1 00:00:00,160 --> 00:00:04,760 This week Sam Altman, perhaps AIS biggest influencer, just sat 2 00:00:04,760 --> 00:00:07,600 down with a journalist and answered a really interesting 3 00:00:07,600 --> 00:00:10,400 question whether or not we're in a bubble. 4 00:00:10,480 --> 00:00:15,560 And he said, and I quote my opinion is yes, in the same week 5 00:00:15,680 --> 00:00:20,480 that Chachi BT had a mass rebellion over users desperately 6 00:00:20,480 --> 00:00:25,400 wanting the drug that is Chachi BT4-O and other loved models and 7 00:00:25,400 --> 00:00:28,360 Sam Altman announcing billions of dollars in spending on data 8 00:00:28,360 --> 00:00:30,720 centres and infrastructure over the next 10 years. 9 00:00:30,800 --> 00:00:33,000 Today we're going to examine this contradiction and try and 10 00:00:33,000 --> 00:00:35,840 answer the question of whether we think AI is really in a 11 00:00:35,840 --> 00:00:37,400 bubble. This is in the leap with Jack 12 00:00:37,400 --> 00:00:39,280 Horton. I hope you enjoy the show. 13 00:00:51,090 --> 00:00:54,290 Let's start right at the top with what he actually said now 14 00:00:55,210 --> 00:00:56,810 with the journalist and The Verge. 15 00:00:57,010 --> 00:00:59,010 He actually said he thinks we're in a bubble. 16 00:00:59,720 --> 00:01:01,920 And what he said was actually quite interesting and quite 17 00:01:01,920 --> 00:01:05,680 clever because when he was asked are we in a phase where 18 00:01:05,680 --> 00:01:08,920 investors as a whole are over excited about AI? 19 00:01:08,920 --> 00:01:12,880 And he said his opinion is yes. And he called it insane that AI 20 00:01:12,880 --> 00:01:16,640 startups would say 3 people and an idea are getting massive 21 00:01:16,640 --> 00:01:19,640 valuations. And he warns that someone's 22 00:01:19,640 --> 00:01:24,280 going to get burned here and that person who loses out are 23 00:01:24,280 --> 00:01:26,560 going to lose in a phenomenal amount of money. 24 00:01:26,720 --> 00:01:29,760 Now Almond's point didn't come out of nowhere. 25 00:01:29,760 --> 00:01:31,160 It was just obviously days after church. 26 00:01:31,160 --> 00:01:35,280 GBTS work probably could be only described as a disappointing 27 00:01:35,280 --> 00:01:37,520 launch, or at least the reception of that launch has 28 00:01:37,520 --> 00:01:40,600 been disappointing. And actually you could argue 29 00:01:40,600 --> 00:01:43,640 that it's just damage control after maybe that big flagship 30 00:01:43,640 --> 00:01:46,000 product of the year didn't really go down as well as maybe 31 00:01:46,000 --> 00:01:48,720 they hoped. Now, his point in particular was 32 00:01:48,720 --> 00:01:53,520 that when bubbles happen, smart people start to get overexcited 33 00:01:53,800 --> 00:01:56,400 about, let's say, a kernel of truth, as he puts it. 34 00:01:56,680 --> 00:02:00,280 And if you look at most bubbles in history, most of the time, 35 00:02:00,280 --> 00:02:02,560 like the tap bubble, there was a it was a real thing. 36 00:02:02,560 --> 00:02:05,280 And that's that kernel of truth. We know this thing is going to 37 00:02:05,480 --> 00:02:08,080 reshape the world. And he thinks AI might be doing 38 00:02:08,080 --> 00:02:10,240 the same to people. Everyone knows it's going to 39 00:02:10,240 --> 00:02:12,200 reshape the world. And that's that kernel of truth 40 00:02:12,200 --> 00:02:15,880 that people obsess over. And therefore that's so big, but 41 00:02:15,880 --> 00:02:19,040 that's so badly. But by comparing it to the.com 42 00:02:19,040 --> 00:02:23,280 bubble, actually what he's doing is implicitly arguing that for 43 00:02:23,280 --> 00:02:26,600 all of the failed companies and the companies that I guess 44 00:02:26,920 --> 00:02:29,680 popped during that bubble, there are the Amazons. 