1 00:00:00,000 --> 00:00:08,480 Any geek with a calculator has known for months that while AI is here to stay in 2 00:00:05,600 --> 00:00:14,000 some form, the financial house of cards that has propped up the recent explosion 3 00:00:10,639 --> 00:00:16,320 in investment is at best unsustainable 4 00:00:14,000 --> 00:00:21,520 and at worst threatens to destabilize the world economy. But the cracks are 5 00:00:19,439 --> 00:00:26,800 starting to show in a way that I don't think anyone can ignore anymore. video 6 00:00:23,760 --> 00:00:29,279 cards reports that DDR5 prices have 7 00:00:26,800 --> 00:00:34,640 begun to drop in markets around the world. And while many AI leaders claim 8 00:00:32,000 --> 00:00:39,360 that we are not in a bubble, in an interview with the Verge way back in 9 00:00:36,399 --> 00:00:44,320 last August, OpenAI CEO Sam Alman himself acknowledged that when bubbles 10 00:00:41,840 --> 00:00:48,239 happen, smart people get over excited, acknowledged that the way that some 11 00:00:46,160 --> 00:00:54,399 money is flowing into the AI space is not rational, and also said someone's 12 00:00:51,360 --> 00:00:56,320 going to get burned. I mean, of course, 13 00:00:54,399 --> 00:01:00,879 he doesn't think it's going to be him, but his credibility took a pretty big 14 00:00:58,640 --> 00:01:06,479 hit recently when he told two of the biggest DRM producers in the world, 15 00:01:03,280 --> 00:01:08,720 "Yes, I think I'll buy 40% of your total 16 00:01:06,479 --> 00:01:15,760 production capacity. Here's a letter of intent to that effect." Followed by, 17 00:01:11,520 --> 00:01:17,759 "Actually, lol, never mind." What? 18 00:01:15,760 --> 00:01:21,680 That's the big business equivalent of, "Oh, oh, yes, Mr. Furley. I know it's 19 00:01:20,159 --> 00:01:26,240 the first of the month, but don't worry, the check's totally in the mail. So, 20 00:01:23,680 --> 00:01:30,640 let's talk about why I think that we are right now at the beginning of the end 21 00:01:28,560 --> 00:01:34,560 for the irresponsible money that's getting dumped into AI infrastructure 22 00:01:32,400 --> 00:01:38,960 and why I think this will finally be a break for consumers after months of 23 00:01:36,799 --> 00:01:44,000 beatings. Now, to continue my own beatings until Segue improves 24 00:01:42,720 --> 00:01:47,759 to our sponsor, >> Proton Mail. Even if you're not looking 25 00:01:45,759 --> 00:01:52,240 for an alternative to your current email provider, maybe it's time to start. Many 26 00:01:50,079 --> 00:01:58,759 providers can use your inbox to track your behavior. Learn more about Proton 27 00:01:54,399 --> 00:01:58,759 Mail using our link down below. 28 00:02:06,079 --> 00:02:11,599 Let's start with the very basics. What is a bubble? Probably the best plain 29 00:02:10,160 --> 00:02:16,800 English definition that I could find came from Investopedia, who defines it 30 00:02:13,520 --> 00:02:18,879 as an asset's price rising far above its 31 00:02:16,800 --> 00:02:23,280 real value because many people rushed to buy it, hoping to sell it at a higher 32 00:02:20,720 --> 00:02:29,280 price. The earliest recorded example is tulip mania. A period during the mid600s 33 00:02:26,400 --> 00:02:33,920 when tulip bulbs became the status symbol duour in Holland, causing values 34 00:02:31,760 --> 00:02:39,120 to soar uncontrollably with rare specimens apparently rivaling the price 35 00:02:36,000 --> 00:02:41,120 of a fine house on the Amsterdam canal. 36 00:02:39,120 --> 00:02:46,000 When the inevitable crash occurred, it wiped out as much as 99% of the total 37 00:02:44,160 --> 00:02:51,040 market valuation, leaving many speculative investors deeply indebted. 