← Digital Disruption with Geoff Nielson
Digital Disruption with Geoff Nielson
The AI Market Must Crash: Ed Zitron on Why the Bubble Will Burst
00:58:52
Loading summary
Transcript44 lines
- [00:00]
A
Hey, everyone. I'm super excited to be sitting down with Ed Zittran. Ed is the host of the Better Offline podcast and a leading critic of the current AI Boom. What I love about Ed is not just the passion he brings to his podcast, but that he always has the facts and the data to back up his skepticism and has a damn solid case that we should take everything we hear about AI with a grain of salt. I want to know what he thinks AI is good for and what's just bs and maybe most of all, what will happen if he's right and this stuff can't deliver. And what should we be doing to prepare? Let's find out. Ed, super happy to have you on the podcast today. You know, you're obviously the host of Better Offline. You know, certainly one of my favorite AI skepticism podcasts, if I can call it that. And, you know, for those who haven't seen it, maybe I'll. I'll start off by, you know, know, sharing one of the quotes I heard you say recently that I think sort of sums up the whole thing, which was you were talking about the AI revolution, and you said the thing holding back AI is it doesn't fucking work. Can you, you know, unpack that statement and, and kind of the components of it?
- [01:08]
B
So if you look at what. What generative AI was meant to be and large language models were meant to stand for, they kind of were always set up to fail. They were meant to be this panacea of we're going to be the future of consumer software, were going to be the thing that kind of restarts growth in software as a service. As I'm sure you all know, software as a service has been slowing since 2021. Actually kind of before that, if I'm honest, people have been freaking out for several years before COVID in fact. But generative AI was meant to be this thing. You plug it into anything and it just creates new revenue. The problem is that generative AI and large language models are inherently limited by the probabilistic nature of these models. What they can actually do is they can generate, they can summarize. You can put hats on a hat. You can say, oh, they could do some coding things, but that's really what they can do. And they have reached a point where they're not what they can't learn because they have no consciousness. So what they can actually do as products is very limited. It's very limited indeed because what people want them to do is they want them to create units of work. They want to create entire software programs. You can't really do that. Oh, can you create some code? You can create some code, but if you don't know how to code, do you really want to trust this? You probably don't. So inherently you've got all of this hundreds of billions of dollars of capex being built to propagate large language models that don't have the demand and don't have the capabilities to actually justify any of it. Right.
- [02:36]
A
And I think a lot of the argument that you're hearing is, well, you know, the next version will get even closer and even closer and you just around the corner from being able to deploy this at scale. I was going to ask you if you buy that. It's, it's pretty clear you don't buy that. But, but you know, tell me, tell me a little bit more about, you know, why you think the inherent nature makes that, you know, decreasingly likely.
- [02:57]
B
Well, I mean we've been told that for several years now. We've been told for years, it's just around the corner, any minute now. It's kind of like waiting for, waiting for a bus, except the bus will never come. More of a waiting for Godo situation. Yeah, everybody's been saying all these things are getting exponentially better. They really haven't. They've been getting more powerful based on benchmarks that are rigged specifically for them because you can't just test them on real things. So everyone's been saying, oh, they'll get more powerful, they get more, they'll get cheaper. And while they've got more powerful, their actual capabilities have mostly stayed the same. Other than reasoning models which have really all kind of a technological hat on a hat, they are, oh, instead of saying you ask it to do something and it just spitting out what it thinks is the answer, it will sit there and generate an action plan and execute that as best it can. The problem is the more complex that reasoning effort, the more likely it is to hallucinate. So the upper bounds of coding LLMs getting better has been at the cost of more token burn because requires more token burn for reasoning models, but more propensity to hallucinate. So basically the more complex the work you ask it to do, the worse it gets. But really it comes down to a far simpler point which is where are the new things? Where are the new products? Where are the new products that actually matter? Microsoft has fa. I just reported this out actually very recently that Microsoft only has 8 million active paying licenses for Microsoft's 365 AI copilot, which is a product category, I think business and productivity is 29 billion and a quarter, something like that. If Microsoft can't do it, nobody can. Microsoft is the software juggernaut. They are the software mob. If they can't work this out, nobody can.
- [04:43]
A
And it's. Copilot's a really interesting example. And, I mean, this is a personal experience. Don't, you know, take it as, you know, the standard truth. But my experience with Copilot has been, like, in a word, underwhelming, I would say. Like, I'm still struggling to figure out what it's actually useful for other than summarizing meeting minutes. And, you know, we put out here a piece of content over a year ago, basically saying, no, Copilot is not ready for prime time. And when we put that out, I mean, I'll be the first to say I. I thought, like, Ready for Primetime was going to be, like, weeks away based on some of the hype. And it still feels like we're not. We're not there. Is that your experience as well? Is that what you're seeing and hearing?
- [05:25]
B
Yes, and talking to a number of Copilot customers as well. People are just underwhelmed. People don't know what to do with it. And I think that that's probably the biggest problem. It's not just that it's underwhelming at the things it's meant to do. It's not really obvious what it is you're buying it for. They just announced a couple weeks back. Oh, they've released Vibe code. Sorry, Vibe. Working for Excel and Word. It's like they're quite literally saying, you work out what it is you're meant to do with this. It's very. It's actually kind of nasty. I don't like it. I don't like the idea of paying for software that I have to work out the value of. That's not. That's not how this works. It's not buying a computer. This is a software product. And I mean, the truth is that LLMs can't do anything that we really need them to in Copilot. Sorry, in Word or in Excel, they can't really do that. I can't even think of what I'd want them to do. Like what. What possible. I'm writing the words, I'm filling in the data in Excel. I guess the only time I've ever found Chat GPT useful was when I asked it how to do an Excel formula. But how would I work this out in Excel and then I went back to Excel and I worked out how to do it from scratch. I wouldn't trust an LLM to do anything with my own data. Not even because I think though, and people really get ahead of themselves on this one. I think it's important to say I don't think Microsoft is trading their models with your copilot data in 365. That's not what I'm scared of. I'm scared of not bloody trusting the numbers. Also, I want a human to look at that. When we hire a person, we're not just hiring them to do a job, we're hiring them so we can kind of distribute the risk well.