45 00:02:30,080 --> 00:02:33,480 And those Amazons are the ones that will take the lion's share 46 00:02:33,480 --> 00:02:36,760 of the profits. And obviously he's implying that 47 00:02:36,880 --> 00:02:39,320 Open AI is going to be one of those Amazon companies. 48 00:02:39,440 --> 00:02:41,880 Now, I think a lot of this narrative, and it's been a real 49 00:02:41,880 --> 00:02:45,080 narrative that's emerged online in the last few days can be 50 00:02:45,080 --> 00:02:50,240 boiled down to this, I guess, summary and tweet by Antonio 51 00:02:50,240 --> 00:02:54,080 Garcia and Nick Bauman. And the reason it's such a good 52 00:02:54,080 --> 00:02:57,520 explanation of, I guess this narrative around the bubble is 53 00:02:57,520 --> 00:03:00,160 because of the historical analysis it makes. 54 00:03:00,240 --> 00:03:03,920 What basically they say is that during the Web 2 revolution, so 55 00:03:03,920 --> 00:03:07,000 when the Internet first emerged and you have the Amazons, 56 00:03:07,240 --> 00:03:10,800 essentially what they say is that venture capital money for 57 00:03:10,800 --> 00:03:14,880 very wealthy individuals were basically going towards paid ads 58 00:03:14,880 --> 00:03:17,920 to pump the number of monthly active users on the platforms. 59 00:03:18,560 --> 00:03:22,480 And then crypto essentially use a very similar mechanism, but 60 00:03:22,480 --> 00:03:25,760 raising money from retail investors and people who are 61 00:03:25,760 --> 00:03:28,520 desperate to make some money. And what they were doing was 62 00:03:28,520 --> 00:03:31,920 spending money on user rewards to then try and get more people 63 00:03:31,920 --> 00:03:36,160 using the platform, again, inflating usage with artificial 64 00:03:36,160 --> 00:03:39,520 access to cash and AI. What these people are claiming 65 00:03:39,520 --> 00:03:42,760 is actually using venture capital money like Web 2 did. 66 00:03:43,880 --> 00:03:46,720 And what they're arguing is that in the AI industry right now, 67 00:03:47,400 --> 00:03:50,920 what's happening is this venture capital money being used again 68 00:03:51,280 --> 00:03:55,960 to subsidize compute costs. So costs to pay API providers 69 00:03:55,960 --> 00:03:59,640 such as open AI or data centers and infrastructure, which as a 70 00:03:59,640 --> 00:04:03,280 result inflates the number of people using it for the amount 71 00:04:03,280 --> 00:04:05,200 they're paying. And what they say is that 72 00:04:05,200 --> 00:04:07,160 eventually this is just going to blow up at some point because 73 00:04:08,160 --> 00:04:10,680 the venture capitalists, whether they stop putting money in, 74 00:04:10,680 --> 00:04:13,120 which means that nobody can afford to get usage, and because 75 00:04:13,120 --> 00:04:17,040 they never delivered real value, the entire industry just kind of 76 00:04:17,040 --> 00:04:19,320 collapses like a souffle. So in other words, every 77 00:04:19,320 --> 00:04:23,200 technology boom basically is being driven by external 78 00:04:23,200 --> 00:04:26,440 capital, by investors that subsidizes much of the costs, 79 00:04:26,960 --> 00:04:30,520 which helps create artificial demand that then evaporates when 80 00:04:30,520 --> 00:04:33,760 eventually the money runs out. And he thinks that following 81 00:04:33,760 --> 00:04:36,480 this playbook, that AI is just therefore in a bubble, and 82 00:04:36,480 --> 00:04:38,640 eventually it's going to follow the exact same path. 83 00:04:38,800 --> 00:04:41,120 Now, if you've been following this podcast, you might guess 84 00:04:41,120 --> 00:04:44,160 that I don't agree with the analogies and the comparisons 85 00:04:44,160 --> 00:04:46,480 they're making. But I guess before I give my 86 00:04:46,480 --> 00:04:48,920 personal opinion, I'll go through some of the arguments 87 00:04:48,920 --> 00:04:50,480 and the data supporting this opinion. 