Carlson Rebuts Zitron AI Scam Thesis as Enterprise Adoption Surges

企业应用激增，Carlson反驳Zitron的AI骗局论 Zitron称生成式AI是一场循环融资骗局的说法不攻自破，因为数年期的数据中心资本支出不能以过去的营收来衡量，而且数百万付费企业用户正在产生真实的现金流。 Joseph Carlson系统地分析并反驳了科技评论员Ed Zitron广泛传播的AI看空论调，后者认为生成式AI是一个夸大其词、资金内部循环的骗局。在回顾了Zitron好坏参半的预测记录后，Carlson指出Zitron的怀疑态度源于其个人未使用，而非技术现实。Carlson详细阐述了企业（包括他自己的平台Qualtrim）如何在软件工程、司法数据审计、法律研究和医疗领域从AI中获取巨大的切实价值。最后，Carlson驳斥了Zitron的宏观经济学论据，解释了为何将数年期的基础设施资本支出与早期的软件营收进行对比在经济学上存在缺陷，并通过指出真实的企业采纳以及数以亿计的付费终端用户，证伪了“循环融资”的说法。 科技评论员Ed Zitron将生成式AI定性为欺诈骗局，掀起了一股市场怀疑浪潮，然而他的论据建立在现代AI几乎没有现实实用价值的主张之上——这一前提与激增的企业应用背道而驰。 (1:52) 将数年累计的数据中心资本支出与单一年度已确认的软件营收进行对比的看空批评，依赖于存在缺陷的经济学逻辑，忽视了大型基础设施投资可在数十年内进行摊销并产生现金流的事实。 (31:49) 认为AI增长只是科技巨头之间资金来回流转的人为闭环的主张，未能考虑到终端用户的商业化变现，其中包括超过10亿的ChatGPT月活跃用户和数百万付费企业账户。 (38:03) 超大市值科技公司正从机器学习中获得巨额间接收入，例如Meta利用AI推荐算法在广告追踪面临结构性阻力的情况下，仍推动了约30%的营业收入增长。 (35:10)

Joseph Carlson challenges tech commentator Ed Zitron's viral argument that generative artificial intelligence is an orchestrated corporate con, contending that Zitron fundamentally misunderstands the technology's real-world operational utility and misrepresents long-term capital investments as an unsustainable bubble. Zitron's Bearish Thesis and Historical Track Record (3:34) Tech commentator Ed Zitron garnered widespread attention after a podcast appearance drew five million views in four days, asserting that generative AI is fundamentally a con and an overpromised, unreliable product pushed non-consensually by tech executives. Zitron argues that AI tools are unwieldy, micromanagement-heavy software that fail to perform basic tasks reliably. However, Carlson notes that while Zitron accurately forecasted the 2021–2022 collapses of Peloton, Web3 economics, and cryptocurrency exchange failures like FTX, his predictive record deteriorated significantly from 2023 onward. Zitron's subsequent misses included claiming Mark Zuckerberg would destroy Meta, predicting crypto would never experience another bull run, asserting Reddit's IPO would trade below offer price, dismissing Anthropic's projected $9 billion run rate as false, and forecasting in January 2026 that Nvidia could lose 40% to 90% of its revenue—shortly before Nvidia reported 106% year-over-year revenue growth to $96.2 billion. Real-World Adoption Contradicts Claims of Inutility (17:31) Carlson identifies Zitron's dismissal of AI's core usefulness as the primary structural flaw in the bear case. In his interview, Zitron acknowledged he does not personally use AI, relying instead on two isolated negative experiences—a single error in a financial model and spending 30 minutes troubleshooting a Minecraft mod—to generalize that the entire sector lacks value. Zitron further asserted that AI cannot perform basic search, which Carlson refutes by citing ChatGPT's real-time search APIs and Google AI Overviews serving over one billion users daily. Contrasting Zitron's anecdotes with commercial realities, Carlson details how enterprises and his own small business, Qualtrim, rely on AI as an indispensable operational partner. AI tools handle software refactoring, automated testing, customer support ticket triage, and data auditing across tens of thousands of transaction records to produce CPA-reviewed financial statements. Across the broader economy, generative models drive ambient clinical documentation in healthcare, dynamic tutoring on platforms like Duolingo, molecular prediction in biotechnology, and document discovery in legal practice. Carlson emphasizes that enterprises expand software budgets not out of corporate charity, but to capture tangible margin improvements. Economic Flaws in Comparing CapEx to Revenue (31:49) A central component of Zitron's critique contrasts the tech sector's cumulative, multi-year infrastructure commitments—exceeding $1 trillion, exemplified by the multi-gigawatt Stargate Abilene data center project—against modest immediate software revenues. Zitron highlights that outside of OpenAI and Anthropic, direct industry AI software revenue stands at roughly $22 billion. Carlson describes this comparison as economically flawed, explaining that capital expenditure (CapEx) represents upfront assets meant to compound cash flows over a decade or more, akin to Amazon constructing logistics fulfillment networks, Costco building retail warehouses, or Texas Roadhouse investing $10 million into a single location that generates $150,000 per week. Furthermore, Carlson points out that AI delivers immense economic returns that do not appear as dedicated AI software line items. Following Apple's tracking restrictions, Meta deployed advanced machine learning models directly into its recommendation systems and ad-targeting engines. Rather than raising prices, Meta adapted its algorithms to fuel roughly 30% year-over-year revenue growth, demonstrating direct commercial returns from AI infrastructure spending. Dismantling the Closed-Loop Financing Narrative (38:03) Zitron's third pillar alleges a system of "circular financing," arguing that tech giants like Amazon, Microsoft, and Google pass capital to startups like OpenAI and Anthropic, which in turn account for an alleged 70% of cloud AI revenue, creating the illusion of organic demand. Carlson argues that this closed-loop narrative collapses because it completely ignores end-user economics. A circular financing loop exists only if capital never leaves corporate balance sheets. In reality, real revenue flows inward from hundreds of thousands of paying enterprise accounts and over one billion monthly active users on ChatGPT, proving that end-user monetization supports cloud infrastructure consumption. Carlson concludes that while alarming, conspiratorial warnings reliably attract viral traffic and media attention in the short term, real wealth is built by recognizing technological productivity shifts. The source maintains that artificial intelligence represents the most significant general-purpose software evolution since the internet, powered by non-deterministic reasoning capabilities that legacy programs cannot duplicate.

[1:52] Ed Zitron: generative AI is at its heart a con [1:52] Ed Zitron: Well, what do you call something where from the very beginning they've sold it in the terms of magic, but it's just a half-ass archery machine? They are misleading the entire world... the fact that this is happening is insane and the fact it's not a scandal is insane. And I've been in the tech industry for 16 years now and I love technology and I'm enthusiastic about it, but I don't like being misled. And this is the largest non-consensual push of technology in history. [10:15] Ed Zitron: Oh, I couldn't possibly say. I'm too shy. [12:35] Joseph Carlson: Ed does not understand AI. He simply just doesn't understand it [12:35] Ed Zitron: There's some value, but... they have spent over a trillion dollars in capex. [15:12] Ed Zitron: I really don't. [15:12] Ed Zitron: it still took me half an hour and kept getting things wrong. [17:31] Joseph Carlson: We don't even need to wait. This is not something that I need to wait 5 years to prove. We can prove this right now. [19:52] Joseph Carlson: It's not because they want to give away money. Businesses are all about saving margins... That's capitalism to try to make as much money as possible. [19:52] Joseph Carlson: the most useful product that has come along for businesses in decades, likely since the internet itself. [29:06] Ed Zitron: Everyone is telling me it's the best thing since sliced bread. It can't even do the basics. It can't even do search well... [40:20] Joseph Carlson: In the short term, the pessimistic bears get the attention. In the long term, the optimists make the money.