- [07:06]
A
And with that in mind, that's one of the promises of AI that is feeling a little bit dubious or at least that there's a lot of different views on is there's this view of oh, you know, there's going to be these, this widespread job loss caused by AI in one camp. There's also, you know, the utopians who say, oh, it's going to create a job, boom. Do you believe either of those or do you have a different perspective?
- [07:27]
B
I think they're both making it up right now. The jobs that this is replacing are translators which have already been. They've been trying to automate them out for years and years and years with machine read translation and SEO people who, I mean for the most part I think that's a job where you already operating on the fringes of what a job it is. And I think there are good SEO people and a lot of bad ones and I think that the good ones will be fine. Because good SEO isn't generating copy, it's understanding how the Internet works. But nevertheless the, the chum level. But I think the only jobs that are being really replaced, our jobs where the boss was already trying to fire people and had already been doing it, I think that on both sides it's a farce. If you use these models or talk to anyone who uses them regularly, you are not coming away with it saying this can replace a person. You are not coming away from it saying from six months ago to here, I am inspired to believe that this can replace a person. Go and sit on any of the subreddits of Replicate or Lovable or Cursor or Claude and you will see that there is no one on there building their own software. Just with LLMs. The people that are able to build functional software with LLMs are people who can already code. And there was a study a couple of months ago and came out in the register where it was people engineers using large language models were actually slower. So it's kind of the whole thing's kind of farcical. And the idea that there'll be more jobs again, that's just the kind of wanky way that people go about it. It's like, oh, there'll be jobs training. The model already exists. They're already jobs training the models. But it's like, oh, they'll. There'll be AI factories. What does that mean? None of these people actually think further than one sentence.
- [09:14]
A
If you work in it, Infotech Research Group is a name you need to know no matter what your needs are. Infotech has you covered. AI strategy, covered. Disaster recovery, covered. Vendor negotiation covered. Infotech supports you with the best practice research and a team of analysts standing by ready to help you tackle your toughest challenges. Check it out at the link below. And don't forget to like and subscribe. You mentioned like the subreddits and the idea of asking people who are actually interacting with these on a daily basis. And this is one of the strange and kind of disturbing things about this to me is it feels like what you hear from CEOs or from business leaders is starting to feel more and more out of sync from what you're hearing from Boots on the Ground. So when you hear some of these clips where a CEO says, oh, well, we're actually doing mass layoffs because we had AI efficiencies, do you just see that as pure opportunism or is there.
- [10:14]
B
More under the COVID It's opportunism. But I challenge anyone who reads any announcement where they said the CEO says, oh, we've replaced AI, We've replaced people with AI, actually read what they say because they never say that. Not once. What they say is all the we are being more efficient and using the power of AI to be more efficient, they're not saying they're replacing anyone other than Marc Benioff, who is just a liar. He's a liar. I rarely just say that people lie, but I think he says 30 to 50% of jobs inside Salesforce are automated or something. Just nonsense. And I've talked to people in Salesforce. They, they back up what I'm saying. And that's the thing. These people, they will say what they need to because they know the people investing in the market. They know the big hedge funds and all that. They know for a fact that those people don't know anything. I don't know Shit about fuck. They just. They're. Everything is a great big vibe for them. And thus people are. I mean, these are just craven, cynical liars. They're not replacing anyone with AI. They want you to believe that because the media will pick it up and run with it.
- [11:17]
A
So let's stick on Benioff for a while. And some very direct words there that we should probably unpack a little bit. So, I mean, Benioff's big thing, whether you believe him or not, is going all in on agentic AI and they're building these products to help people go agentic in your own view. Why is that bs? And assuming that it is bs, how do you see the next handful of months or years unfolding for that organization as they have to cope with not actually having the product or service they've been suggesting they do?
- [11:54]
B
Well, let's start with the most obvious thing, which is that Salesforce's own CFO said back in March that they will not see growth from AI this year. That feels like the biggest indictment of all of it. Second June 26, 2025, Marc Benioff says AI is doing up to 50% of the work at Salesforce. That is a lie. Like that is. That's a lie. Like, I'm sorry, he never has to prove himself. He just says this crap. He is lying. That is a lie. There is no way that that's happening. And to say that I'm actually shocked that there is not regulatory action taking place because that's. That is nonsense. But Agent Force sucks. People don't like it. The information is reported repeatedly how people don't like Agent Force. It doesn't do what they say it needs to. There was an internal Salesforce research paper that came out that said the, the agents fail, I think 52%. They only succeed 52% of the time on single step things like single step tasks. So I asked the agent to do something and then it does something. And they only succeed 32% of the time on multi step. So that means ask agent to do something. Agent books an appointment, Agent goes and looks something up. That's terrible. That's runs completely contrary to agents and Agent Force. And a few weeks, a few weeks ago, they announced something along the lines of, oh, they've got a new internal AI and autonomous coding tool called Vibe. Cody, they're out of idea. Everyone's out of ideas at this point as Vibe this, Vibe that. But if you look in the illustrious history of Salesforce, he's been promising Einstein AI since like 2016. Marc Benioff loves getting on stage and saying that AI is the future, but AI never seems to do anything at Salesforce.
- [13:43]
A
So let's come back to that idea of, you know, agentic is, you know, all hopped up and you know, half or more of these capabilities don't actually exist in your view. We kind of touched on it but like what are the actual reasonable capabilities or use cases we can expect from it? Like it clearly in your mind it's a lot more limited than what it can't do. Like what can you actually expect from this stuff?
