Transcription
We're reaching this thing where we're realizing that everybody made a huge mistake. Everyone has invested in this thing for really effectively no reason. The actual data keeps showing that there is no productivity growth.
What we're seeing here is the beginning of a hysterical episode following another hysterical episode where everyone said that something was happening when it wasn't. And I sound sarcastic, but look at the data. 90% of them said no impact on employment or productivity. Literally nothing changed other than they spent a lot of money on something and uh everyone had to use it. But yes, this is we work too except it's so much worse.
Joining me on the tech report today is writer of Where's Your Edat and the host of the Better Offline podcast, Ed Zetron. Thanks for coming on.
>> Thanks for having me.
I saw a study by the National Bureau of Economic Research which revealed that 6,000 CEOs that it surveyed across the US, Europe, and I think Australia as well. 90% of them saw no impact on employment or productivity in the last three years from the adoption of AI. The study points to the information revolution of the computer where computers were sold and productivity growth was expected. But in fact, due to the overwhelming amount of information that these computers are producing, productivity actually slowed. I mean, anyone that is familiar and uses AI is probably hearing alarm bells at this point because they know that that's kind of what's happening now. There's just too much information being made by AI that it's actually slowing things down. Do you think history is repeating itself?
I mean 1948 uh we'd also lost a lot of people very recently due to a war and on top of that the introduction of the computer was nothing like the introduction of AI. Mini computers sound small, they were very large, these were room-sized computers and then eventually terminal-sized computers. What you would now find as a server act was one computer. It was the very early days of what they could actually do. In this case, when it comes to AI, it's one of the most evenly distributed and accessible things of all time, you actually can't get away from it. It's harder to not use AI than it is to use it because if you load a product, go use Google, you type in a search, natural language model, [ __ ] You have to use that now. You pick up your phone. Oh, Apple Intelligence wants to read you your text messages. You didn't ask, but it's there. You load every single app and it's got some weird pop-up. You go into Instagram. Meta AI wants to harass you and generate an image. If AI was gonna help, it would have been helping already. And I think we're entering this point where this is nothing. We're at the beginning of history. We're not at the end of history, Mr. Fukuyama, which I realize is a misquote, but we're not at a point where, oh, this is just the beginning.
We're reaching this thing where we're realizing that everybody made a huge mistake. Everyone has invested in this thing for really effectively no reason. The actual data keeps showing that there is no productivity growth. There was another study uh the other day where it was like 75% of chief information officers were finding that they would have to in the next 6 months prove that AI hadn't worth or they'd have their budgets cut. I realize we've seen over a trillion dollars invested since 2022. Now, if you count the venture capital and all the capex, we've seen all of this money. We've had AI screaming in our ears. Every time we open a draw, there's an AI in there. and this large language model in everything now. And nothing seems to have happened other than the birth of a new, more annoying Twitter personality.
I think what we're seeing here is the beginning of a hysterical episode following another hysterical episode where everyone said that something was happening when it wasn't. And I sound sarcastic, but look at the data. 90% of them said no impact on employment or productivity. Literally nothing changed other than they spent a lot of money on something and everyone had to use it. Accenture is now saying that they are forcing people to use the AI. There will be checking and there will be your performance evaluation will be be based on your use of AI. It's what happens when something isn't really working.
Great example when the iPhone came out for example, we were still in the early days of bring your own device with software companies or companies. So you would wouldn't be able to just bring any phone with you. You would have to have one provision by the company. There was a growth of companies that did BYOD. But you had this thing called shadow IT happening. You had people secretly using their iPhones. And indeed, shadow IT spread into other kinds of software. People were using the software they wanted to use at work. It was the opposite of being forced. In fact, they were using it without their IT managers liking it. This is the opposite. This is the boss saying, "No, use my software. Use it. Use it, you pig. I want to see you using AI because we're in the era of the business idiot. The people that run these companies don't do real work. They're pushing AI because everyone's blaring in their a their ear that AI is important. When it comes to actually doing real work, just doesn't seem to do it. Even the supposed speed ups for software engineers, what does that actually result in? What is the result of this? What is the thing that's actually happened? No one seems to be able to tell, which is crazy considering the amount of money invested.