[0:00] Today on the Joseph Carlson Show, at any [0:02] given time during an investing career, [0:04] there always seems to be the bear, the [0:07] one prominent figure that's brave enough [0:09] to go against the crowd to say, "All of [0:11] you investors investing in this thing, [0:14] all of you are wrong, and I know [0:16] better." And they wave that red flag, [0:18] warning investors of the impending doom [0:20] ahead. We've seen the bears before. In [0:22] fact, there's many of them that are [0:23] still prominent today. We have Michael [0:25] Bur, the one who sniffed out the 2007 [0:28] mortgage crisis by looking through [0:29] mortgage back securities line by line. [0:31] He still remains one of the most [0:32] prominent bears today with one of the [0:34] most popular financial substacks. Then [0:36] of course we have Jeremy Grantham who [0:37] hasn't called any specific financial [0:39] crisis like Michael Bur, but Jeremy [0:41] Grantham has been more focused in his [0:42] bearish commentary on overall valuations [0:45] market cycles. He has what he refers to [0:47] as super bubbles or two sigma bubbles [0:51] where he believes that valuations get so [0:52] far above their historical norm that [0:55] everything is going to collapse. Jeremy [0:57] Grantham has made massive CNBC [0:59] appearances that garner a lot of [1:00] attention. We have people like Ray Dio [1:02] that manage massive amounts of money [1:03] that write constant opeds about the big [1:05] downfall of America, the super cycle, [1:07] the credit problems. He's also been [1:09] calling for doomsday for some time now. [1:12] And there's more of them to be sure. But [1:14] in this competition of who can be the [1:16] biggest bear, it seems like there's a [1:18] new bear on the block. He has entered [1:20] into the competition waving a critical [1:22] red flag. And his message is much more [1:25] focused and pronounced than bears of the [1:27] past. Rather than talking about general [1:29] valuations or super cycles or economic [1:32] cycles of the United States, no, he's [1:34] focused specifically on AI. He is the AI [1:38] bear. And his message and his rank [1:40] amongst the bears has risen incredibly [1:42] fast. He is now basically the general of [1:45] the bears, the one leading the pack, the [1:47] one garnering the most attention today, [1:50] even more so today than Michael Bur. [1:52] Normally, I don't mean to address every [1:54] single bear or every single bear [1:55] argument as many of them are just trying [1:57] to gain attention. But in this case, Mr. [1:59] Ed Zitron has already accomplished that [2:02] goal. In fact, he's gained so much [2:03] attention, 5 million views in 4 days, [2:06] that he is now one of the most notable [2:08] people in all of the financial markets. [2:10] He is likely being watched more than any [2:12] other individual. I don't know of anyone [2:15] garnering this much attention in [2:16] financial media today. And part of the [2:18] reason why, of course, is because many [2:20] big claims are being made. I think [2:22] generative AI is at its heart con. And [2:25] seeing these ultra rich, ultra powerful [2:27] people lie through the table. Turns my [2:30] stomach. The word con is a strong word. [2:32] Well, what do you call something where [2:33] from the very beginning they've sold it [2:35] in the terms [music] of magic, but it's [2:37] just a halfass arcery machine? They are [2:38] misleading the entire world. [2:40] >> You are the first person that I've [2:41] spoken to that has that opinion. [2:42] >> Well, the fact that this is happening is [2:44] insane and the fact it's not a scandal [2:46] is insane. And I've been in the tech [2:48] industry for 16 years now and I love [2:50] technology and I'm enthusiastic about [2:52] it, but I don't like being misled. And [2:54] this is the largest non-consensual push [2:57] of technology in history. Now, in this 2 [2:59] and 1 half hour podcast, Ed makes a [3:01] series of massive claims regarding AI. [3:04] Not only calling it a fraud, but many [3:05] other huge claims throughout this video. [3:08] And as I've looked at this, I have not [3:10] seen anybody offer a sound rebuttal. I [3:12] haven't seen anybody go point by point [3:14] giving the truth about this. And as this [3:17] video has 5 million views, I think [3:18] that's overdue. So, what I plan to do is [3:21] go point by point through this video, [3:23] address Ed's strongest arguments that he [3:25] makes, and share a perspective that I [3:27] believe is the truth because I believe a [3:28] lot of people, in fact millions here, [3:30] are being misled by a lot of the things [3:32] Ed is saying. So, let's go ahead and [3:34] jump in. Now, before jumping into this [3:35] interview and going point by point with [3:37] my rebuttal to his arguments, I think [3:39] it's good to first take a look at who Ed [3:40] is, how he became so popular so quickly. [3:43] I don't know much about him, but I do [3:46] know that Reddit surfaced a list of his [3:48] predictions of the past. So, this is not [3:50] the first group of predictions he's [3:52] made. He's been making predictions for [3:53] year, and this is the collection of Ed [3:55] Zitron's predictions. Now, looking at [3:57] this list of predictions that was [3:59] compiled by some Redditor, it became [4:01] clear to me that every single prediction [4:03] on this list was wrong. Every one of [4:05] them was incredibly wrong. And this went [4:08] viral. It got spread around and a lot of [4:10] people saying, "Look, Ed gets everything [4:12] wrong. Why should we listen to him?" But [4:14] usually, usually bears don't get [4:16] everything wrong. And it was a little [4:18] bit skeptical to me that Ed would [4:20] literally get every prediction he makes [4:22] completely wrong. Even bears that get [4:24] the majority of things wrong, like a [4:25] broken clock, they're usually right [4:26] twice a day. They get some things right [4:29] eventually just by the law of average. [4:31] If you make enough guesses, some of them [4:32] will be correct. But in this case, none [4:34] of these were correct. So, I went and [4:36] did my own research. To treat Ed fairly, [4:38] I tried to look at all of his [4:40] predictions and not just cherrypick the [4:42] ones that he got wrong. And I think [4:44] these are worth looking over. So this is [4:46] a more holistic comprehensive group of [4:49] Ed's predictions, both the ones that he [4:51] got right and the ones that he got [4:53] wrong. Well, back in 2021, this was kind [4:54] of during the co era, he said Pelaton [4:57] would get hammered when people returned [4:59] to normal life. And that of course is a [5:01] great call. Pelaton went down 95%. So he [5:04] made a great call there, although one [5:05] that I think was mostly obvious to most [5:08] people. He also said in 2021 to 2022 [5:10] that web 3 and NFT economics were mostly [5:13] unsustainable. Now, this one was also [5:15] very correct, even though I would say [5:18] again this was somewhat of an an obvious [5:20] call. Now, in 2022, I believe this is [5:22] his most important and his best call to [5:24] his credibility. He said a major crypto [5:26] exchange would be discovered insolvent [5:29] and withdrawals would trigger contagion. [5:31] Now, he didn't say exactly which one, [5:34] but he just said one of these major [5:35] crypto exchanges was going to go belly [5:37] up and FTX collapsed almost exactly this [5:40] way 3 months later. He also said in 2022 [5:42] that Reality Labs in the metaverse would [5:44] burn enormous amounts of money. And of [5:46] course, this was another good call. Now, [5:48] again, this one is more on the obvious [5:50] side. It was very clear to a lot of [5:52] people back then, me included, that [5:54] Reality Labs was going to burn a lot of [5:56] money. So, up until around 2022, it [5:59] looks like he has a very strong track [6:00] record. All of these are correct. Ed [6:02] seems like he's just