- [14:10]
B
Not most of the stuff they're saying. I think the Carl Brown of the Internet of Bugs said it really well. He's a great YouTuber covering 35 years in software engineering makes the easy start easier and the hard stuff harder. So if there are sections of code that you need to very quickly do, if you have a very character heavy coding project. Nick Suresh over Ludic as well said this. If you just need to generate a lot of quick bits of code that you can look over and manually see great. That's, that's really it. You can't really build entire software projects with coding LLMs. I guess Rag search is kind of useful. It's a slightly more more efficient. I don't even know if it's more efficient search. That's the thing. I'm not even being a cynic here. I'm really must be clear. I'm not sitting here being like, oh, it all sucks just because I want, like I don't care, I don't want it to or don't want it to. I'm simply stating what I've seen. I've yet to see one thing that maybe go, oh, that's interesting. And I'm. I'm an early adopter. I'm sitting in my office right now. I have a tonal workout machine over there. I have a Corsair 1000D2 motherboard thing back here. I've got all manner of tech crap on the floor. I love weird new stuff. I have a GPT Win 4 which is a tiny little PlayStation portable looking Windows PC. The buttons are kind of terrible, but I find it fascinating. I can tolerate bad or weird or crappy. If there's something interesting, it's doing nothing out there doing this. I really mean that's really, I think the crowning glory of this era. How goddamn dull it is, how really just unexciting and lifeless it feels. Because I am also at my PR day job, I'm spreadsheet heavy I'm in the spreadsheets all goddamn day. I hate it. I pray for the end. Sometimes with the amount of links I have to put in a document, you would think that I'd be able to say, look up everything written in the last week by this client and I could hit a button and it would spit it out. I tried to get Manus, an agent, an agentic AI to do for my own name. I said, from the last year or two years, get me every link that mentions me. Should be real easy, right? Should be the thing that this thing is. I watched it write constant like lines of Python, just like crunching numbers, obliterating compute to get me 10 goddamn links. I think I've had at least 50 citations. And I'm not even tooting my own horn. I'm just saying very easy to get this right. Like very should be very easy. But even on the block and tackle most obvious use cases, it doesn't do it. I don't have much use for generating text. I really don't. But even if I did, the text it generates sucks. I truly. And I've sat here and tried to work it out. I can't think of an actual use case that would change my mind anymore because, oh, it could organize your calendar. Can it? Because my count, the whole reason that you have a calendar organizer or you have an assistant is it's not so that they can manage your calendar. It's so that you can trust them to. It's so that you can say, oh, find me a time like Tuesday, the Thursday and make sure that it doesn't get in the way of some crap I have on you don't. If you have to handhold an assistant, it's not an assistant, it's an intern at best. It's an intern that never learns. And so I'm really sitting here trying to think, okay, what would be the thing if I guess generative search was really powerful, but even then it isn't. And I've used the very high end generative search products as well. It's just also distinctly underwhelming. And I'm really, I'm trying to, trying to be fair here, but it's just, it's really underwhelming.
- [17:59]
A
You summed it up, I think really well with the word trust, right? Like it feels like trust is the biggest barrier to this thing being.
- [18:08]
B
No, I mean, it's, you're right, but I think it's, it's simple. It's you, it. Trust is part of it, but it's the fact that it doesn't work.
- [18:19]
A
Right.
- [18:19]
B
And, and even if it never will, because the limitations are probabilistic models. So you can't trust them. So you're right. But I think when you say trust, it's very easy for people to get tied up in that and go, well, well, maybe they could build up trust. No, it's that because you can't. Like, if you give them, if you let them off the hook and say, well, build up trust, it means that they can get you to kind of efficacy. No, you can't. The whole point of automation is it needs to be unilateral. You can't leave something to control other things if you can't trust it. To your point. But the reason you can't trust this is because probabilistic models are not operating on defined rules. You can try, and they're really trying with deterministic stuff, but damn, is it not working.
- [19:05]
A
Well, and that's exactly where I was going with this, that the trust is not, in your view, an iteration away. It's an impossibility of the fundamental structure of how this technology works, the probabilistic model. But to me, listening to everything you have to say about this, it feels like, and you mentioned your love of technology and early adoption, it feels like this not working is only a fraction of what's kind of drawn you into this. My sense is that what really pisses you off is this layer of what feels like collusion in tech and in the media. So what's going on there and how have you seen it develop over the last few years? And do you have any sense that it's going to get better or worse?
- [19:55]
B
So collusion I don't think is happening with. I don't think the media is colluding. Just want to be abundantly clear there. I don't think the media is doing their job. I don't think the media is sitting here trying to work out what this stuff does. I think they are accepting the company line. I think across the tech industry there is a. They've realized they've hit a wall, that there are no more hypergrowth ideas. We don't have a new cloud, we don't have a new smartphone, we don't have a new anything. We don't have a new SaaS. We don't have a new fuzzy thing to shake at the markets. So everyone chose AI. The collusion, I would say, is the fact that all of them are agreeing to do the same thing so that no one has to stop the moment that someone stops the moment, that's when it ends. It's the moment one of them says, we don't want to do any of this anymore. None of them want to. They all hate it. It's so obvious. You've got all of them saying it's a bubble or an overbuild at some level other than Sundar. I think at this point Satya Nadella said it might be an over overbuilder. Dwarkesh Patel. Yes, Dwarkesh. I think a few months ago, Mark Zuckerberg just said it a few like a month ago to Alex Heath. Sam Altman just said it two months ago. I mean they're all saying it's a bubble, so they want to be done. But I think the thing that drew me into it was the fact that the media is acting like it can do things it can't. And also the insane numbers, the billions of dollars burned for bugger all revenue. I think that that drew me in because I think that the way when you look at how this is sold in the media, you look at how they talk about it in the. On earnings calls and such without ever revealing revenue, they talk about it as this massive revenue driver. When you look at the underlying numbers, they're insane. They're terrible. From my calculations, best case scenario, Microsoft's making about. And they're definitely not making this much. If they had 8 million paying active users and then 4 million more inactive paying users of Copilot AI and they were all paying 30 bucks, you're looking about $4 billion of annual revenue, which is piss poor for Microsoft. It's absolute. That's dog shit. Like they've. Those numbers are. Steve Barber would throw a chair through you and yeah, you've got what, 80 something billion dollars of CapEx this year for this. And by the way, they give tons of discounts. More likely they're making like 2 billion on this, which is still piss poor. Now Microsoft also stopped announcing their AI revenue. They said it was annualized January 20th. Whenever they did their January earnings was the last time they mentioned it. 13 billion annualized, so about 1.08 billion a month. Again, they stopped saying that number. Yet they still say AI is this amazing thing. If you think something's amazing, why aren't you telling me how much money you're making? And the answer is they're not making the money. It's very clearly not going well. So what's pissing me off is I'm seeing everyone act like and this has been over a year And a half of this now everyone acting like this is the next big revenue driver, the next big economic driver. And the only thing it's been driving is hundreds of billions of dollars of Capex into data centers full of GPUs from one vendor.
