 00:05:54,520 --> 00:05:57,880 Speaker 1: and the products themselves are extremely expensive to run. Deep 113 00:05:57,920 --> 00:06:00,400 Speaker 1: research is perhaps the best example using O open ai 114 00:06:00,480 --> 00:06:03,359 Speaker 1: is extremely expensive O three model, which can cost in 115 00:06:03,400 --> 00:06:05,720 Speaker 1: some cases as much as one thousand dollars per query. 116 00:06:05,960 --> 00:06:10,040 Speaker 1: Deeper search is probably cheaper, but not that much cheaper, probably, 117 00:06:10,360 --> 00:06:12,680 Speaker 1: I would. I've heard rumors, and this is a rumor. 118 00:06:13,279 --> 00:06:15,400 Speaker 1: It's a rumor. I've heard like a dollar or two 119 00:06:15,400 --> 00:06:18,479 Speaker 1: per query. If that's the case, that's fucking insane. Anyway. 120 00:06:18,720 --> 00:06:22,200 Speaker 1: While open Ai could absorb the remaining capacity at say Crusoe, 121 00:06:22,279 --> 00:06:25,200 Speaker 1: Lambda and core Wave, this creates a systemic risk where 122 00:06:25,200 --> 00:06:28,200 Speaker 1: every GPU provider is reliant on open AI's money, and 123 00:06:28,240 --> 00:06:30,880 Speaker 1: this assumes that they'll actually have enough to begin with. 124 00:06:31,680 --> 00:06:35,000 Speaker 1: Open Ai also just close the largest private funding round 125 00:06:35,040 --> 00:06:38,880 Speaker 1: in history, forty billion theoretical dollars, valuing the company at 126 00:06:38,880 --> 00:06:42,159 Speaker 1: a ridiculous three hundred billion dollars raised from he gets 127 00:06:42,240 --> 00:06:45,120 Speaker 1: the soft Bank and other investors. That's good news, right, 128 00:06:46,120 --> 00:06:49,520 Speaker 1: Not really? In truth, open Ai really only raised ten 129 00:06:49,560 --> 00:06:51,560 Speaker 1: billion dollars, with seven and a half billion of those 130 00:06:51,600 --> 00:06:53,880 Speaker 1: dollars coming from soft Bank and another two point five 131 00:06:53,920 --> 00:06:57,680 Speaker 1: billion dollars coming from other investors, including Thrive Capital and Microsoft. 132 00:06:58,080 --> 00:07:00,600 Speaker 1: The remaining thirty billion dollars off, where which soft Bank 133 00:07:00,680 --> 00:07:02,719 Speaker 1: is on the hook for twenty billion dollars off, will 134 00:07:02,800 --> 00:07:06,320 Speaker 1: arrive at the end of the year. That's all we've gone. 135 00:07:06,480 --> 00:07:09,160 Speaker 1: But open Ai will only get ten billion dollars from 136 00:07:09,200 --> 00:07:11,920 Speaker 1: soft Bank, so bringing it down to a thirty billion 137 00:07:11,960 --> 00:07:14,720 Speaker 1: dollar round total. If open ai fails to convert from 138 00:07:14,760 --> 00:07:17,120 Speaker 1: a nonprofit to a for profit company by the end 139 00:07:17,160 --> 00:07:21,240 Speaker 1: of twenty twenty five, a massive acceleration there. As a reminder, 140 00:07:21,440 --> 00:07:24,080 Speaker 1: open ai is a weirdly structured nonprofit with a for 141 00:07:24,200 --> 00:07:26,720 Speaker 1: profit arm, and their last round of funding from October 142 00:07:26,720 --> 00:07:29,480 Speaker 1: twenty twenty four had another caveat that if open ai 143 00:07:29,640 --> 00:07:32,119 Speaker 1: failed to become a for profit company by October twenty 144 00:07:32,120 --> 00:07:35,720 Speaker 1: twenty six, all investment dollars would convert into debt. I've 145 00:07:35,760 --> 00:07:38,160 Speaker 1: also read that they would have to hand the money back. 146 00:07:38,560 --> 00:07:41,600 Speaker 1: I'm not sure whether that's the case. Debt is the 147 00:07:41,600 --> 00:07:44,920 Speaker 1: one that's been reported the most. Furthermore, open Ai loses 148 00:07:44,920 --> 00:07:48,000 Speaker 1: money on every single prompt on Chat GPT, even from 149 00:07:48,040 --> 00:07:51,200 Speaker 1: their two hundred dollars a month chet GPT Pro subscribers. 