38 00:02:49,200 --> 00:02:56,319 That doesn't mean that tulips became worthless. They've just fallen a little 39 00:02:53,599 --> 00:03:00,239 bit behind relative to other assets. With that out of the way, then what 40 00:02:57,920 --> 00:03:05,360 makes AI a bubble? Well, the biggest one is the valuations of the companies who 41 00:03:02,879 --> 00:03:13,040 are leading its development. In no particular order, anthropic 380 billion, 42 00:03:09,040 --> 00:03:16,879 XAI, 200 plus billion. Open AAI an 43 00:03:13,040 --> 00:03:18,560 eyewatering $852 billion. Like 44 00:03:16,879 --> 00:03:23,360 seriously, they're practically knocking on the trillion dollar door, which isn't 45 00:03:21,040 --> 00:03:28,080 inherently a problem. There's plenty of legitimate members of the Tea Club. But 46 00:03:26,080 --> 00:03:33,920 here's the thing. With some notable exceptions, those other members also 47 00:03:31,040 --> 00:03:39,680 have Mondo revenues to go with their Mondo valuations. Samsung, for instance, 48 00:03:37,440 --> 00:03:44,560 while totally participating in the AI boom in the form of HBM memory 49 00:03:41,440 --> 00:03:47,680 manufacturing for GPUs, also happens to 50 00:03:44,560 --> 00:03:50,879 have revenues that exceed $220 billion 51 00:03:47,680 --> 00:03:54,159 US a year. How has their value fallen 52 00:03:50,879 --> 00:03:56,560 below OpenAI? I mean, in fairness, 53 00:03:54,159 --> 00:04:01,280 OpenAI did recently boast that they have exceeded revenues of $2 billion a month 54 00:03:59,040 --> 00:04:05,120 and hit this milestone faster than internet era defining companies like 55 00:04:03,040 --> 00:04:09,840 Alphabet and Meta, which is definitely impressive. But, uh, let me let you in 56 00:04:07,680 --> 00:04:15,439 on a little secret. As someone who runs a business, I am acutely aware that 57 00:04:12,400 --> 00:04:19,199 revenue and profits are not the same 58 00:04:15,439 --> 00:04:23,280 thing. See, Samsung is making over $28 59 00:04:19,199 --> 00:04:26,400 billion a year in profit. Open AI is 60 00:04:23,280 --> 00:04:28,639 doing about that in revenue while still 61 00:04:26,400 --> 00:04:34,720 burning billions of dollars a year after over a decade of operation with no end 62 00:04:31,120 --> 00:04:37,919 in sight. So what's their plan? More 63 00:04:34,720 --> 00:04:39,600 consumer AI subscriptions. Open AAI says 64 00:04:37,919 --> 00:04:43,440 this is going to require massive investments in infrastructure to flood 65 00:04:41,759 --> 00:04:47,919 the world with intelligence as cheap and ubiquitous as electricity. After which I 66 00:04:46,160 --> 00:04:52,320 assume the plan is to leverage their position to squeeze their locked in 67 00:04:49,840 --> 00:04:57,360 customers for profit. it. I mean, they already have 900 million users. That's 68 00:04:55,040 --> 00:05:02,000 about one in every 10 men, women, and children on the face of the earth. Sure, 69 00:04:59,680 --> 00:05:04,240 they can go after the other nine, but I kind of have a hard time believing that 70 00:05:03,360 --> 00:05:08,880 that's going to be the difference between losing billions and making 71 00:05:06,720 --> 00:05:13,120 billions. Enterprise tools are an interesting path. Except that paying 72 00:05:11,520 --> 00:05:18,160 customers are the most likely to be critical of an AI that, for being 73 00:05:15,600 --> 00:05:23,039 honest, is nowhere near true artificial general intelligence. And as for how 74 00:05:20,880 --> 00:05:28,479 it's going so far, the integration of enterprise AI has been, let's call it, 75 00:05:25,759 --> 00:05:33,440 uneven at best. Personally, I find it more of an irritant than a help from a 76 00:05:30,400 --> 00:05:35,199 day-to-day perspective. But even if I 77 00:05:33,440 --> 00:05:41,280 accept the premise that widespread corporate adoption of AI is inevitable, 78 00:05:37,759 --> 00:05:43,680 how much of let's say my internal tools 79 00:05:41,280 --> 00:05:48,479 helper LLM or my customer service chatbot would I really outsource to 80 00:05:46,080 --> 00:05:55,280 OpenAI instead of just hosting locally on a quantized open- source model? 81 00:05:52,320 --> 00:06:00,720 And so far, I've completely ignored the threats from commercial competitors like 82 00:05:57,600 --> 00:06:02,880 Anthropic and probably more importantly, 83 00:06:00,720 --> 00:06:07,600 Google, who hasn't just been sitting there. So then, wait, why have I only 84 00:06:04,800 --> 00:06:12,560 talked about Open AI so far? I guess it's because kind of like you see in 85 00:06:09,520 --> 00:06:15,120 crypto, where as Bitcoin goes, so does 86 00:06:12,560 --> 00:06:20,560 the rest of the industry. Open AI tends to act as a pretty decent proxy for the 87 00:06:17,520 --> 00:06:22,800 current vibes around AI as a whole. And 88 00:06:20,560 --> 00:06:28,639 right now, things aren't looking so good for them. Let's look at some examples. 