- [22:57]
A
Well, and that's the piece that has me most concerned these days. It feels like more and more you're hearing stories of yes, these big Capex purchases here, but also it feels like every week there's some new story about big tech vendor, a promising however many tens or hundreds of billions of dollars to big tech vendors. See, and it just. You used a word on your podcast that was rattling around in my head and I haven't heard anyone use it yet, which is it feels, it feels Enron Y. I don't know if you said Enron Y or you said Enron.
- [23:34]
B
Enron, Enron, Nortel.
- [23:37]
A
It just feels like what you would expect to see pre a bubble bursting. Like it, like I don't know if it feels like the Russian economy or what. Like it just does not feel.
- [23:46]
B
Oh, it's very like very similar, yeah, it's very similar to Enron. But Enron was insane. I don't think people recognize how insane Enron was. Like, this is not Enron. This will never get there. Because Enron was making SPVs and giving their debt to other companies. Honestly insane, but kind of impressive how insane. It was just like, yeah, we're going to make a company just to offload all our debt to and then we're going to get revenue from them fraud masters. However, one thing Enron did but more Nortel and the telecoms giants did during that bubble was they would give these generous vendor financing packages to people so they would sell, sell their stuff to someone else with generous credit package when. Don't worry about paying me for a minute, I just need this for the earnings. I don't know what Nvidia is doing. I do not have any privileged access. Their accounts receivable are climbing. If those things don't start coming. And to be clear, AR generally climbs. It's not necessarily always a bad sign, it's in fact in this case it could be a good one. Nvidia loves selling chips. It's great. But revenue growth isn't growing at the same rate as accounts receivable. So I'm kind of like kind of weird. There's also the fact that 39% of their last quarter revenue came from two resellers, Super Micro and Dell. Supermicro last year was under massive fraud investigation. So Always good that that's one of your larger customers. So I think the simplest way of putting it, I've kind of got myself in a tizzy because whenever I think these numbers, I feel insane. It comes down to something simple, which is if there was a healthy market for GPUs, Nvidia would not be doing this. Nvidia seeds. Companies like CoreWeave or Lambda, they invest in them, they become their largest customer, which they sell them GPUs, those companies, these Neo clouds. So Core weave will take the debt and the contract, sorry, the contract and the GPUs, use those as collateral to raise debt, which they will then use to buy more GPUs from Nvidia. That's not. That's not good. That's not a good sign. That's not good at all. That's not what you want to see. But then you get to the much more worrying thing, which is when you go and look at these Neo clouds. And just to be clear, a Neo cloud is just an AI specific compute company. So like there's Microsoft Azure, there's Google Cloud, there's Amazon Web Services. Those companies sell all kinds of compute. The specific Neo clouds, the core weaves, Nebbyses and Lambdas of the world, they just sell AI compute. Now, the problem with those is when you look at their revenues, their biggest customers are Nvidia, Microsoft, Amazon and Meta and OpenAI. That's not good. That's not a diversified revenue base. And also when most of your revenue is from hyperscalers that build their own capacity, that's basically them saying, we will quit one day. Like we otherwise, like, they are just using this so they don't have to raise their own capex guidance. Again, this isn't a sign that the industry is doing well. This isn't a sign that people want AI Compute. It's a sign that there are hyperscalers that have a vested interest in keeping AI alive that I would call a form of. I don't know if it's actual collusion. I don't know what they're. That's actually my biggest question right now. What are they doing with all this Compute? Because it certainly isn't useful. I don't know. It's all very weird to me.
- [27:20]
A
It is weird. And it, you know, you keep seeing these headlines and to your point, if you read past the headline, it feels like every number you read should have an asterisk beside it, right? Like every time you see something going on, it's like, well, we're going to invest this much money, but it's contingent on A, B and C. And it's actually circular and like, it's tough to really understand what's going on. And maybe the most confusing part is the market seems to be rewarding them for it in general like that. The skepticism doesn't seem like it's permeated the investor layer. And I don't know if that's because there's big institutional investors that are, you know, in on the take or can't have this fail. But I mean, do you have a sense of the, the mechanism there that's allowing this to continue to like, grow versus shrink?