saying truths here, [6:04] predicting the future accurately. But [6:06] then, as we move on, things start to [6:08] take a turn. This is where we get into a [6:10] lot of very bad calls by Ed. Back in [6:13] 2023, he said that Zuckerberg would [6:15] ultimately destroy Meta. So Meta instead [6:18] became vastly more profitable and [6:20] valuable. The exact opposite of his call [6:23] that Zuck would destroy Meta. In 2023, [6:26] he said that there would never be [6:27] another crypto bull run after FTX. [6:30] Crypto subsequently had another huge [6:32] bull run. In 2024, he said that Reddit [6:34] was a terrible business and that shares [6:36] will quickly fall below the IPO price. [6:38] Reddit became a major post IPO winner. [6:40] He said that generative AI had [6:42] essentially peaked and it was a dead-end [6:44] technology and then what really happened [6:46] was usage, capabilities, revenue and [6:48] investment subsequently exploded. So far [6:52] he is very very wrong on that call. He [6:55] also said in 2025 that Anthropic's [6:57] projected $9 billion revenue run rate [7:00] was total BS. Anthropic actually reached [7:02] roughly that level. SoftBank couldn't [7:04] fulfill its enormous OpenAI funding [7:07] commitments. SoftBank ultimately funded [7:09] those commitments. And then his most [7:10] recent prediction tracked here is that [7:12] Nvidia could lose 40 to 90% of its [7:14] revenue within a couple quarters. That [7:16] was January 2026. Nvidia reported 106 [7:20] year-over-year revenue growth to 96.2 [7:23] billion. They even guided revenue growth [7:25] next year of 70%. The point here is that [7:27] Ed gets things right. He got a lot of [7:29] calls right back in 2021 and 2022, and [7:32] that likely gave him a lot of confidence [7:33] to continue making calls. But his most [7:36] recent ones, especially regarding AI, [7:37] have not been correct so far. But that [7:40] hasn't stopped Ed from making even [7:41] bigger and bolder calls that are getting [7:43] more attention from now millions of [7:45] people. Now, let's go ahead and dive [7:46] into this viral interview, and we'll [7:48] just start off with his introduction and [7:50] his overview of how he views AI. This is [7:52] what's getting so much attention. I [7:55] think generative AI is at its heart a [7:58] con. I don't think it is sold as honest [8:01] software. I think that they overstate [8:03] both what it can do, what it will do, [8:05] and the underlying financials to the [8:07] point that they are misleading the [8:09] entire world. And they're actively [8:10] exploiting the weaknesses in journalism, [8:12] in our economies, and indeed within the [8:15] responsible parties with sellside [8:17] analysts, governments, and all over the [8:18] shop. [8:18] >> The word con is a strong word. [8:21] >> Yeah. I mean, what do you call something [8:23] where from the very beginning they've [8:25] sold it in the terms of magic as this [8:27] thing that will replace all jobs, that [8:28] will cure cancer, as all of these [8:30] things? And when you look at it, it's [8:32] boring cloud software that's extremely [8:34] expensive and unprofitable and also [8:36] unreliable at its core. [8:38] >> So Ed's description of a con is [8:39] basically when all these different [8:41] groups, they have different incentives [8:43] to kind of play along with this game [8:44] that AI is better than it really is. And [8:46] he describes it as being talked up way [8:48] too much. Like it's magical. It can [8:50] solve all these problems. In reality, [8:51] it's just boring cloud software. And he [8:54] continues building his case against AI. [8:56] >> So that's the funny thing is people say [8:57] he's not going to finance experience. [8:59] So, he's not going to take I've been in [9:00] the tech industry for 15 16 years now in [9:03] PR, but still had practical experience [9:05] and I love technology and I'm [9:07] enthusiastic about it and this thing [9:09] just comes along that everyone is [9:10] telling me is the best thing since [9:11] sliced bread. It can't even do the [9:13] basics. It can't even do search well. [9:15] Whenever you ask an AI person, well, [9:17] what's your setup? They describe this [9:19] PeeWee's Playhouse thing of like, well, [9:20] you got a harness here and you got to [9:21] use the right prompt. Well, you don't [9:23] want to use that prompt. You want to use [9:24] this prompt here with this model, but [9:26] don't use this model for the beginning. [9:28] But at the end, you're going to want to [9:29] use this model. And this is meant to be [9:31] artificial intelligence. It's meant to [9:34] be smart. It's meant to be autonomous. [9:36] It's meant to be something that you set [9:37] and forget. So, a second argument I [9:39] would define is him describing AI as [9:41] being complex. You kind of need a nerd [9:44] out to know how to use it. It's [9:46] unwieldy. You have lots of different [9:47] settings and toggles and use cases. The [9:50] contingencies are being changed all the [9:51] time of how you use it. And that's not [9:54] magical or easy to use. What you'd [9:56] rather see in a magical product, one [9:58] that's really AI, is you just tell it to [10:00] do something and it does it and you [10:01] don't have to overthink it. You don't [10:03] have to set everything up like you're an [10:05] IT admin. So AI is not really as smart [10:07] or as special as it's pointed out to be. [10:10] And this is one of the pillars of his [10:12] arguments that AI really isn't as good [10:14] as it's talked about. The other one has [10:16] to do with the financial incentives and [10:17] the fundings and profitability of the [10:19] major companies behind AI. Well, their [10:22] revenues are not really coming from AI. [10:24] Up until fairly recently, none of their [10:26] revenues were coming from AI. Like [10:27] dribbles a bit. Right now, 70% of all AI [10:31] revenues across those three companies [10:32] are from OpenAI and Anthropic to [10:34] unprofitable, unsustainable companies [10:36] that literally cannot afford to exist [10:38] without these very same companies giving [10:40] them money. Amazon sent $50 billion to [10:44] OpenAI this year. They sent $5 billion [10:46] to Anthropic. Google sent $10 billion to [10:49] Anthropic. And in the next 3 and 1/2 [10:51] years, OpenAI and Anthropic based on [10:53] actual sellside analyst evaluations, [10:56] their estimates that inform whether [10:58] stock is going to go up or down after [10:59] earnings, they are expecting 400 or more [11:02] billion dollar of revenue, 30 or [11:05] something% of cloud growth just from [11:08] these two unprofitable companies that [11:09] will need to be given the money from [11:10] somewhere. And on top of that, these [11:13] companies have such low respect for the [11:16] average investor, for the analyst, for [11:18] everyone really that they don't even [11:20] disclose their AI revenues. The few [11:22] times they dain us worthy, they use [11:24] something called a run rate, an [11:25] annualized run rate, which means well [11:27] nothing. They never define it. It can [11:29] mean month 12. It can mean months 13. It [11:32] can mean last 4 weeks time 13. It's [11:34] different every time and they never [11:36] define it. And then they sometimes just [11:37] don't mention it. So, you've got this [11:39] big thing that is meant to be the [11:40] biggest, most influential change to [11:43] software ever. And whenever you ask them [11:45] about it, when you say, "What? How much [11:46] you making from this?" They go, "Oh, I [11:47] couldn't possibly say. I'm too shy." [11:49] These are public companies, or at least [11:51] the ones that aren't anthropic and open [11:53] AI. When they have good news, they'll [11:55] tell you. And when they don't tell you [11:56] something, well, that actually speaks [11:58] volumes. So, you can see the second leg [12:00] of his argument. The first part is that [12:02] AI overpromises. It's