>> I definitely want to get onto your assessment of whether or not you think the assumed productivity growth from from this study as well is worth the money. But first of all, do you think that AI or more specifically the AI that people are familiar with like Gemini or Chat GPT, do you think that will eventually mature enough that we'll also it will become as sort of commonplace as as a laptop is now?
I mean, it already is, but not through choice. It already is very common. They're shoving it down our throats. It's as common as a laptop because when you load a Windows 11 laptop, now co-pilot's there. It's like a vagrant moved into your basement. It's like a a person's crawled through your vents and starts telling you that it could generate a summary of your emails. it it's common place uh but not through any consent from the user side. I don't think these models are improving in a way that makes them really essential at all. And I think that the thing is you can never fundamentally trust them. There is the ultimate the ultimate technology behind large language models makes them inherently untrustworthy. your use and trust of large language models ultimately comes down to how much you care about the information you have being correct or how much trust you have in your ability to verify it. Now, if the one of the few use cases that people can actually quote of large language models is research, how in the world can you trust research from something that hallucinates? Oh, they're getting rid of hallucinations. No, they are not. OpenAI put out a study last year that said that hallucinations are an inherent part of large language models. So they can't beat them. So I guess you just eventually trust them. I had someone the other day say, "Oh, we got something right about macrobiology." Did it? Can you say for sure? If the only way for you to confirm whether whether something is correct is that you knew it already, then why the bloody hell did you get research in the first place? Confirmation bias. I don't know. So, it's really what are they improving? How are they improving? What is the end point? That's the ultimate thing of this conversation. It's where's the beef? What what is it that I meant to be looking at and excited about? And where is the point when this is good enough? Because right now it isn't. And that's before you get to the fact that all of this is subsidized.
Do you think then that maybe the the increase in usage figures which have I mean I mean by and large obviously open AI dipped quite significantly recently but by and large as a as a whole I should say are going up. Do you think then that is because it is a AI LLM chat bots are being pushed into areas rather than people are seeking them out to go and use them.
>> Yes. I think that yeah, sure there's you have if you hear about AI and everything, you're going to go, I should probably check this out cuz no one will stop talking about it. But Google did a crazy one. Google Gemini, hundreds of millions of uh users suddenly. How'd that happen? Oh, right. They changed uh Google Assistant on every Google device to be Google Gemini. Traders hate this one trick. Uh Microsoft doing the same thing with Microsoft 365 co-pilot rename a thing. Suddenly you got all these users. Yay. I I'm sure that Google also considers every user of Google Docs who has a Gemini popup or Microsoft considers every single user that has a co-pilot prop popup on Word or Google Docs to be an active user of Gemini or C-pilot. This is just rigging the dice. And this is not something you have to do if something is useful. It's something you do if you're just trying to deceive people, which is what they are trying to do. If these things had to stand on their own two legs, no one would really care.
In that in that study that I mentioned at the start that despite the lack of returns, it's sort of executives are expecting an average increase in productivity of I think it was about 1.4% in the next three years. And it kind of there's been some analysis recently pointing to the J curve. So the the the heavy amount of investment, capital investment at the start creates a flat flat growth and then the payoff comes at the end c creating that sort of the the tool bit of the J. What what is what's your response to that kind of analysis?