- [28:10]
B
I think that it is far simpler. I think the markets are run by babies. I think that, I think that they have the institutional knowledge and emotional intelligence of babies. It's that simple. Oracle announces a $300 billion deal with OpenAI and the market goes psycho. $300 billion over four years or five years. That number keeps changing, by the way, and the market goes, ooh, the squealing, annoying king and rolling on their backs. It's disgusting, but also unrealistic. OpenAI doesn't have that money. And everyone is saying that now. OpenAI doesn't have the money. Boring. That's dull. That's tired. You know what's wired? Oracle doesn't have the capacity. OpenAI can't afford to pay Oracle and, and Oracle does not have the capacity. It takes about two and a half years per gigawatt. They need four and a half gigawatts to serve this contract. By my calculations, they will only have 1.2 gigawatts by the middle of the fiscal year 2027, when OpenAI should be paying them 30 to 60 billion dollars. I forget which one it is. The point is, if the markets even sat there for a second and thought about it, they would be panicking. I'd be panicking. This is panic worthy. And what's crazy to me is I don't have a finance background. I was like the dumbest guy in my private. I went to a private school in England. Miserable time, Worst time of my life. I was the dumbest kid in the year, every year, without fail. I learned all this stuff myself. And this feels very obvious, but when I talk to like hedge fund managers, they go, ooh, really? Oh. And it's like, what the fuck do you do? What do you do all day? Do you read anything other than Excel? Do you Vibe Excel? Do you use Vibe Excel? Do you use Copilot? All of these AI boosters also if you ask them whether they use AI, they get very nervous because they don't really have an answer. But putting that aside, the markets want AI to exist. They want it to do the thing. Because once the markets accept, I don't even know if they fully understand this. Once the markets accept that AI is not the thing, they will realize that there are no more hyper growth markets in tech and they will start punishing every tech stock. That's inevitable. But right now Nvidia is the fastest, most, strongest, most biggest company that ever did big company anything. So yay, more number go up. Who knows if number will continue going up. But I think the on top of all of that, I think the market has an unhealthy relationship with Nvidia. Nvidia's last quarterly earnings they had 55% year over year growth. The markets were pissed. The market's like, what is this? What? What? Because I think, I think 2024, the same quarter they had 265% year over year growth. So the markets are babies, still babies. What's crazy is if Nvidia was to in a year's time from their last earnings, make 55% year over year growth, they'd be selling $72 billion worth of GPUs in one quarter. It's not sustainable. None of this makes any sense. It's all this short terminism bullshit and it's going to hurt retail investors.
- [31:20]
A
So let's talk about those retail investors because there's, you know, we're talking about words like inevitable correction and which I completely agree with you on by the way.
- [31:29]
B
Yeah.
- [31:30]
A
Assuming there is an inevitable correction and it, you know, I'm curious, you mentioned earlier that as soon as the first tech company kind of says okay, we're like putting our foot on the gas here. That's going to be the first domino in the series. Who is most at risk here? Is it the tech companies? Is it retail investors? How. What kind of impact could this have on the, on the broader economy?
- [31:52]
B
So Nvidia makes up 7 to 8% of the S&P 500. So that's where we should stop then. Yeah, it's really, it sucks, man. The Magnificent Seven makes up 35% of it, I think, which is Nvidia, Microsoft, Meta, Google, Amazon, Tesla and Apple. I think that's eight. There's seven of them. Jesus Christ. Nearly said the Magnificent Eight there. Very bad. No. So first of all it starts there. So if you are invested in the market, you're probably partially invested in the Magnificent Seven. I think that there are also a lot of unfortunate retail investors. I don't know this for a fact, but I get the sense that there are a lot of unfortunate retail investors who may be heavily shored up with Oracle as well because of Oracle's big growth burst. So what's going to happen is when these stocks tank that like people's retirements, people's 401ks, people's honestly just people's Robin Hood accounts are going to start looking like a dog's dinner and the tech companies are going to be fine. Other than OpenAI, I think OpenAI dies and gets absorbed into Microsoft. Think, think Amy Hood unhinges her jaw and just because Microsoft owns all of the research and IP and has access to all of their models and exclusively sells them anyway, so. And they also run all of their infrastructure so they'll just take the open AI sticker, put a Microsoft one on there, they'll be done. But I think the. There is going to be apocalyptic terms for startups, I think because I think AI took something like 66%. I thought it was 33%. I got told recently it was way more than that of a. Of funding, of startup funding in the us. So you're going to find a bunch of venture capitalists who can't raise anymore. They were just. Their limited partners are going to say go fuck yourself. Because the other unique thing about AI startups is because they're so. Because it's so resource intensive, because the cost of running these models is so horrible. These companies have had to raise all this money and. And then only to find out that they can't sell because it's really easy to clone the products because all of these products are just wrappers of GPT or Claude or what have you. Because in previous booms sure you would be building cloud software and code languages, but your ingenuity as a coder or as a systems engineer, the systems you could build were your own IP. You could build really fundamentally strong IP with LLMs. You can't. You'll notice that no one has a profitable company. Nope. And in fact no one really appears to be able to sell their companies. I found one AI company like a true blue generative AI company that actually sold for money. It was Cognig V, I want to say Cogniggy. They were a customer support AI company to a public company called Nice. I've never heard of. Otherwise there was one that sold to AMD in 2024 and the rest of them are these weird kind of aqua hire situations. Like inflection or wind surf, what have you. Kind of a cludgy point, but there is a reason I'm telling you, which is you're going to have all of these startups, these AI startups that are going to be sat there with nothing to do, they can get acquired or die. So you'll see a wave of aqua hires. I also think that you're going to see something I call the subprime AI crisis, which is right now all of the rates that you're currently paying with say anthropic or open AI are subsidized heavily. I believe all of them were unprofitable. I think that at some point they're going to crank up the prices. This has actually already happened. I reported this back in June, the anthropic June or July. I think that Anthropic and OpenAI released service tiers, priority processing, which means the troll toll. You need to pay this otherwise you won't have proper uptime. Which was never a problem before, but it's a problem now. It's a problem we've created for you. So that is going to get worse. I think they're going to just start increasing the costs of these models on a per to per million token basis. But that's kind of already happening. And indeed reasoning models make that happen. But again, kind of a cludgy point, but there is just a setup here where all of these AI startups have no, they have no horizon, they have no future. There will be no way to turn them profitable, there'll be no way to get them acquired, so they will die. Once that starts happening, it will just become flat out difficult to raise money as a startup, in part because so much venture capital in the US has been jettisoned into AI. So you're going to have venture capitalists who can't