88 00:05:01,100 --> 00:05:04,940 So TechCrunch did an investigation into the cost of 89 00:05:04,940 --> 00:05:07,900 these, for example, AI coding assistance is just one example, 90 00:05:08,140 --> 00:05:12,100 and the cost of compute, so paying for API services and all 91 00:05:12,100 --> 00:05:13,780 the infrastructure that sits behind everything. 92 00:05:14,260 --> 00:05:16,780 And so they did an investigation and they actually found that 93 00:05:17,140 --> 00:05:20,500 margins on many of these products are either neutral or 94 00:05:20,500 --> 00:05:22,100 negative. So that means they're losing 95 00:05:22,100 --> 00:05:24,300 money on almost every single user interaction. 96 00:05:24,460 --> 00:05:27,460 And if the user always wants the best model, because why would 97 00:05:27,460 --> 00:05:31,600 you not, then each competing company is forced to always have 98 00:05:31,600 --> 00:05:34,680 the best model or the newest thing, which just cost more and 99 00:05:34,680 --> 00:05:36,560 more money. And when the gross margins are 100 00:05:36,800 --> 00:05:40,040 sometimes negative, then that screams like a big problem. 101 00:05:40,160 --> 00:05:43,440 And The Information reported that Replits are another big AI 102 00:05:43,440 --> 00:05:46,640 coding platform. Its revenue and gross margins 103 00:05:46,640 --> 00:05:49,920 fluctuated between or its gross margins, shall I say, fluctuated 104 00:05:49,920 --> 00:05:55,560 between 36% and -14% which really showcases that pain and 105 00:05:55,560 --> 00:05:59,120 relying on 3rd party AI models. And for context, as a SAS 106 00:05:59,120 --> 00:06:02,120 companies are technology providers selling software as a 107 00:06:02,120 --> 00:06:04,520 service for a subscription or a license fee. 108 00:06:04,760 --> 00:06:08,120 You're looking for eighty 8590% gross margins. 109 00:06:08,200 --> 00:06:11,560 So this data makes it I guess easy to maybe assume or think 110 00:06:11,560 --> 00:06:15,000 that we must be in a bubble because clearly venture capital 111 00:06:15,000 --> 00:06:18,920 money is just subsidizing these big companies and they're all 112 00:06:18,920 --> 00:06:22,280 going to go bust eventually. Well, personally, I don't buy 113 00:06:22,280 --> 00:06:34,400 that. There are many, many reasons why 114 00:06:34,400 --> 00:06:35,520 and I'm going to go through some of them. 115 00:06:35,760 --> 00:06:38,720 And one of the first ones is that this data, what it didn't 116 00:06:38,720 --> 00:06:41,800 reveal is actually the speed in which these companies are trying 117 00:06:41,800 --> 00:06:44,640 to solve these problems. Because naturally clever 118 00:06:44,640 --> 00:06:48,320 successful companies just adapt. They make something work. 119 00:06:48,480 --> 00:06:51,720 They've got a very good incentive to fix this problem. 120 00:06:51,920 --> 00:06:53,280 A great example of this is Replit. 121 00:06:53,280 --> 00:06:55,880 So one of the companies that I talked about just a minute ago, 122 00:06:56,080 --> 00:07:00,120 they actually replaced their previous pricing model to a 123 00:07:00,120 --> 00:07:03,360 usage based pricing model and that increased the gross margins 124 00:07:03,360 --> 00:07:07,400 to 23%. So from -14 to 23%. 125 00:07:07,680 --> 00:07:11,280 Again, still very small numbers, but big and clever companies 126 00:07:11,800 --> 00:07:13,840 have the incentive to figure it out. 127 00:07:13,960 --> 00:07:17,160 Another reason I don't buy this historical analysis is that 128 00:07:17,160 --> 00:07:19,240 there's just this huge infrastructure build out right 129 00:07:19,240 --> 00:07:21,240 now, which is extremely long term. 130 00:07:21,880 --> 00:07:24,800 Because the historical comparison to say, Web 2 131 00:07:24,800 --> 00:07:28,560 suggests that these external subsidies of venture capital 132 00:07:28,560 --> 00:07:32,560 money, for example, props up completely unsustainable usage, 133 00:07:32,560 --> 00:07:35,120 which completely vanishes when funding stops coming. 