>> I mean just wow. So I just have to wait until something happens, huh? Because that really that really is all that it's just like yeah it costs a lot of money and then something happens. Can they actually point to a time when that's happened? Because the.com bubble is very different to this one. All of that investment was in just on a very basic level. The fiber lasts for 15 plus years versus GPUs that are obsolete in one or two years and Jensen Hang's already talking about uh new GPUs he's going to announce. So cool. Uh but nevertheless, why is it that every one of their predictions, promises, expectations is that that the curve, that J curve bit, it's always Jay keeps going out further and further. It's the longest Jay we've ever seen. One of the biggest beautiful Jay's we've ever seen. It's just very strange whenever you ask them to actually tell you when something will happening is going to happen. It's like, oh, it's it's just coming up. It's just coming up. Sam Alman in India said yesterday he said that we're going to have super intelligence as early as the end of 2028 and to enjoy your job while it lasts. The thing that a guy without a real job would say at Dario Amadate, oh end of 2027, a a data center full of geniuses. Blah blah blah blah blah blah blah. It's always in the future, isn't it? When you ask these people to actually tell you when something will happen, it's always in the future. We need all your money now so that we can spend it so that then we can be rich. I think even though the data center debt doesn't make sense. Pretty much every data center is going to be underwater. Open AI can't afford to pay for the stuff. Oh yeah. All of those annoying things like addition and subtraction. Don't worry about those because um executives expect 1.4% more productivity. The people that do the least work and get paid the most expect more productivity from the thing that they are forcing you to use that no one who does real work seems to be able to explain the point of I guess it's you're faster at some things but not in a way anyone can measure. It's funny. It's when you ask for these people, you ask them how much money they need, they know exactly how much. It's more than you've ever spent on anything. When you ask them for proof, it's always vague, but you got to trust them because, you know, there's always a lot of money up front and then the good stuff happens.
Which leads me to my natural question. Give me another example that isn't Amazon Web Services. Give me an example because there isn't one. And Amazon Web Services was about $69 billion over 9 years before it went cash flow positive. They Open AAI is raising an 100 billion plus round, a teleathon. We are the world. Let's all found Sam Alman's future. But Amazon Web Services, probably one of the most consequential software com like company infrastructural movements in tech history cost less than OpenAI will have raised in the last calendar year. In fact, probably by the time they're done, half of it. Why are we doing this? Like it's I think that that's that's the real question. I like why are we doing this? What's the point? No one's Everyone's losing money. Regular people hate it. It's driving people insane. It's destroying our environment. It's created some of the most annoying people online that we've ever seen. People buying Mac minis and claiming they're running agents on Open Claw. But when you go and look at what they're doing, it's like, "Yeah, create a website. Summarize my texts." It's nothing's happening. We're just burning money for no reason. It's so sad.
If if we take at face value that there will be a 1.4% boost or whatever whatever it might be, 2% we could call it, is that enough to justify the hundreds of billions or even trillions even that are being thrown at the building of AI infrastructure? And then also on the on the open AI hundred billion dollar funding round, do you that's not the last one we're going to see shortly?
>> God, no. No, no, no. I mean, so on, but I want to also be clear about something. Anthropic has done a really good PR job in pretending they're the nice ones that aren't the big lossy company. No, Anthropic is as big, ugly, and billious as Open AI. They raised $30 billion. The information reports they're going to spend hundred billion on training in the next three years. Also, these companies when they say training, they want you to think R&D and capex. They want you to think training will stop. No, training means everything. bug fixes, small updates, stopping model drift. Model drift is when the model can't respond to the real world anymore. Anthropic by that by the split of how they spend their money. That means they're going to spend about $160 billion in compute in the next 3 years. They only got $30 billion. Oh, wait. They also promised $21 billion to broadcast for chips. So yeah, they're going to Anthropic is going to need at least 100 another 100 billion. Open AAI raising over a hundred billion now. That's nice. That should cover a third of the $300 billion they owe Oracle. And uh Oracle's building a bunch more data centers. They're going to actually owe them more than that. Also, they owe Broadcom. They're meant to build 10 gawatt of data centers by the end of 2029. It's going to be another 400 something billion dollars. They promise 6 gawatt to AMD. Unless, of course, they were just lying about all of those. But putting that aside, OpenAI, according to the information, plans to spend over $450 billion in compute. hundred billion dollars is barely going to let them wipe their ass. I don't know why they're pretending they only need this much. But apparently Amazon is going to be a big chunk of this deal. And I assume that $ 38 billion of this will be the 7-year long deal for paying Amazon for Amazon Web Services. If there was a functioning SEC, this would be investigated. It shouldn't be legal to invest cloud credits. Indeed, it shouldn't be legal to invest this much in a single customer. This is going to this is creating a dependency in the revenue streams of big tech on companies that cannot afford to exist. And if the only way they can exist is big tech feeding itself its own money, that's not a real business. It gets back to what I was saying earlier. It's not a real industry. It's not a real business. None of these things are particularly useful.