not really [12:04] magical, and in fact, it's difficult to [12:05] use and unreliable. The second part is [12:08] all about the finances. That these [12:10] companies are really trying to make [12:12] their revenue look like more than it [12:13] actually is. That Open AAI and Enthropic [12:15] give you revenue run rates which they [12:17] don't closely define. And that the big [12:19] tech companies which are already [12:20] publicly traded are very discreet with [12:23] what they share. They're opaque and they [12:25] don't really break out how much money [12:26] they're actually making from AI. That is [12:29] a big part of his argument. Now, he has [12:31] many more points that he brings up that [12:33] we'll continue to go through, but I want [12:35] to loop back to the first one. The first [12:37] one being that he claims AI is really [12:39] not that good. And this is the first [12:41] point that I'd like to rebut. I think it [12:44] is actually the biggest flaw in his [12:46] entire argument. And I think this is the [12:48] biggest flaw with Ed's entire bear case [12:50] for AI. Ed does not understand AI. He [12:54] simply just doesn't understand it. And [12:56] that's self-evident in this interview. [12:58] The things that he says in this [12:59] interview itself are immediately [13:01] falsifiable. They can be disproven [13:03] today. His entire argument rests on the [13:06] idea that AI is really not that useful. [13:09] It sounds ridiculous to anybody that [13:11] actually uses AI, but that is the entire [13:13] core foundation of his argument. [13:16] Everything else after AI not being that [13:18] useful is built at top that flawed [13:20] argument. and he continues to expound [13:24] how much he doesn't understand AI [13:26] throughout this interview at multiple [13:28] points. Let's go ahead and take a look [13:29] at a few. The interviewer literally asks [13:31] him, "Have you used AI? Do you not see [13:34] any value in it?" [13:35] >> Used these tools, the AI tools, Gemini, [13:38] Anthropic, Chat, GBT, etc., and you [13:40] found no value in them. [13:42] >> There's some value, but when asked if [13:44] there's any value in AI, he kind of [13:47] shrugs his shoulder, shakes his head, [13:48] and goes, "There's some value. Like [13:51] there's there's a little bit there's [13:53] some some value in AI. I suppose if you [13:56] have if you have to make me admit [13:57] there's a little bit of value. [14:00] This is AI we're talking about. This is [14:02] chatbt Claude and he says there's [14:05] there's some value and I'm not taking [14:07] him out of context. Let's listen to the [14:09] full thing. [14:09] >> Robic chatbt etc. And you found no value [14:13] in them. [14:13] >> There's some value but it's not there's [14:15] they have spent over a trillion dollars [14:17] in capex. See what he does there? He [14:19] deflects the question, which is, "Have [14:21] you personally found any value in these [14:23] products?" And he says, "Well, they're [14:24] spending a lot of money on them." Those [14:26] are two different questions. The [14:27] interviewer didn't ask, "Do you agree [14:28] with the economics behind funding AI?" [14:31] That wasn't the question. And Ed knows [14:33] that's not the question. Ed was asked, [14:35] "Do you see value personally in using AI [14:37] products?" And Ed says there's some [14:39] value in it. And then he deflects and [14:41] shifts the conversation back over to the [14:42] economics behind AI. Now, to anybody [14:44] that actually uses AI, that implements [14:46] it in a business that actually sees the [14:48] power of it and what it can accomplish, [14:50] the suggestion that AI is simply it [14:52] simply has some value, a little bit of [14:54] value, is delusional. It is profoundly [14:58] ignorant. And in fact, the only reason [14:59] that someone could conceivably believe [15:01] that is if they didn't understand what [15:02] AI is and they didn't use it themselves. [15:05] And Ed actually makes it clear [15:07] throughout this interview that he does [15:08] not understand AI's capabilities and he [15:11] doesn't use it himself. [15:12] >> What do you use AI for? generous. I [15:14] really don't. The interviewer asks him, [15:16] "Ed, what do you use AI for?" And he [15:18] says, "I really don't." Now, that raises [15:20] some questions because again, I already [15:22] have the position that there's no way [15:23] you can believe AI is not that useful [15:25] unless you genuinely don't understand AI [15:27] or you don't use it yourself, which if [15:29] you don't, that's fine. You don't have [15:30] to understand the use of every product. [15:32] But Ed here is a big bear. He's on Wall [15:34] Street. He's giving he's giving video [15:35] interviews that have 5 million views. [15:37] And he's doing so on a product that he [15:38] admitted to himself he does not use. And [15:41] in fact, he outlines why he no longer [15:42] really uses AI. So, how do you know it's [15:44] bad? I've used it. I've put it through [15:46] its paces. I've used it to try and do [15:48] financial models and found one error and [15:50] immediately be like, "Ah." So, Ed used [15:52] AI one time to run a financial model. He [15:55] found one error and immediately [15:57] abandoned AI and concluded that it was [15:59] an con, overhyped, and that the product [16:02] is essentially useless. And this is the [16:03] biggest bear on AI today. Now, this is [16:05] not a one-off experience Ed has had with [16:07] AI. It seems that he views it very [16:09] negatively because his personal [16:10] experience with it so far has not been [16:12] good. And he reveals this even more [16:14] later. E said it might have said this [16:15] earlier. So it makes the easy things [16:16] easy, the hard things harder. When you [16:18] know you're doing a really distinct [16:19] small script for something and it can [16:21] plop that out, it's awesome. I used [16:23] Claude the other day for something [16:24] useful. My kid loves Minecraft. I was [16:26] trying to fix a broken mod cuz he loves [16:28] his wither storm. It's awesome. And it [16:31] still took me half an hour and kept [16:32] getting things wrong. See, what Ed's [16:34] doing here is extrapolating his personal [16:36] negative experiences with AI to say that [16:38] the whole thing is really just a scam. [16:40] It's a con. He says that he encountered [16:42] a problem when he was trying to run a [16:44] financial model with Claude. So, [16:45] Claude's not that useful and he doesn't [16:47] use it anymore. He says when he was [16:49] troubleshooting something with his kid [16:50] that it worked, but it took too long and [16:52] it wasn't that useful. And he even boils [16:54] down the use cases to say the AI is [16:56] really only good at things that are easy [16:58] and it makes difficult things way more [17:00] difficult. See, Ed's personal experience [17:03] is what informs the rest of his [17:05] arguments, and that's the reason that I [17:07] stress it. All the stuff about the [17:09] finances, about open AI or anthropic, [17:11] all the stuff about whether or not it'll [17:13] be adopted in society, whether it will [17:15] pay on all of that, all of that [17:17] investment that's going into it, all of [17:18] that is determined really on how useful [17:21] AI is. If Ed is right and AI is not that [17:25] useful, then of course all of this is a [17:27] con. If AI is really just a small thing, [17:30] not that useful in that many [17:31] applications, if it really is just this [17:33] complicated software that makes things [17:35] more complicated and more difficult for [17:36] difficult tasks, then all of this is a [17:39] con. All the big tech companies that [17:40] we're invested in, anthropic and open [17:43] AI, those revenues will collapse [17:45] eventually. The problem with Ed's [17:47] argument here is that he is profoundly [17:49] wrong. This is incredibly, incredibly [17:51] wrong. in his predictions of