89 00:06:25,600 --> 00:06:31,039 CNBC reports that amid concerns that 90 00:06:28,639 --> 00:06:34,960 expansion ambitions were too great for the potential revenue that would follow, 91 00:06:32,960 --> 00:06:41,840 OpenAI slashed their spending expectations, targeting just 600 billion 92 00:06:38,160 --> 00:06:44,160 by 2030. That uh might affect their 93 00:06:41,840 --> 00:06:48,160 ability to buy 40% of the world's memory, which in turn would take some of 94 00:06:46,479 --> 00:06:53,199 the pressure off of competitors who are trying to keep pace. Energy costs. This 95 00:06:51,039 --> 00:06:57,520 is another big one. It's the single biggest expense for AI companies outside 96 00:06:55,440 --> 00:07:01,919 of the hardware itself, and they have soared over the last half a decade. It 97 00:07:00,319 --> 00:07:06,720 wouldn't be a problem if it wasn't a trend that has only accelerated due to 98 00:07:04,080 --> 00:07:12,080 the ongoing code orange state of current global affairs. Case in point, OpenAI's 99 00:07:09,599 --> 00:07:16,880 abrupt complete cancellation of the Sora 2 video generation app. By all accounts, 100 00:07:14,639 --> 00:07:21,840 it seemed to be just plain costing them way too much money to run. Then there's 101 00:07:19,599 --> 00:07:25,680 Oracle, who meanwhile has been cruising from high to high, announcing a 102 00:07:23,520 --> 00:07:29,599 partnership with OpenAI to build AI infrastructure. They've boasted that 103 00:07:27,680 --> 00:07:34,400 demand for AI infrastructure continues to exceed supply, and they had an 104 00:07:31,919 --> 00:07:39,680 exceptional quarter. And yet, according to The Independent, they just laid off 105 00:07:36,720 --> 00:07:44,479 as much as 18% of their global workforce amidst a broader organizational change. 106 00:07:42,800 --> 00:07:48,639 Now, their stock has actually jumped about 5% between conceiving of this 107 00:07:46,720 --> 00:07:51,199 video and sitting down to write it. And it's probably going to move some more by 108 00:07:49,840 --> 00:07:55,360 the time the editor gets their hands on the footage. But that doesn't change 109 00:07:53,120 --> 00:07:58,560 that the overall trajectory is pretty rough. And they're sitting at about half 110 00:07:57,120 --> 00:08:02,560 of their all-time high back in September. Microsoft is actually going 111 00:08:00,639 --> 00:08:08,639 through a similar correction, sitting just 9% over the highest point it 112 00:08:05,360 --> 00:08:10,639 reached in November of 2021. 113 00:08:08,639 --> 00:08:14,400 Now, not every AI company is a similar story right now. NVIDIA is still selling 114 00:08:12,720 --> 00:08:18,400 golden picks and shovels faster than they can possibly craft them. And Google 115 00:08:16,400 --> 00:08:21,919 and Meta have both taken much smaller hits, more in line with the kind of 116 00:08:20,319 --> 00:08:26,639 thing that happens during times of geopolitical uncertainty as money flows 117 00:08:24,240 --> 00:08:32,320 away from tech bros and then towards weapons bros or their evolved hybrid 118 00:08:29,759 --> 00:08:36,880 tech weapons bros. Anyway, that's a little off topic. Point is, I don't 119 00:08:34,640 --> 00:08:41,839 think AI is going away. Meta got up on stage at ARM's AI CPU launch last month 120 00:08:39,440 --> 00:08:45,440 to talk about their vision for AI's integration into their products and it's 121 00:08:43,599 --> 00:08:50,480 clear that they are going full steam ahead. Every facet of the Meta 122 00:08:48,399 --> 00:08:55,600 experience is going to have AI in it to the point where whether you like it or 123 00:08:52,800 --> 00:09:02,399 even think about it, their 3 billion plus daily users will be using AI every 124 00:08:59,120 --> 00:09:04,240 day. And that is going to take a lot of 125 00:09:02,399 --> 00:09:08,959 infrastructure. Infrastructure that isn't built yet. 126 00:09:06,080 --> 00:09:12,800 But I don't think that the Open AIs of the world are the ones that are going to 127 00:09:10,720 --> 00:09:18,720 own that infrastructure. I could be wrong, but I see Open AI as 128 00:09:16,080 --> 00:09:24,399 more of like a Dropbox like