And on top of that, here's a crazy fact for you. So someone did a mathematical study of the limits on claude subscriptions for Anthropic. They allow you to spend anywhere from 8 to 13.5 times the amount of money you spend on your subscription in API calls. So just give you an example. If you spend $100 a month on a Claude Max subscription using Claude code, you can spend over $1,300 of credits. Yeah. Takes money to lose money, Isaac. And that's why these companies need all this money because they're burning money to keep customers because if they allowed them to use them use these models in a way that was sustainable, they would have to charge probably hundreds of dollars a week or even a day. So they do these ugly subsidized products to speedrun their revenue, to get bursts of revenue, to to beguile the media into thinking that these products at a $20 or $100 a month uh subscription rate will actually be sustainable and usable. No, no, no, no. Deep down, these companies are just it's an illusion and they will keep raising money until they run out of it or the debtors and the creditors and the venture capitalists and the private equities eventually give up on this.
There's a puzzle I've seen that is that Nvidia's valuation has remained somewhat kind of stagnant for the last couple of months, but data centers are still being built, which means chips are still being bought. But that then that kind of that doesn't square in my head with why Nvidia's share price would be remaining relatively speaking flat over the last little while. What what are investors waiting for in that?
I don't think investors have more than two brain cells to rub together at the best of times. I don't think if anyone was actually thinking about any of the logic behind this. They'd say, "Hey, Nvidia GPUs are so expensive that you can't actually just buy them with cash. You cash flow is not enough. You have to raise debt." I've seen predictions that hyperscalers are going to raise hundreds of billions of debt in the next year or two to buy GPUs. If investors actually thought about the mathematics, they'd say, "Huh, this isn't a test of the AI industry. This is a test of the global debt. This is how much debt can be raised." And as I've said before, data centers are all unprofitable the second you don't have a tenant. And even when you do have a tenant, I've heard out of Stargate Abene that Oracle will not be extending past buildings a past building 8 because OpenAI is not making them enough money. These stories should rattle the market. Nvidia should get dumped. But no, Nvidia will beat and raise because there is a weird capital situation where private equity is still feeding debt to it. It's flat because what's meant to happen with Nvidia now? It's just they're going to make more money forever. I think everyone's just kind of waiting for the shoe to drop with this company because I've said it before, for them to actually keep up with the growth rate they're on, they need to be by the end of the year, I think, selling 90 to$100 billion of GPUs in a quarter. In a quarter. That's an insane amount of money. And that will come from private equity. It will come from private credit, which borrows its money from banks. And that's the really scary thing because private credit you think well it's not really it's this kind of private industry you know it doesn't really it's not banks lending the money it's just private equity where do you think private equity gets their money from their parents? No from the banks. The banks lend them the money. So Nvidia's valuation being flat or up or down. I mean, at some point, the markets aren't operating rationally. And people will say in the comments, "Oh, the market can stay irrational longer than you can stay solvent." Very clever. You're the hundth person to say it. But the truth is that Nvidia being flat means that investors aren't really sure what's happening anymore. The boom, it will go 190, go 170, 190, 170, just keep bouncing between them. it will go up on invest on um earnings day. Everyone will go, "Yay, more money beat and raise. They've said they'd make this much and they've magically they'll magically make exactly 500 million to 1.5 billion more than analyst estimates every single time in a way that's kind of weird and suspicious, but you know, C, they've also got a customer concentration problem where like four customers are most of their revenue. I mean, that's not a problem either. Overall, in everything, I feel that everyone is just waiting for the other shoe to drop. They're either waiting for something to happen that will prove that all of this is worth it or they're waiting for something to explode.