the things [17:54] that he got right and wrong. Saying that [17:56] AI is not that useful is right now today [17:59] one of the ones he got wrong. We don't [18:01] even need to wait. This is not something [18:02] that I need to wait 5 years to prove. We [18:05] can prove this right now. Ed makes most [18:07] of his judgments about the usefulness of [18:09] AI based on his own personal anecdotal [18:11] experience. So, let me first before [18:13] broadening this out, just go with my [18:16] personal anecdotal experience with AI [18:18] and go through a couple of ways that I'm [18:19] using AI currently in my business and in [18:22] my process. Now, in Qualrum, I use AI [18:24] all the time. Not just to vibe code the [18:26] app by entering in a couple sentences. [18:28] It's more complex than that, but to look [18:30] at industry standards, to look at how [18:31] certain user interface designs are [18:33] supposed to work, to look at how [18:35] different buttons and messages and [18:36] haptics, all of that works together. to [18:38] look at the chart behavior, workflows, [18:40] all of the debugging. A lot of that is [18:43] handled by AI today. It improves the [18:45] product dramatically. And any developer [18:47] that that works in the industry today [18:49] will tell you how much AI is being [18:50] implemented into the workflow. Debugging [18:53] technical issues, currency conversion [18:55] issues, chart issues, data issues. [18:57] Adding a layer of AI to be able to [18:59] validate all the data has been [19:00] tremendously helpful reducing those [19:02] issues. In customer support, AI can read [19:04] and categorize and prioritize based on [19:06] the problem. It can automatically sort [19:09] different tickets by severity. It can [19:11] make it so that we can have a one [19:12] customer support team do the work of [19:14] five. It is dramatically more effective [19:16] in content production. Now, I never have [19:19] AI give me scripts to read. I never read [19:21] scripts to begin with, but in [19:22] researching topics, I use AI all the [19:24] time. I look at the bullcase and look at [19:26] the bare case. I research different [19:28] notes about it. A lot of that is getting [19:30] a base level through artificial [19:32] intelligence. And anyone that's not [19:34] using any AI to research any topic, they [19:36] might proudly say that while they get [19:38] left behind. AI makes it dramatically [19:41] easier to learn about a topic at a much [19:42] quicker rate than doing other methods [19:45] than reading through random websites. AI [19:47] can look through hundreds of websites. [19:48] It can think and analyze data and it can [19:50] surface exactly what you're looking for. [19:52] AI is incredibly useful all across [19:54] business administration. Even in a [19:56] business as simple and small as mine, [19:58] where there's only a handful of [19:59] employees, I can bring in all my revenue [20:00] lines. I can connect it with Qualram. I [20:02] can connect it with Patreon, with [20:04] YouTube, with Spotify, with everywhere [20:06] that I earn money. Have it bring in all [20:08] of these thousands of transactions, tens [20:10] of thousands. And AI can look through [20:12] all of it, look for fraudulent [20:13] transactions, duplicate transactions, [20:15] mistakes, and errors. It becomes an [20:17] informed data analyst. Somebody's so [20:19] good at it that I would normally have to [20:21] pay $100,000 a year, but instead, I can [20:23] have an AI do that for me. Outside of [20:25] looking at all the payment issues, it [20:26] can give me flowcharts and [20:28] categorizations, visuals of all my [20:30] finances. It can show me in live time [20:32] how much money my business is making or [20:33] losing. It can also show me red flags [20:36] and give advice and suggestions on [20:37] different things to consider with your [20:39] business. And it does that by analyzing [20:41] tens of thousands of rows of data in [20:43] only a minute. This is incredibly useful [20:46] stuff. The business administration is [20:49] endless with AI. It looks at contractor [20:52] communication. It can inform you on tax [20:54] documents, policy language, compliance. [20:56] I've gone as far as to make custommade [20:58] bookkeeping software for my business [21:00] that specifically addresses all of my [21:03] specific needs and nothing more. And the [21:05] monthly statements that it prints out, [21:07] I've passed by a CPA who said they look [21:09] excellent. And this is a business of one [21:12] YouTuber that has a couple people [21:14] working on it. Imagine the use cases [21:16] that AI has for major corporations. [21:19] Imagine the reasons why they're paying [21:21] billions of dollars a year to use these [21:23] products. It's not because they want to [21:25] give away money. Businesses are all [21:27] about saving margins. They're all about [21:29] spending less to get more efficient. [21:30] That's capitalism to try to make as much [21:32] money as possible. All these [21:34] corporations throughout the United [21:35] States are not implementing AI deeply [21:38] into their processes, embedding it into [21:40] their workflows out of the goodness of [21:42] their hearts. They're not just wanting [21:44] to donate to anthropic or open AI. [21:47] They're not doing it because they want [21:48] to lose money. They're doing it because [21:50] they're finding it so incredibly useful [21:52] every single day that it's worth it. Not [21:54] only is it worth it, but it is the most [21:56] useful product that has come along for [21:58] businesses in decades, likely since the [22:01] internet itself. Now, that's [snorts] [22:03] just me. That's my personal anecdotal [22:05] experience with my small business. And I [22:08] find it extremely useful. In fact, I'd [22:10] say extremely is an understatement. I [22:13] need something that's more of a [22:14] superlative, something that's bigger [22:15] than extremely. And when I look at what [22:18] Ed says, the reasons that he doesn't [22:19] believe it's useful, I've encountered [22:21] those, too. AI has made mistakes before, [22:25] it's getting better all the time. It's [22:27] dramatically better just just where it [22:28] was 6 months ago. But it is true in some [22:31] cases it's made mistakes, but humans [22:33] make mistakes as well. Everybody makes [22:35] mistakes. AI makes mistakes as well, but [22:38] it does a lot of work really quickly. In [22:40] many cases, much faster than humans can. [22:42] And for many tasks, it makes far less [22:44] mistakes than humans. What Ed did here [22:47] is he extrapolated a couple negative [22:49] experiences to now saying a claim that [22:52] AI is overhyped and it's a con. And [22:54] that's just not true. And you don't need [22:56] to rely on my anecdotal experience. This [22:58] is blatantly disprovable today. For [23:01] example, the suggestion that AI is not [23:03] that useful and that it can only do easy [23:05] tasks. One of the things that AI is [23:07] exceptional at is coding. And as a [23:10] previous coder myself, I can assure you [23:12] coding isn't easy. If it was easy, [23:15] everyone could do it. And not many [23:17] people could do coding. Not many people [23:19] could do it well. Coding takes a long [23:21] time to learn. App architecture takes a [23:23] long time to learn. I know Ed probably [23:24] doesn't have an appreciation for that [23:26] because it doesn't sound like he's done [23:27] any coding in the past. But if you have, [23:29] if you built entire apps by coding every [23:31] single line of code, you can appreciate [23:33] how good AI is at it and how very few [23:36] mistakes it makes. AI has revolutionized [23:39] one of the most important fields in the [23:41] world, which is writing software. The [23:44] modern economy, making digital products, [23:47] AI is incredibly good at it. It doesn't [23:50] just code. It debugs. It tests. It [23:53] documents. It