raise, startups who can't raise. Tech companies are going to take a haircut in the public markets. This is going to fuck over retail investors. Retail like regular people that invested in this, who believe the headlines, who believe that AI was taking jobs, that believe that AI was getting exponentially more powerful, saddled in Nvidia and Microsoft. Those, those companies are going to take a real hit. I also think we're going to see net after effects. I think core weave will die. I really despise core weave there. They are antithetical in my belief to what technology is. They're not about providing AI compute. They are a debt vehicle wearing computer hats like they are there to raise debt to buy more GPUs. They do not have a ton of real customers. I think they're, they're going to be dead. I think Nebius is going to have similar problems. I think Lambda, I think all of these AI specialists are going to start collapsing because they're so. Core weave has $25 billion worth of debt. They lost $300 million in the last quarter. They're going to lose more next. None of this makes sense, but those stocks will get completely plowed. But we're going to see net after effects with companies like Broadcom. Broadcom is meant to get $10 billion from OpenAI next year for AI chips that they built. Special bad news. The information reported a few weeks ago that apparently those chips will have modest rollouts. That's what you want to hear when you spend $10 billion. The word modest is what you're really looking to tell people. Modesty, everyone. So on a grander level, once the illusion breaks, because this has all been vibes. It's funny they use the word vibes so much because that's really what this is because it has been vibes the entire time. Once the vibe shifts, once it. When fully everyone accepts what they know to be true, what they see in front of their eyes, the collapse will be thorough and horrible. And I think everyone that touched AI is going to get burned on some level. And I think there are. I don't think any of the. The problem is that all of the ultra rich people will be fine. They already got their money, they already cashed out. I think Sachin at Della is on the way out. I think Amy Hood boot boots him hard. I think Judson, although who took over a few weeks ago as the CEO of Commercial Applications, what Satcher do anymore, I think that we're going to see him get booted out. And I think that I reckon Meta is going to be the first to pull out of capex. But long story short, retail investors get fucked. They always do. It's always, always, always regular people.
- [38:44]
A
So with the regular people in mind and with this forecast which as I said, like to me, inevitable is a word that is in my vocabulary for it as well. If you're a business leader right now or you're an investor right now, like is there anything you can do differently to you know, limit your exposure or make this less bad or even help like at an individual level or even at a broader level to try and make the, the, you know, the popping of this less dire for the broader.
- [39:18]
B
Economy really depends who you are. I don't have any stocks. I'm a psychopath. I live in cash people now. People like you should have short positions. No, it's far weirder when you know that I just do this for the, for the thrill. But my general thing is like batteries and EVs and solar, like they are the actual future. I don't know for this. The current tech industry is so divorced from actual value creation that I don't know what the hell to do with them. I think that on an organizational level, it's time to walk away from AI commits. There's no reason to be doing this anymore. It's a waste of time and money. If you have said the words train people for AI, you are being conned. The wallet inspector has got your wallet. This whole thing of we need to train people for AI is gaslighting. You are being gaslit by software. The software is, it is inadequate. You are training yourself to. It's kind of like occasionally you'll get like people talking about their dogs, training them. This is what's happening. You are learning bad habits because the software sucks. Your dog is also more intelligent. So the thing to do I think on a very basic level is to say, yeah, walk away from these AI commitments, Stop pushing them. If anyone says AI training, say, why do I have to be trained to use this? What is the outcome? That is actually the big question to ask anyone. Just what's the outcome? What do you think? What, how will we measure the success of this? And if they can't come up with that, say we shouldn't do this. This is expensive and bad. I think the anytime I've written, I just put out a few weeks ago an 18,000 word kind of compendium of what the case against generative AI. Read it. I know it's kind of selfish, but it's free. Like it's literally. I don't like read it or not, it doesn't really matter, but I go through it in there. But it's learn the talking points which are these models are unreliable, these products don't do much. And honestly everyone saying AI is the future doesn't use them. That really is it. Bill McDermott from ServiceNow around the launch of chat GPT said everything had to be AI AI AI AI AI. And that is a quote from the information. Anyone speaking like that just immediately think this person doesn't use it. Because if they did, in pretty much every software revolution you can point to them saying, and then I did this with this. Go and watch the original iPhone announcement. Steve Jobs, horrible monster of a person. Deadbeat dad sued by the California district attorney for child support for his kid. And I'm not kidding, it's wild. Nevertheless, his first presentation, he goes, oh, yeah, we've got an ipod. Got email, got this. Now the one device, the iPhone is coming out here. He just says what it does. He just says what it does. He says the. The functions. And then you go, wow, I use function. Wow. I too, listen to music and use email and website. Wow. You didn't have to do. You didn't have to solve the riddle of the Sphinx to you to know why the future was here. Like, that's the biggest insult of all of this to me. It's. You can't just have someone go, you've got to use this. It does this. Every single product I've ever used that I've been excited for, someone said, you've got to fucking see this, man. Like I mentioned. So to nal, it's. This workout machine has magnetic resistance. It can check your form with a camera. You can adjust it for the lifting. When I show people this, they go apeshit because they're like, wow, that's really cool. It's insanely expensive. If you don't use it, it's a terrible waste of money. But the point is, I don't have to explain it much. They see it and they go, wow, that's sick. Even, like, the iPhone. Air. IPhone is awesome. It's awesome because it's. It's thin like an iPad pro and it's got a big screen. It looks nice. Wow, how great. I don't have to advert. Every person I talk to about AI is emphatic in trying to convince me. And it feels like a mlm. It feels like MLM or talking to a polyamorous couple. Like, they're constantly trying to convince you that their way of life is. Is the right one and that you are. You simply don't get it. You need to practice conversion so that you can use ChatGPT. It's utter bollocks. I. I shouldn't. I. Like I said earlier, I'm an early adopter. I love new tech shit. I'm willing to look past wonkiness because I'm like all of the current handheld gaming PCs, for example, like the ROG, Ally X or the Lenovo Legion. I think they're wonky. They have wonkiness them, but they're cool. You. You're like, wow, I can see the future in this. It's like a PC, but it's like a game Console. Wow, how amazing. The new AirPods 3. Well, they fit better, they sound good. Oh, they can measure my heart rate. I know what I do with that. Chat GPT Analyze data. Indeed. When you go on OpenAI's website and you try and read what they say, it's like, oh yeah, ChatGPT could brainstorm ideas, you can analyze documents. I guess we need $1 trillion. We need $1 trillion because more even their own literature, they say like 90% of people just use it to share. It's not a product, it's a parasite.