134 00:07:35,320 --> 00:07:38,120 But I don't really see that. For example, infrastructure 135 00:07:38,120 --> 00:07:41,680 investments look nothing like what you see in a big bubble 136 00:07:42,000 --> 00:07:45,520 because the scale is just unprecedented and it's long term 137 00:07:45,520 --> 00:07:47,400 focused. A great example of this is 138 00:07:47,400 --> 00:07:51,520 Equinox who has just secured a GW of nuclear energy. 139 00:07:51,640 --> 00:07:56,200 These are 20 year power purchase agreements for factories that 140 00:07:56,200 --> 00:07:58,600 are still in development. To me, this just shows that AI 141 00:07:58,600 --> 00:08:02,520 is reaching such an industrious scale it's now being treated 142 00:08:02,520 --> 00:08:06,160 like aluminium or steel mills in power planning. 143 00:08:06,320 --> 00:08:08,760 And Bloomberg just announced that U.S. data center 144 00:08:08,760 --> 00:08:13,680 construction is actually on pace to now take over office building 145 00:08:13,680 --> 00:08:16,880 in America. So that is now more data centers 146 00:08:16,880 --> 00:08:20,200 being built out than there are general office buildings being 147 00:08:20,200 --> 00:08:21,840 built. In a world still adjusting to 148 00:08:21,840 --> 00:08:25,040 remote work, this slack in office construction is now being 149 00:08:25,040 --> 00:08:29,720 filled by server farms to basically just fuel AI. 150 00:08:29,880 --> 00:08:33,760 Another really big thing to me is that adoption is through the 151 00:08:33,760 --> 00:08:36,039 roof. So when adoption metrics are 152 00:08:36,039 --> 00:08:38,000 disconnected from real value creation. 153 00:08:38,000 --> 00:08:42,000 So Web 2's monthly active users or cryptos total value on the 154 00:08:42,000 --> 00:08:45,160 chain, then I'd worry about a bubble. 155 00:08:45,160 --> 00:08:48,160 But AI presents a different dynamic because each interaction 156 00:08:48,360 --> 00:08:51,680 with an AI system is representing usually some form 157 00:08:51,680 --> 00:08:55,000 of work being performed. So usage data just as of this 158 00:08:55,000 --> 00:08:56,880 month. So in August 2025. 159 00:08:57,120 --> 00:09:00,800 It's staggering in both scale and what it shows us about 160 00:09:00,800 --> 00:09:03,000 value. We're literally a point where 161 00:09:03,640 --> 00:09:07,080 ChatGPT weekly users have hit 700 million, which is up from 162 00:09:07,080 --> 00:09:11,800 400 million in just February, and ChatGPT is on track to have 163 00:09:11,800 --> 00:09:15,720 more conversations per day than all human beings combined. 164 00:09:15,880 --> 00:09:18,920 And even more telling is token processing, because Google 165 00:09:18,920 --> 00:09:22,480 announced that it's now processing 980 trillion tokens 166 00:09:22,640 --> 00:09:25,320 monthly, which is double the level of just two months 167 00:09:25,320 --> 00:09:29,680 earlier, which represents 104% growth in usage just in two 168 00:09:29,680 --> 00:09:31,920 months alone. Unlike previous bubbles, each 169 00:09:31,920 --> 00:09:36,400 token represents a specific number of words and is actually 170 00:09:36,400 --> 00:09:40,200 a request for AI to perform actual work or something, often 171 00:09:40,200 --> 00:09:43,640 quite helpful, whether that's writing code or analyzing data 172 00:09:43,640 --> 00:09:47,480 or drafting documents. And the ChatGPT 5 backlash, to 173 00:09:47,480 --> 00:09:49,760 me, validates real usage and value. 174 00:09:49,960 --> 00:09:53,240 Users weren't suddenly abandoning ChatGPT, and AI tools 175 00:09:53,240 --> 00:09:55,120 because they weren't happy with GBT 5. 