refactors. It can jump [23:55] into a massive codebase, explain all the [23:57] issues with it, give you [23:58] recommendations. It can mitigate issues. [24:00] It can check for vulnerabilities. It can [24:02] do rapid prototyping with design. In the [24:05] field of coding and software itself, AI [24:07] is endlessly useful. It is endlessly [24:09] useful. The suggestion that it's it's [24:12] just okay. It's it's a a little [24:14] software, some boring cloud software. It [24:16] is blatantly false. It is incredibly [24:19] misleading. And you may say coding is [24:21] the only thing that AI does really well. [24:22] Well, that would be wrong. AI is rapidly [24:25] going across different categories and [24:26] the use cases are becoming more and more [24:29] seen by different companies. It's [24:30] becoming more and more implemented. For [24:32] example, if you go into customer [24:33] service, AI is already there. It's [24:35] answering calls. And in many cases, you [24:37] have a better experience if AI picks up [24:38] than an actual human. It's more [24:40] attentive. It's quicker at resolving [24:42] issues. It's getting rapidly better. The [24:44] suggestion that, "Oh no, I got an AI bot [24:46] on the phone is no longer that good. [24:49] I've had experiences with an AI bot that [24:51] was far better than talking with a [24:52] human, and it's only getting better." [24:54] The legal field is being revolutionized [24:57] by AI. When you have attorneys that need [24:59] to look up contracts, look up reviews, [25:01] look up red flags, discovery, previous [25:04] cases, all of that background work, AI [25:07] is exceptionally good at it. It can [25:09] summarize things much faster. It can [25:11] read through documentation. There's a [25:13] reason that every lawyer is using AI. [25:15] All of them. Even if they say they [25:17] aren't, they're likely using AI a little [25:19] bit in their practice or they'll soon be [25:21] out of business. The amount of headache [25:22] that AI can save in legal reviews is [25:24] incredible. It's yet another field [25:26] that's being completely revolutionized [25:27] by this technology. In medicine, we have [25:29] increasing amounts of uses of artificial [25:31] intelligence. The most obvious one was [25:33] reading images, looking at scans, and [25:35] scanning for things that maybe the [25:36] radiologists missed. AI is exceptionally [25:39] good at looking at little nuances and [25:40] pictures that humans can't quite pick [25:42] up. Now, there's still radiologists that [25:44] need to review the work, and of course, [25:46] there needs to be the human touch, but [25:47] they can do far more work. They can view [25:49] far more scans and they can do so with [25:51] something else, checking their work, [25:53] making sure they don't miss things, [25:54] which of course will reduce the amount [25:56] of false positives and false negatives, [25:59] reducing the amount of patients that get [26:00] told they don't have cancer when they [26:02] really do have something to treat. This [26:04] is another massive benefit. But it [26:06] doesn't just end at radiologists. You [26:08] also have documentation. You have all [26:09] the business operations behind [26:11] healthcare. All the background stuff, [26:12] the accounting stuff, the legal stuff, [26:14] the HR stuff. All of that again is being [26:17] implemented and proved by AI. You have [26:19] doctors that can now bring a simple [26:21] recording device in a meeting with a [26:23] patient. The recording device records [26:25] the entire transcript and then [26:26] automatically enters in those patient [26:28] notes highly categorized. They [26:30] previously had to hire scribes to do [26:31] this in hospitals. People to take notes [26:33] of every single meeting. That was [26:35] literally their job. Now scribes still [26:38] exist, but again this makes it so they [26:39] can do much more. They can focus on [26:41] categorizing and not just listening to [26:43] every single meeting. In finance, the [26:45] use cases of AI are becoming more and [26:47] more prominent and they are again [26:49] virtually endless. With reconciliation, [26:51] much easier to do with AI, with [26:53] financial modeling, much easier to do [26:55] with AI. With fraud detection, much [26:57] easier to do with AI. You can do invoice [26:59] processing, financial analysis, audit [27:02] work, reporting, and so on. These [27:04] companies are implementing AI into every [27:06] facet of their business because it [27:07] speeds up so many processes. It makes [27:09] these companies dramatically more [27:11] efficient. With education, the use cases [27:13] of AI are never ending. Of course, you [27:15] can now have one-on-one AI tutors. [27:17] That's exactly what Dualingo is doing. A [27:19] company that it took them years to build [27:21] course content to learn languages. Now, [27:23] they can build course content much [27:25] faster. Engagement is picking up. More [27:27] and more users are using the app [27:28] specifically because they're building [27:30] more content. They're making it so that [27:31] there's more step-by-step guides. And [27:33] this isn't just with Dualingo. It's [27:35] across the entire education field. [27:37] They're implementing AI not only into [27:39] the administrative part and the back-end [27:40] part, but also the teaching part. AI can [27:43] give you dynamic and immediate feedback [27:45] when you make mistakes. Unlike [27:47] deterministic logic, AI is already being [27:49] used across science and biotech and [27:51] protein design, molecular prediction, [27:53] synthesizing previous research, being [27:55] able to have more predictive outcomes in [27:56] marketing and media. It is incredibly [27:58] useful. You can make drafts, copies, [28:01] images, you can make designs immediately [28:04] with these companies. It has [28:06] revolutionized this industry. In fact, [28:07] in fact, it makes marketing and media [28:08] much easier for businesses to launch [28:10] products because they don't have to [28:11] spend as much time with a full marketing [28:13] team. With managing sales, AI is [28:15] incredibly useful. Used across the CRM, [28:18] prospect research, outreach to [28:20] customers, managing relationships. The [28:22] non-deterministic nature of sales is [28:24] exactly why AI is so useful in those [28:26] cases. You can see why Salesforce, a [28:29] company that has been in sales forever, [28:31] just partnered with Anthropic. And it's [28:33] making it so that they can crawl through [28:35] their data and do all of these valuable [28:37] things. They're not doing that because [28:39] they want to. They're doing it out of [28:41] necessity. Salesforce knows that they [28:43] need to have the best for their [28:44] customers. They need to have AI [28:46] integrated. And we could continue on [28:48] across all different verticals, all [28:49] different fields across the entire [28:51] economy from engineering to logistics to [28:53] cyber security to so on. If there is a [28:55] category of work, AI will impact it. And [28:59] this isn't just me saying so. This isn't [29:02] just a narrative that I'm building. It's [29:04] also proven by the numbers today. The [29:06] reason that Anthropic and Claude are [29:08] growing so fast is because customers are [29:11] using their product. The reason that [29:13] customers are using their product is not [29:15] because they believe it's a con or [29:17] overhyped or they don't find any value. [29:19] If that was the case, Ed, then they [29:21] would have used the product for a while [29:23] and like you concluded it was mostly [29:25] hype and not worth it and they would [29:27] have discarded it. They wouldn't have [29:29] kept using the product and kept using [29:31] more of it and kept finding every single [29:33] week additional ways to implement AI. [29:35] That's not what you do for useless [29:36] products. You considered it useless. You [29:39] used it once to do analysis and a couple [29:41] things for your kids uh Roblox account [29:43] or whatever it was. You decided it [29:45] wasn't really worth the hassle. And so [29:47] now you're saying that it's essentially [29:48] overhyped and it's all a con. That's not [29:51] the normal experience that businesses [29:52] have. Normally they use AI, they have [29:55] some headaches getting it implemented at [29:57] first, but then they have some factor [29:59] that wows them. There's some feature [30:00] that does something so much better than [30:02] what they expected that they start to [30:04] use it more and more. And we're seeing [30:05] that throughout the growth rates of [30:06] anthropic and open AI. The reason that I [30:09] stress the usefulness of AI again is [30:11] because everything that Ed says [30:13] throughout this entire two-hour [30:14] interview, all of it is built on a [30:16] foundation that AI is not useful. That [30:19] is his argument. It's not really useful. [30:22] It It's got some uses, but not much. And [30:25] it's not that good, and it can't do [30:26] this, and it can't do that. And it's [30:28] overpromised. It's overhyped. That's the [30:30] core of his argument. Everything else, [30:32] all the finances, builds upon that. And [30:34] Ed is just profoundly wrong on this [30:37] topic. He's so wrong that he doesn't [30:38] even understand the capabilities of AI [30:40] today. For example, in one part of this, [30:42] he says that AI is not even good at [30:44] search. [30:44] >> Everyone is telling me it's the best [30:45] thing since sliced bread. It can't even [30:47] do the basics. It can't even do search [30:49] well whenever you [30:50] >> AI can absolutely do search and it can [30:52] do it very well. Chat GBT has an entire [30:55] API that's dedicated to live search [30:58] data. The second you call this API, [31:00] it'll search. It'll search across all [31:02] different search engines, hundreds of [31:04] different results, and it will give you [31:06] live data that second. Google has a [31:08] billion people using AI overviews, which [31:10] is live AI search every single day. And [31:13] they noted that people that use AI [31:15] search use it more, more frequent, more [31:18] frequent searches, more frequent [31:19] follow-ups. They like the product so [31:21] much that they actually get more use out [31:22] of it and they use it more. It is [31:24] directly contrary to your claim that AI [31:26] can't do search. Now, ironically, Ed [31:29] believing that AI is not that useful is [31:31] something that is not a new belief. [31:32] Again, when we look back at his [31:34] predictions back in 2024, he made the [31:36] assertion in the same prediction that [31:38] generative AI had essentially peaked was [31:40] a deadend technology. Ed has continually [31:43] underestimated the use of AI, and he's [31:45] doing so today in a strikingly [31:47] falsifiable way. Now, you may believe [31:49] that Ed sees the different use cases of [31:51] AI and doesn't think they're all that [31:53] good. And this is the trick that Ed [31:55] tries to play. He tries to say even [31:57] though AI does have all these use cases, [31:59] it's not improving the numbers. These [32:01] companies really aren't doing much [32:02] better because of AI. And he does this [32:05] with a lot of subtle tricks. There's a [32:07] lot of tricks throughout this interview [32:08] and I want to highlight a couple of them [32:09] here. One of them is how he does [32:11] comparisons of the capex spend to the AI [32:14] revenue. Let's go ahead and just outline [32:16] what he does here. Need a bunch of [32:18] power. So an example, OpenAI and Oracle [32:20] are building a data center in Texas in [32:22] Abalene, Texas. 1.2 in 2 GW called [32:25] Stargate Abalene. Within that, with each [32:28] one of the eight buildings, there'll be [32:29] 50,000 Nvidia GB200 GPUs. So, city of [32:34] Bristol takes about 7800 megawatt of [32:38] power a year, right? Well, Stargate [32:41] Abene is condensing more power than [32:43] that, 1.2 gawatt into a space around [32:45] 1,172 [32:47] times smaller. City of Bristol is about [32:49] 1.2 billion square ft. Star Elyn is [32:52] about 998,000. [32:54] So you're condensing all of this power, [32:56] all of this money, all of this labor [32:57] into this one spot. And all of these [33:00] data centers cost billions of dollars. [33:02] All of these companies other than [33:03] Microsoft have now to take out debt. And [33:06] the thing is they've spent over a [33:07] trillion dollars so far and they want to [33:08] spend another trillion dollars next [33:10] year. And for what? To make tens of [33:12] billions of dollars, most of which comes [33:14] from two unprofitable companies, [33:16] Anthropic and Open AI. He always says [33:18] one number which is the trillions and [33:20] trillions of dollars these companies are [33:21] spending and then he compares that to [33:23] the tens of billions of dollars of AI [33:25] revenue today. And this is the [33:27] juxtaposition that helps his argument [33:28] but it's also very misleading. See what [33:30] Ed is doing here is he's taking the [33:32] commulative cost outlined for years and [33:35] years of AI construction. This is the [33:37] capex bill that these companies are [33:38] outlining for for five plus years [33:41] they're going to be spending. That's the [33:42] trillions of dollars. And then he's [33:44] comparing that to the current AI [33:46] revenue. But that's not a fair [33:48] comparison, Ed, because these companies [33:50] aren't going to earn money for AI just [33:52] this year. The the way things work is [33:54] you you spend a lot of money upfront to [33:56] build an investment and then that [33:58] investment can earn you returns for [33:59] multiple years. You can take the example [34:01] of an Amazon warehouse. Amazon pays [34:04] billions of dollars upfront to build a [34:06] warehouse and then they earn money from [34:08] that warehouse for years and years and [34:10] years down the road. Amazon doesn't have [34:12] to keep paying for the cement, the land, [34:14] the steel, everything every single year. [34:17] No, they have to pay some costs [34:18] continually, but the majority of them go [34:20] down dramatically as they earn continued [34:22] returns. It's the same thing with a [34:24] Costco warehouse. Costco spends a couple [34:26] billion dollars to build a warehouse to [34:28] buy the land. It's a lot of [34:29] construction, a lot of expense, and then [34:31] they gain income from all their members [34:33] that shop there for the next 10 to 20 [34:35] years. See, this is the game he plays. [34:38] He compares the upfront massive expense [34:41] of the capex buildout to the current AI [34:44] revenue today. Here's another example of [34:46] Ed playing the same game with these [34:48] false comparisons. We have the apex [34:50] predator of cloud software, Microsoft. [34:54] They can only get singledigit billions [34:56] from selling AI software. And Christ [34:59] almighty, outside of OpenAI and [35:00] Anthropic, we barely get $22 billion. [35:02] And the thing is, $22 billion is a large [35:05] amount to you and me. It's not a large [35:07] amount of money when you spent a [35:08] trillion plus dollars when you have [35:10] anthropic and open AI with $1.1 trillion [35:12] worth of cloud commitments. That's the [35:15] comparison. The amount of money they're [35:17] spending to build infrastructure that [35:18] will serve them for half a decade. And [35:20] he's comparing that to the line item [35:22] revenue of these companies AI revenues [35:24] today. So this is a strikingly false [35:28] comparison that he continually makes [35:29] across all of his appearances to make [35:31] the false impression that AI is in some [35:33] massive bubble. But this isn't even [35:35] theoretically how any investment works. [35:37] This is like saying, "I can't believe [35:38] Texas