This whole chaotic mess around Claude code where people are claiming they if you reading about Claude code, you think AGI already happened. Except what really happened was people were able to clumsily put together a website in an hour that looks exactly like the rest of the training data that Anthropic trained on. They see that and they're like, "Wow, I can make half functional software that barely runs. Wow, this is going to replace software stocks. Salesforce, Microsoft, they're going to hell." We are in the realm of the irrational baby logic where it's just oh you know oh well I've seen a slight sign that something might be changing. Well that means we need more bloody capex. We need Amazon to spend $200 billion. It's the only way any of this makes sense because it's wy coyote running at full speed. Don't look down because the moment you look down the moment you look down and you think about the numbers you think about the trillions of dollars that they're claiming they're going to spend. You think about these giant deals between Nvidia and Meta. The moment you think about it for a second, you go, "H, but none of this is very good, and it doesn't seem to be getting better." No, no, no. Let's all be stupid for a while longer.
>> With that in mind, what do you make of SoftBank going sort of all in on Open AI, increasing their I think 60% of their portfolio is now just Open AI, while at the same time, Microsoft has both detached itself from Open AAI to try and develop its own uh its own models while at the same time garnishing 20% of OpenAI's revenue for two years longer than it was supposed to. What what what does that tell you? I mean, is is SoftBank risking another Wei Work mistake?
>> Oh, they they we are already in Weiwork 2, man. This is Nobody does it better than Masoshi's son. I love him. I think he's a nutter. I think he is just it's a I think actually this era can be defined as less of people who are very successful but more by people that have taken the piss enough times and let's see how many more times they can do it. Soft Bank is allegedly putting another $30 billion into Open AI. They cannot afford this by the way. They're saying they're going to do it in three10 billion tranches. They're going to have to borrow even more. They're going to have to sell off things. I got no idea how they do it. But yes, this is we work too except it's so much worse. Truthfully, I think anthropic might be weiwork too. Dario Amade was on the Dwarvesh podcast and he said that the way that he he said using something using the term stylized facts, he said that well actually the way you should calculate profitability isn't um revenue minus COGS, it's uh cost of goods sold. No, it's um whether how much a model costs and how much money was generated by the model. No. No. profitability is it's revenue minus cogs Dario. Anyway, all of these companies are doing the same community adjusted EBIT duh much like we work. So yeah, SoftBank's doing it again. Microsoft's playing the field though. They invested in Anthropic. They put a little money into that 37 investor round. Very normal, very good. But also, rumor has it Microsoft's going to put a little money into OpenAI as well, which is great because OpenAI promised to spend $250 billion on Microsoft Azure. So, you know, just uh taking taking a tenor out your wallet, putting it back in your wallet. Very cool. Yeah, this is we work too. It's the same thing except dumber because we work, they had leases, they had real estate too much. They couldn't afford it, but there was a thing you could point to. OpenAI and Anthropic actually have very few assets. They have a lot of talent, a lot of AI scientists who work for 12 hours a day, but nothing seems to come out. You're not really sure what they do all day other than a lot of lot of yap yap yap. Don't lot of talking. Doesn't seem like the models are getting crazy better. Uh you've got a lot of contracts that you've promised, but no real assets, so nothing really to pick apart when this all falls apart. I'm not sure what Masoshi Sun will do when this explodes. His best hope is that OpenAI actually goes public and they can dump the stock. The problem with that will be is that even with a large float, dumping the size of their holdings, which will be over $50 billion, I guess maybe over $60 billion at this point. Jesus Christ. Um, dumping that will actually be quite difficult indeed. any fluctuation in that stock will affect their earnings directly. But also, I don't think OpenAI wants us to see what's inside their guts when they file for going public. Same with Anthropic. I can't wait to see those S1s. I'm rubbing my hands together thinking about it because I think the real economics of these companies are so bad. And exactly the same thing happened with Weiwork. By the way, that phrase I used, community adjusted bit, duh, was the wacky accounting of weiwork. Just how they found ways to be like, yeah, actually we're profitable if you just remove this and move this here and you take this, you add this, you take that away, that you multiply by this. I think OpenAI and Anthropic are doing the same dodgy stuff. And even if they're not, they're just bad businesses.
>> Well, on that note, Editron, thanks for taking the time.
>> Thanks for having me. If you enjoyed today's episode and again you'd like to keep me employed, please consider liking and subscribing. Also, if you're still not aware, you can listen to episodes of the Tech Report wherever you get your podcasts.