176 00:09:55,400 --> 00:09:59,280 They were actually demanding online in a movement that they 177 00:09:59,280 --> 00:10:01,760 bring back the old models that people used to live. 178 00:10:01,880 --> 00:10:04,960 There are literal Reddit posts online about how people feel 179 00:10:04,960 --> 00:10:08,360 like they've just lost a friend because the GPT 4-O model has 180 00:10:08,360 --> 00:10:10,280 been killed. This is the behavior of people 181 00:10:10,280 --> 00:10:13,480 who have integrated AI so deeply into their lives that they 182 00:10:13,480 --> 00:10:16,720 couldn't imagine not having this thing in their life anymore. 183 00:10:16,840 --> 00:10:19,880 So this bubble thesis also assumes that the subsidized 184 00:10:20,120 --> 00:10:24,920 costs remain super, super high, making there for a eventual 185 00:10:24,920 --> 00:10:26,960 pricing correction almost inevitable. 186 00:10:27,440 --> 00:10:30,040 But I don't agree with that either because what we're seeing 187 00:10:30,040 --> 00:10:32,680 is pricing improvements happening faster than ever. 188 00:10:33,840 --> 00:10:38,120 So for example, in the past year open air has GPT 4 with an 8000 189 00:10:38,120 --> 00:10:42,880 context window had cost $30 per million input tokens and then 190 00:10:42,880 --> 00:10:45,360 $60.00 per million output tokens. 191 00:10:45,720 --> 00:10:50,880 Whereas GPT 4 turbo just a year later costs 50% less and then 192 00:10:50,880 --> 00:10:55,440 67% less than that. And ChatGPT 5 is coming in even 193 00:10:55,440 --> 00:10:58,600 cheaper than that. So this pricing model validates 194 00:10:58,600 --> 00:11:01,640 underlying demand. When Replit shifted to compute 195 00:11:01,640 --> 00:11:04,800 time billing so usage based Millet bidding margins 196 00:11:04,800 --> 00:11:08,440 immediately increased. When cursor added usage charges 197 00:11:08,440 --> 00:11:11,680 to their product, customers started paying more money rather 198 00:11:11,680 --> 00:11:13,840 than churning. I'm not saying that this startup 199 00:11:13,840 --> 00:11:16,280 world is going to be easy for many of these companies in the 200 00:11:16,280 --> 00:11:21,320 space, but there's evidence of real value creation, not 201 00:11:21,320 --> 00:11:24,120 artificial demand propped up by extremely low prices. 202 00:11:24,240 --> 00:11:26,760 And finally, to me, this ignores a very important discussion 203 00:11:26,760 --> 00:11:29,600 because many of these metrics are talking about consumer 204 00:11:29,600 --> 00:11:32,560 dynamics, whereas most sustainable businesses are going 205 00:11:32,560 --> 00:11:34,400 to be built on enterprise integration. 206 00:11:34,440 --> 00:11:36,880 So this survey found the integration complexity was the 207 00:11:36,880 --> 00:11:40,320 number one barrier to enterprise adoption, ranking even above 208 00:11:40,320 --> 00:11:42,320 cost. It's expensive, it's time 209 00:11:42,320 --> 00:11:46,840 consuming, it's really specific to individual businesses, but 210 00:11:46,840 --> 00:11:48,680 it's where the recurring revenue gets built. 211 00:11:59,660 --> 00:12:01,740 So let's come back to where I think this could be heading. 212 00:12:02,140 --> 00:12:06,340 So scenario 1 is that we have, I guess, small pop of the bubble 213 00:12:06,900 --> 00:12:09,660 within a very large build out of infrastructure. 