Roadhouse pays $10 million to open [35:41] up a new restaurant when they only earn [35:44] $150,000 a week per location." $150,000 [35:49] against 10 million. Jeez, this is such a [35:52] waste. Well, those are the actual [35:54] numbers, but Texas Roadhouse is an [35:56] economically profitable business. He is [35:58] placing two things together, hoping to [36:00] trick people that are financially [36:02] illiterate enough to believe him. Now, [36:03] even when presented with examples of how [36:06] AI is impacting businesses positively, [36:08] when they say that AI is leading to [36:10] faster growth, more efficiency, more [36:12] customers, a better product overall, all [36:15] of the good stuff that the CEOs and [36:16] executives will outline, Ed, of course, [36:18] is very skeptical and he believes that [36:19] these companies are growing for [36:21] different reasons. Their AI bets have [36:23] paid off. Wow, their AI bets have paid [36:25] off. as these companies refused to say [36:26] how much they're making from AI, but [36:28] because their existing businesses [36:30] continued to grow and did so, by the [36:32] way, through price increases, changes to [36:34] how Google and Meta uh did advertising, [36:37] Amazon bumped up prices and changed how [36:39] they did actually Amazon started a [36:41] remarkable ad business during this whole [36:43] time as well and the selling through [36:44] Amazon platform. Anyway, nothing to do [36:46] with AI, but because number go up, [36:48] because revenue go up, everyone went AI [36:50] because these companies wouldn't spend a [36:51] trillion dollars for for no reason, [36:54] right? So even when we have companies [36:56] that say they're growing in part in a [36:58] large part because of AI, he says no, [37:00] they're not. They're growing because of [37:01] old-fashioned price increases, if it was [37:03] simply price increases that led to the [37:05] growth of these companies and the [37:06] success of them, then how did Meta [37:09] become successful? Meta is a free [37:10] product. They didn't increase prices on [37:12] anything. Yet, their company grew 30% [37:15] revenue year-over-year. It's one of the [37:17] fastest growing, largest companies in [37:18] the world. Meta actually had a lot of [37:20] disadvantages leading into this growth. [37:23] Apple cut off their tracking, which was [37:25] supposed to be a huge blow to Meta. It [37:28] temporarily was until it wasn't because [37:30] they found this magical tool that all of [37:32] a sudden solved that problem. A tool [37:35] that could use more nuanced ways of [37:37] measuring what people liked. Meta [37:39] implemented AI directly into their [37:41] social graphs and it led to substantial [37:44] revenue growth, tens of billions of [37:46] dollars of their core product. Meta [37:47] doesn't need to break out an AI revenue [37:50] line item. You can simply look at the [37:52] core product. itself and how much it's [37:53] being aided by AI. And Meta is not the [37:55] only case. The proof of AI helping [37:57] businesses out is so incredibly [37:59] apparent. It's so indisputable, yet he [38:01] continues to dispute it. Now, after [38:03] addressing how Ed doesn't use AI [38:05] himself, doesn't seem to understand or [38:06] appreciate the usefulness of it across [38:08] different industries, and after [38:10] addressing how he inaccurately or [38:12] falsely portrays AI as not having a [38:15] return potential because he's comparing [38:17] commulative long-term spend on capex [38:19] compared to today's revenue. I want to [38:21] finally address what I believe is the [38:22] third most important point and it's the [38:24] argument of circular financing that [38:27] there's not really natural organic [38:29] demand for AI. It's all just a bunch of [38:32] companies handing money over to each [38:33] other taking money out of one pocket and [38:35] putting it into the next. 70% or more of [38:38] all that demand comes from these two [38:40] companies who are funded by these three [38:42] companies. And that's the funny thing. [38:45] The reason that they don't want to break [38:46] out their AI revenues is because it [38:48] would become alarmingly obvious that [38:51] this was the case. It turns out that the [38:53] only real big customers because it's not [38:55] like they're building a few data [38:57] centers. They're building trillion plus [38:59] revenue potential. They believe they'll [39:01] get speculative. It's entirely [39:03] speculative. They're building it because [39:05] they saw the biggest companies in the [39:07] world buy a bunch of GPUs and they said, [39:08] "I want in on that." They must have [39:10] diverse customers, right? they wouldn't [39:12] just have two unprofitable fail sons [39:14] that they're propping up with Christ, [39:16] they've raised $217 billion just in [39:19] 2026. [39:20] >> Now, even when you look at the way that [39:22] he illustrates this, it's with all these [39:24] different companies and he just talks [39:25] about how the the money's going between [39:27] hands. All of it is just kind of [39:29] changing hands from one company to the [39:30] next. And the party that he leaves out [39:32] intentionally is the customers, the [39:35] actual end users of AI. The reason that [39:38] Ed seemingly forgets to mention them is [39:41] because it completely defeats the point [39:42] that he's making. The number of actual [39:44] customers funding money into Claude and [39:46] OpenAI is staggering. They have millions [39:49] of businesses using their product, [39:51] millions of businesses. We have Cloud [39:53] that has hundreds of thousands of [39:54] businesses using it more and more, more [39:56] and more spending every single year. And [39:58] we have billions of end users. CatchBT [40:01] has over a billion monthly active users. [40:03] Even their ad business, which they just [40:05] launched, is growing at a blisteringly [40:07] fast pace. See, the problem with the [40:09] circular financing argument is that only [40:11] works if it's really companies just [40:13] moving money around between their own [40:14] hands. But that's not what's happening [40:16] here. There are billions of customers [40:18] that are adding to the revenue of these [40:20] products. So, when we look overall at [40:21] Ed's claims here, there is a reason that [40:24] they gain so much exposure. It's because [40:26] they're very big claims. That all of [40:28] this is a con. is because it fits with [40:30] the same trope that all these hands are [40:32] working together to make this whole [40:34] thing boil up in a way that it doesn't [40:36] deserve. The real truth is much more [40:38] boring. It's much more mundane. AI is [40:41] just an incredibly useful technology, [40:43] the most useful since the internet. And [40:45] lots of businesses are finding a lot of [40:46] uses with it. So, they're paying for it. [40:49] And they're not stopping. As it is [40:51] proven to be useful and indispensable [40:53] because of its ability to have [40:54] non-deterministic logic, companies are [40:57] going to continue using it. in more and [40:59] more diverse ways. That's a more boring [41:01] truth because there's no big conspiracy [41:03] here. Now, the boring truth, [41:04] unfortunately, does not get as many [41:06] views. It has been proven throughout [41:08] time that the bears get the most [41:09] attention. It is over and over again [41:11] that we see the ones raising the big red [41:13] flags always garner the most clicks, the [41:15] most attention, and this is no [41:17] exception. Even as faulty as this bear [41:19] case is, it's garnered more attention [41:21] than any mundane bull case. And that [41:23] just goes back to the fact that in the [41:25] short term, the pessimistic bears get [41:27] the attention. In the long term, the [41:29] optimists make the money. That's going [41:31] to be it for this episode. Hope you [41:32] enjoyed. See you next one.