company, one that will probably survive the bubble 129 00:09:21,600 --> 00:09:29,040 pop due to its ubiquitousness and its brand value, but that might struggle to 130 00:09:26,800 --> 00:09:32,959 convert free users to paid users and ultimately have to settle for a niche 131 00:09:31,040 --> 00:09:36,080 role in the industry that they defined. Meta and Google, on the other hand, have 132 00:09:34,800 --> 00:09:42,320 something in common that I think has shielded them from the brunt of the bubble's deflation so far. They aren't 133 00:09:40,560 --> 00:09:46,720 simply landgrabbing as much infrastructure as possible in hopes that 134 00:09:44,080 --> 00:09:50,800 it'll be useful someday. Their valuation is based on their products usefulness 135 00:09:48,560 --> 00:09:56,399 outside of AI and revenue that they derive outside of AI. So when OpenAI 136 00:09:54,480 --> 00:10:00,959 stumbles, which they seem to be doing now, a bit of that competitive pressure 137 00:09:59,040 --> 00:10:05,920 gets lifted and their infrastructure investments can maybe return to a more 138 00:10:03,519 --> 00:10:11,680 as needed scale as they continue to find uses for this admittedly sometimes very 139 00:10:08,880 --> 00:10:16,720 cool and useful new technology. Really what it comes down to for me is I think 140 00:10:13,279 --> 00:10:20,079 that OpenAI's 40% DRAM decommmit episode 141 00:10:16,720 --> 00:10:22,079 has rightly spooked a lot of the folks 142 00:10:20,079 --> 00:10:26,320 who have been making multi-year, multibillion dollar investments, 143 00:10:24,560 --> 00:10:30,880 assuming that the future revenue that they were building for was some kind of 144 00:10:28,560 --> 00:10:35,200 guarantee rather than just a signature on a worthless piece of paper. And now 145 00:10:32,880 --> 00:10:38,880 that that realization has happened, I think the bean counters are on the verge 146 00:10:36,880 --> 00:10:43,680 of getting back in the driver's seat, which will hopefully see investments in 147 00:10:40,720 --> 00:10:48,399 AI return to a more reasonable level as the winners determine what kind of real 148 00:10:45,839 --> 00:10:52,959 revenue they can actually derive from AI services. With all that said, I do still 149 00:10:51,279 --> 00:10:57,519 think it's going to take some time. Tulip prices didn't crash overnight 150 00:10:54,880 --> 00:11:01,279 either. So, uh, hey, hang in there everyone, guys. I think there's light at 151 00:10:59,600 --> 00:11:05,360 the end of the tunnel and it might not be a train. Oh, you know what else is at 152 00:11:04,079 --> 00:11:10,160 the end of the tunnel? This cool not slop t-shirt. ltstore.com. 153 00:11:07,839 --> 00:11:14,480 And also the segue to our sponsor, >> Proton Mail. There was a time before you 154 00:11:12,399 --> 00:11:19,200 had a computer, a tablet, or even an email address. A time when you 155 00:11:16,560 --> 00:11:23,279 maintained some level of privacy. But the fact of the matter is that in this 156 00:11:21,040 --> 00:11:27,760 digital landscape, it is assumed that we just don't have the privacy online. And 157 00:11:25,600 --> 00:11:31,920 we've kind of come to accept that. Something like your email address acts 158 00:11:29,600 --> 00:11:36,079 as a lynch pin for just about everything you do online, which makes it a perfect 159 00:11:34,240 --> 00:11:39,920 hub for the big corporations to track your behaviors. And then that 160 00:11:37,680 --> 00:11:44,640 information can be used for things like uh training AI. Proton Mail doesn't 161 00:11:42,480 --> 00:11:49,120 participate in any of that nonsense. There's no ads, tracking pixels, or 162 00:11:46,800 --> 00:11:53,839 email scanning. Their platform focuses more on protecting you than profiting. 163 00:11:51,600 --> 00:11:58,000 Your inbox should be for you to use, not for you to be used. So, protect your 164 00:11:56,160 --> 00:12:01,680 digital privacy by checking out Proton Mail using our link in the video 165 00:11:59,680 --> 00:12:04,880 description. If you guys enjoyed this video, you might like the one we did 166 00:12:03,120 --> 00:12:11,120 recently showing that you don't actually need the most expensive DDR5 memory to 167 00:12:07,680 --> 00:12:11,120 get peak gaming performance.