- [44:35]
A
It feels. There's a word that caught my attention earlier that to me encapsulates the best use case for it in business or commerce, which is. You said it's like an intern and it feels like that's what we've built. We've built this army of free interns and they're not very good. They don't really know what they're doing, but they're free. So everybody should have their own free Internet. Do you buy that logic or is there.
- [45:02]
B
I, I think that that is the closest you'll get and I think it's an insult to interns because an intern, even the dumbest intern you've ever met, who does not understand anything, understands more than this. Because these models don't understand anything, people. Also, even the silliest morons can learn. You can learn. I'm not even insulting. I don't. Not calling interns morons. If interns don't know stuff, it's because you're a shitty boss. Your job is to teach an intern. That's the whole point. That's why they're free. They're free because you are giving them something in return, which is the value of your experience so that they can do a job and potentially do a job for you in the future. So you can say these are like interns in the sense they're useless in the sense that you need to handhold them on everything. But guess what? Even into like, I, I started as an Internet of games magazine, I built shelves, I got tea. And indeed I tried to make the tea wrong once. And they, they showed me how they knew that scam. Respect to Micah over at CVG magazine, but it's about as close as you'll get, but not close enough because an intern can learn. And also an inter. The very existence of an intern is that it will grow, that an intern will grow. That's the whole point of having them. An intern that stays trapped in amber at a certain level of incompetence. Is useless and also antithetical to what an intern is. Like I said, it's about investing in the. In the future. But I guess if you're an asshole and you think an intern is just a vessel that you allocate outputs to, fine. But I think that that's actually the overall problem. I said this in the article from a few weeks ago. It's people that these executives see every job as a unit of work. We are not on a podcast talking. We are just talking for a certain amount of time. There is no. The value here is the fact that our mouths open and noises come out. It's not really about the noises themselves when you were writing, it's just about making 800 words and there's meaning in there. I guess it's not really about they see everything as outputs. They'd see software engineering is just coding when that's not true at all. It's software engineering. It's systems architecture. It's making sure things work now. And you leave enough comments so that things work in the future. And you build things in a logical way that doesn't break when you add stuff. When you. And of course people would think, oh, that's what an intern is. They're just units of work that happens to have a heart and a body and a brain, allegedly. When in fact an intern is promise an opportunity and hope. And if you see them as something else, you're an asshole. But you're also just wrong. So, yeah, if you see interns as this empty vessel for your theoretical job, great. But I don't know, man, I. I've never taken on an intern. I've mentored a few people. But generally the most exciting and fun thing about having a young person working for you on any level is that they want to learn that they're excited to. That they grow with you and grow for themselves and then eventually leave you and go and do their own thing. Having a trapped in Amber intern that vaguely resembles a work product, that's useless. But I guess when you're an asshole who doesn't do any real work. Not saying this is you, by the way, just to be abundantly clear. Just really want to not insult you because it's not what I mean. It's. Yeah, of course that is it. But it's so weird as well. It's like, wow, even if, Even if that was true, even if this intern thing was true, wow, we've given. We've given These companies think OpenAI's effective cost if you include Microsoft's investment in capex is like $100,000,000,000. Wow, that's a really expensive intern. Man, that's so expensive. We made crappy interns. We made flash frozen interns. Jesus Christ. How depressing. 20 years ago we had the ability to condense computers into little frames. Laptops are incredible. We've actually had innovations in the last 20 years that are insanely amazing. Like this is actually a side rant which I'll stop. But like we've had amazing innovations. This isn't one of them.
- [49:14]
A
So I want, you know, I want to talk about amazing innovations. I think the intern piece, I mean, well said. I think we've probably beat that topic to death at this point.
- [49:23]
B
We're aligned.
- [49:25]
A
I think so too. Innovations. You know, you mentioned things like batteries like solar power. Know you're, you're, you're someone who's obviously very passionate about new and emerging tech. What technology that's up and coming actually excites you. That's not, you know, AI related. And, and what tech companies, if any, do you actually see doing really interesting work?