214 00:12:10,260 --> 00:12:13,660 So we might see some correction of the markets over this next 215 00:12:13,660 --> 00:12:15,020 period. So maybe there's going to be a 216 00:12:15,020 --> 00:12:20,460 drop in the value people are placing on equity by maybe 30 to 217 00:12:20,460 --> 00:12:22,580 50%. So you're not going to get these 218 00:12:22,580 --> 00:12:28,000 hugely inflated valuations really quickly, which may be 219 00:12:28,000 --> 00:12:32,160 results in many start-ups being bought by other companies and 220 00:12:32,160 --> 00:12:35,040 then resulting in less venture funding going into the space 221 00:12:35,040 --> 00:12:36,920 potentially. But infrastructure spending will 222 00:12:36,920 --> 00:12:41,280 just continue because major tech companies see this as absolutely 223 00:12:41,280 --> 00:12:45,960 essential no matter what the short term sentiment is. 224 00:12:46,160 --> 00:12:47,520 And to track this, it'd be quite simple. 225 00:12:47,520 --> 00:12:51,720 We track GPU utilization rates, maybe data center construction, 226 00:12:51,720 --> 00:12:55,000 build outs over the world, despite the fact that stock 227 00:12:55,000 --> 00:12:57,480 markets might go up and down. So let's go to scenario 2 of 228 00:12:57,480 --> 00:13:01,400 what could happen, which is I guess a very soft landing and 229 00:13:01,400 --> 00:13:04,400 massive scale up. So eventually markets slowly 230 00:13:04,400 --> 00:13:07,840 adjust what they expect to see. Revenue growth from many of 231 00:13:07,840 --> 00:13:11,480 these companies starts to catch up from the amount they've spent 232 00:13:12,400 --> 00:13:16,080 and gross margins for all these companies again start to go up, 233 00:13:16,080 --> 00:13:19,160 not down. In this scenario, enterprise 234 00:13:19,160 --> 00:13:22,400 adoption will just accelerate as integration challenges get 235 00:13:22,400 --> 00:13:25,080 solved. Both scenarios assume again that 236 00:13:25,080 --> 00:13:26,920 infrastructure build out continues. 237 00:13:27,440 --> 00:13:30,480 And I guess the question really to me is just how bumpy the 238 00:13:30,480 --> 00:13:33,440 financial ride becomes. So I guess to conclude, this 239 00:13:33,440 --> 00:13:37,800 discussion definitely captures something real about the current 240 00:13:37,800 --> 00:13:41,760 AI market because we are seeing subsidized usage, we're seeing 241 00:13:41,760 --> 00:13:47,280 massive valuations and really unsustainable unit economics and 242 00:13:47,280 --> 00:13:49,680 gross margins for many of these companies. 243 00:13:50,280 --> 00:13:53,840 So the application layer absolutely shows bubble 244 00:13:53,840 --> 00:13:56,440 characteristics. But at the same time, there's a 245 00:13:56,440 --> 00:13:59,080 lot of evidence saying the absolute opposite, the immediate 246 00:13:59,080 --> 00:14:02,760 utility that's driving massive adoption, the infrastructure 247 00:14:02,760 --> 00:14:06,160 commitment spanning decades, the rapidly reducing costs of all 248 00:14:06,160 --> 00:14:09,800 these incredible services. And most importantly for the 249 00:14:09,800 --> 00:14:13,920 success of these products, the deep psychological integration 250 00:14:13,920 --> 00:14:17,000 that ChatGPT really showcased. And this, to me, points to 251 00:14:17,000 --> 00:14:20,320 something that's very durable, more durable than a typical 252 00:14:20,320 --> 00:14:22,920 bubble. So the next six months are going 253 00:14:22,920 --> 00:14:26,920 to test these narratives. And I might turn out to be 254 00:14:26,920 --> 00:14:29,200 completely wrong, but if infrastructure spending 255 00:14:29,200 --> 00:14:33,640 maintains its current trajectory despite these corrections that 256 00:14:33,640 --> 00:14:37,120 we'll probably experience, and if enterprise adoption continues 257 00:14:37,120 --> 00:14:43,360 to grow, and if users keep demonstrating a love and in some 258 00:14:43,440 --> 00:14:47,960 very bad cases, a dependency on these AI tools, we'll know that 259 00:14:47,960 --> 00:14:50,920 we're witnessing a build out disguised as a bubble. 260 00:14:51,120 --> 00:14:54,000 Anyway, I hope you enjoyed today's episode and I'll see you 261 00:14:54,240 --> 00:14:54,840 next week.