150 00:07:51,600 --> 00:07:53,840 Speaker 1: The burdens some interest payments would make it even harder 151 00:07:53,880 --> 00:07:56,960 Speaker 1: for open ai to reach break even, which right now 152 00:07:57,000 --> 00:07:59,760 Speaker 1: it doesn't even seem like they can do anyway. As 153 00:07:59,800 --> 00:08:02,600 Speaker 1: an another reminder, soft Bank is a company that has 154 00:08:02,640 --> 00:08:05,720 Speaker 1: now invested in two different fraudulent schemes, Wirecard and Green 155 00:08:05,760 --> 00:08:08,240 Speaker 1: Silk Capital, the latter of which helped put the nail 156 00:08:08,240 --> 00:08:10,640 Speaker 1: in the coffin of credit sweee back in twenty twenty 157 00:08:10,680 --> 00:08:15,040 Speaker 1: three and put sixteen billion dollars into we work. It 158 00:08:15,080 --> 00:08:19,360 Speaker 1: will be incredibly some might say impossibly difficult, and I'll 159 00:08:19,400 --> 00:08:21,480 Speaker 1: cover this in the future episode to convert open ai 160 00:08:21,560 --> 00:08:23,760 Speaker 1: into a for profit company, and the fact that soft 161 00:08:23,840 --> 00:08:26,640 Speaker 1: Bank is putting this caveat on their investment heavily suggests 162 00:08:26,800 --> 00:08:29,000 Speaker 1: that they have doubts it will happen. And I must 163 00:08:29,080 --> 00:08:31,960 Speaker 1: be clear. When the monopoly man is getting nervous, you 164 00:08:32,000 --> 00:08:35,800 Speaker 1: should get nervous too. The fact OpenAI accepted these terms 165 00:08:35,800 --> 00:08:38,760 Speaker 1: also suggest they're desper and I don't blame them. They've 166 00:08:38,760 --> 00:08:41,720 Speaker 1: committed eighteen billion dollars to the Stargate Data Center project, 167 00:08:41,800 --> 00:08:44,440 Speaker 1: will spend thirteen billion dollars on Microsoft Computer alone in 168 00:08:44,440 --> 00:08:47,400 Speaker 1: twenty twenty five, according to the information, and they've now 169 00:08:47,440 --> 00:08:51,640 Speaker 1: created an incredibly popular product that will guarantee people come 170 00:08:51,679 --> 00:08:54,040 Speaker 1: and use it like twice and then never use it again. 171 00:08:55,360 --> 00:08:57,960 Speaker 1: Now keep a keen eye on any restrictions that open 172 00:08:57,960 --> 00:09:00,400 Speaker 1: ai makes on chat GPT in the coming month. I 173 00:09:00,440 --> 00:09:02,440 Speaker 1: do not see how this company survives, nor do I 174 00:09:02,480 --> 00:09:05,680 Speaker 1: see how they expand their capacity much further. Price increases, 175 00:09:05,800 --> 00:09:08,000 Speaker 1: rate limits and other ways of slowing down the pressure 176 00:09:08,040 --> 00:09:11,240 Speaker 1: on their servers will likely suggest that open eye is 177 00:09:11,320 --> 00:09:13,400 Speaker 1: up against the wall, both in their ability to support 178 00:09:13,400 --> 00:09:15,600 Speaker 1: the services they provide and the costs they must bear 179 00:09:15,640 --> 00:09:18,680 Speaker 1: to provide them. We are entering the hysterical era of 180 00:09:18,679 --> 00:09:20,960 Speaker 1: the bubble, time when the craziest stuff will happen as 181 00:09:21,000 --> 00:09:23,240 Speaker 1: the money does everything it can to keep the dream alive. 182 00:09:23,880 --> 00:09:25,720 Speaker 1: I look forward to telling you what happens next.