- [49:48]
B
So the. I like Anchor A N K E R because they keep making interesting and fun things. They've got these little b. They like, look. I don't really understand woman buying shoes before until anchor. Now I understand because I do it with batteries. They release a new battery. I'm like, ah, credit cards out because they keep making. Using I was Gallium Nitride. They found a way to basically condense power to a way that you can get a little power brick. I'm trying to think like half the height and width of a Diet Coke can with a retractable cable that charges at 45 watts. This sounds minor, but that can charge a Nintendo Switch. That can charge a MacBook. I think a MacBook Air. That's cool. That's cool as hell. They have this thing called the Soundcore X1 Pro, which is just a giant wheelie projector like, and it can keystone like you can put it at a weird angle, it can project onto things. It has the speakers built in. It's all like wireless. That's cool. I think that was it Bid like the electric car company in China. Their cars aren't nice, but driving down the price of electric vehicles is cool. I also think I have complex feelings about Waymo, but I've been in one and it's the first time in a while I've gone, ooh, wow, that's new. That's cool. I mean that's something. And I also think That I don't know, you've got. What is it? Framework. There is a laptop company that I really wish I remember the name of where you can put the laptop together yourself. It's built specifically for. Right. To repair. Cool as hell. I love it. And I also think that the new generation of. I think within five years we're going to have insanely good handheld PCs. I think handheld PC gaming is going to be huge. I think it is way too expensive right now. But even now it's so cool. It's so cool. We're in the, we're in the golden age of mobile gaming right now, which is insane. I thought it was dead because I thought cell phones with microtransaction crap had ruined it. But no, it's exciting, it's cool. And I think just batteries are getting smaller and more powerful. With that we will have new form factors, we will learn new things about what we can build. And I'm going to say something really controversial. We are not there yet and I don't know how long it will take. But the Apple Vision Pro proved that there is a new computing form coming. I think There were entire 10 minute periods with the Vision Pro where I was like, wow, this is the future. And then there was the rest of the time. Then there was the rest of it. When the thing moves slightly and you can't see properly or it's slightly out of focus and you're dicking with it and you're not sure if it works properly, sometimes the poking doesn't work, but when it does, and to be clear, I'm very clear, this product should never have been released. It's way too early. But man, you see it and you're like, oh, these are surmountable problems. These are like, these are things that they can work out. You can see the promise. I don't see any promise in generative AI. Nothing that's happened has made me go, ooh, what could they do next? Vision Pro? Yes. Even I haven't tried them yet, the Meta Ray Bans. I don't think anyone should wear them. I think they are just cruising for a bruise and wearing them. I think the idea is sick and I think that there are accessibility options. Like friend of mine works on a company that does accessibility stuff with ar. I think that that is incredible. I don't think AR is ever going to scale, but I truly think like, like front worn, whatever you call it. Like I want to say wearables because that can mean anything, but I think that AR has possibilities. I Don't know if it's a hypergrowth market. I'm none of these things I think are hundred billion dollar market cap companies. What I do think they are is cool. I think they are cool. I think it's an interesting way of looking at the world and I don't know, I'm excited to see some new stuff. I just. And it kind of. To wrap it back around. That's kind of the depressing thing about AI. It's so dull, it's so boring. It's so. I've never been more bored with the tech revolution in my life and I used to do. I used to represent platform as a service companies. I don't. I've not had to say kubernetes in a long time but even thinking about it makes me slightly sad. That was more exciting than generative AI. At least that's like, oh, we're going to. Sorry, I'm rambling. It's just the tech industry is capable of doing cool shit. It's not doing it right now at scale.
- [54:21]
A
Right. So there's, there's one more angle I wanted to take kind of around the AI thing, which is one of the, one of the quote promises of AI is that it's going to in some way transform the future of work. And work is going to look so different in six or 18 months than it does right now. And we've debunked a bunch of that in the past hour. But I did want to ask you, Ed, from where you're sitting, what does the future of work look like over the next handful of years? If it's at all different from now? What do you think we'll see more of? What do you think we'll see less of? And how is that going to evolve with or without AI?
- [55:01]
B
I think it's going to look much the same. I think the biggest problem that they have right now is that on top of their oh, this is good, it's just an idea. But I think on top of that there is no innovation within business software left. Maybe there, maybe there is somewhere. But to innovate business further, like really, this is scary to say, I realize to actually like enterprise software and stuff like that. They're out of innovation because enterprise software was never about innovation, it was about taking industries and disrupting them. There was all that that was. There was never. It was modernizing rather than innovation. So the future of work is a problem that they've had for a few years with the remote work era because they desperately wanted people back in the office because they realized, oh crap, that's remote work. Makes it really obvious who doesn't do anything about. At my company, AI comes along and they're excited because, well, it means they can fire some people, but it also means that they can make some of their useless people useful. Problem is, AI doesn't do that. AI actually doesn't really do anything. So the future of work is going to look very much the same in a few years. I truly think that. I truly think that AI is not having a meaningful effect on the enterprise. It's kind of been proven through multiple studies at this point. And that's because the modality of our jobs have stayed mostly the same. We have the ability to record stuff thousands of miles away, hundreds of miles away. We have the ability to send large documents and process large documents in the cloud. We have the bits and pieces. What we don't have is technology that makes us better at our jobs for the most part. Because a lot of the jobs that actually humans, A lot of the jobs that humans do within software are very much human driven. The actual tasks are not the output itself. They are the ability to understand all of the stakeholders and all of the context and all of these things and bring them together in a task that might seem simple but is actually a combination of institutionalized knowledge and context. You can't automate that. You can't automate person to person. You can try and they've been trying, but it doesn't work. So I think that we're going to see very possibly a kind of existential crisis within the world of work. I think once the AI boom goes away, CEOs are going to sit there and not really know what to do next. Because the big thing, the reason that CEOs love AI as well, is when you're a CEO, it's not really obvious what you do unless it's spend money. If you spend money as a CEO, that means you're doing stuff. AI lets you spend money. And once AI is gone, what the hell are they going to do? I don't know. Maybe that didn't work.
- [57:37]
A
Yeah, the picture you paint, I think in some ways comes full circle of why the promises of AI and its capabilities are so alluring. Because they present there's this, oh, we know what the future of business or the future of work or the future of enterprise software looks like and it's going to be so much more efficient and we're going to have so much more capability or quality. And if the alternative to that is saying you know, putting up your shoulders and saying, we've got nothing. That's a. That's a hard pill to swallow, right?
- [58:09]
B
Yeah. And I think the remote work in the COVID era really showed how many bosses don't do a goddamn thing. And I think it showed a bunch of managers that didn't either. AI gives them the exciting idea that they can pretend they do work, which is what they've been doing already, but actual work would come out. And I don't think they have an answer. I don't think anyone does. I think that this is. This is. Everyone's caught with their pants down here because no one was thinking too hard.
- [58:34]
A
Well said, Ed. I know we're kind of at time here. I wanted to say a big thank you for coming on the program today. Really love the conversation, and I really appreciate it.
- [58:44]
B
Thank you.