Is OpenAI a Goner? Don’t Miss Adir’s AI Deep Dive, Netflix goes Pacman, Bending Spoons goes Old Skool, Social Media Ban, Amex Kills Points and Pollies Behaving Badly

24:50Yeah. which some people have been able to get. I've been able to get. Qantas impossible to get something that's not, what's the plus thing, whatever it's called. You've got to get crazy lucky to get. I've got occasional, like without having to call the call centre, I've got to do it, but it's very hard. The day before jumping on a flight to Japan, for example, or LA, maybe something will open up, but for one seat, right? Yeah, exactly. But it's very hard to get. It's super hard. So is the answer to this, here's my question to you. Are you going to keep your card? I'm looking at other options. We've got to get Capital One to open in Australia and bring their Sapphire card and have a real card here.

25:25Yeah, exactly. I signed up to, on a tangential note, but I'll let you keep talking about this, but just before I forget. So I just get pumped with invitations, so a million different cards. When I say invitations, it is like it's free, but if you want anything, you'll get to pay. Oh yeah, Centurion, I endlessly get that. As if I pay 15K to join that thing. 15K now. 10 or 15K. For what? Yeah. It's the same redemption rate as MX Charge. I thought it was slightly higher redemption rate. Same. Oh, okay. Same. I would never pay it. Yeah. so for people that don't know because they're living on planet earth yeah the centurion card is like so already I pay what$1450 for the Amex charge card charge platinum charge platinum but at least I get I get$850 of flight credits that's pretty good yeah that's pretty good right and I think I can work out I can get to$1450 even without the points benefit like I basically get all my money back because there's a dining credit and some other stuff okay okay fine and you use that stuff don't even use it yeah I use it I've got a I've got like because I'm so bad at operations and I just wasted everything.

26:19I've now got a spreadsheet and my mum is across it and makes sure that I use the stuff. And so, but, so if you think, well, it's not fancy enough to have a metal, it's made of metal, a metal charge card in my wallet. Not that I ever use that thing. I use it on my phone, right? But if that's not fancy enough, then I want to go and get this black card. American Express Black. What's it called? Centurion. Now, I don't know. It wasn't great being a centurion. and your chances of dying were pretty high, by the way, in the Roman Empire, but putting that to the side. Like, you pay, I don't know, it's at least 10Ks a joining fee.

26:55Yep. $10 ,000 a year. The benefits are marginal. I never used a concierge service on Platinum. Yeah, it's... I never used it. I think the only reason people get Centurion cards is that it's like 35 to 40-year-old dudes trying to pick up. There's really no other point to it. Or, yeah, or something, let's call it ego. Yeah. It's something to do with ego, right? Yeah. Because... I'm so rich I can waste money on this ridiculous card because we know benefits. And so there is a business lesson to learn from the American Express Centurion. Yeah. When you go to the, have you been into their Centurion lounges at the airport?

27:25Well, they're not lounges for anyone, but yeah. They're called Centurion lounges. Yeah, of course. Yeah, yeah. So if you've got platinum, you can go in. But there's a little rope. And if you've got a Qantas Amex card, you can go in as well. Can you? Yeah. There's a little roped off area for Centurion card members. Yeah. I've never noticed that roped off area. I don't know. There's nothing different behind the rope. You get your food from the same buffet and you sit on the same chairs, but behind a rope. It's like there's a fine line between - It's like you're a cage, a little cage animal. No, there's a fine line between a private area and a prison, right?

27:51Like what's the difference? Or a cage or a goldfish bowl, right? But I will say there is a business lesson hidden in here, which is there is some proportion of people that will pay for something just because it's the most expensive thing. And so you should always have some version of your product that has some notional benefits that seem a bit ridiculous. and confer some status. I thought it was a classic status. That's like all it gives you. And price it at a ridiculous price because the signaling of that price increases the conferring of status. And some percentage of your customer base will probably buy that.

28:27It's a Birkin, essentially. Yes, probably. You can argue Birkin quality is maybe slightly better than other things, but really what you're paying for a Birkin is a centurion of handbags. I'm paying 40 grand for something that's really the same as the$1 ,000 version. So I signed up to this card called the Founders Card. It's quite good though, isn't it? You get that pumped into your socials. I mean, I'm not even on socials, but I think LinkedIn. Yeah, I think it's quite good. You can get it for free. Yeah. So I signed up for free. Yeah. And it says for$200 instead of$800 or something, you can get the version that will actually be useful to you.

29:00Yeah. Okay, I paid it. Yeah. I just was interested to see what it did compare to the Amex one. And as it happens, I bought something from Dell. Yep. A monitor. And I got made back in the savings with the code, all of the savings, all of the costs of that$200. So that was good. Basically, it's not very Australian-centric. It's predominantly US. They say that you can book Qantas flights to the US more cheaply, but it's never available when I've searched all the dates. It's like a higher fare. And some hotel stuff or whatever. And it gives you a whole lot of boosts up the status for hotels and stuff like that.

29:35So I think for$200, if you travel a lot, it's probably – And it gives you Stripe credits and things like that. Do you actually have to be a founder? No. You have to be someone with a few hundred dollars to spend. Okay. That's the prerequisite. And so, but, you know, it is good. I tell you what's good for founders. You get benefits too. I think you used to get AWS benefits, but you don't get them anymore. You get Stripe benefits and other things like that. There's a lot of services that give you credits. Yeah. And so I think if you had a small business, you could make it back on those credits. I'm assuming for like a catapult, you're not going to get$200 Stripe credit on your, on your thing.

30:10The hassle in getting that would be far more. But if I had a business, like if I was starting a business and I get a$500 strike credit, I'll take it, right? Yeah, totally. For sure. And it's like Google Cloud stuff. So I think if you've got an Amex Platinum and you're just like a, just a traveler in corporate or something, it's not worthwhile. But if you've got your own small business, I'd look at it. But you can always get it for free and then$200 to upgrade it. Like it's, there's always deals floating around. I was trying to help that Amex stuff. So part of the reason why Amex devalued obviously it's a great profit boost for Amex and Qantas to be in now.

30:43They've been able to use the cloud cover of the ridiculous RBA, Jim Chalmers credit card fee whole debacle, which has now been pushed back three months and may not even happen. It won't happen. So all that's happened is these morons, arguably one of the worst treasurers ever, Jim Chalmers, and these lackeys at the Reserve Bank, have just screwed over every cardholder who's now got worse terms because the bank's trying to hide behind this ridiculousness. What a disgrace this has been. I would say... Thank you, Jim Chalmers, for screwing every Amex holder. If you have an Amex card, you should vote against Jim Chalmers.

31:16Absolutely. But I suspect everyone with an Amex card is already voting against Jim Chalmers. There's plenty of Chardonnay socialists who will be supporting him, I reckon. Maybe. We'll let those... I can't even say the word Chardonnay. Chardonnay socialists. That's a tongue twister. The problem is you've got an incompetent opposition as well. Well, the thing is that all of those people that wanted to vote for the left, they're now going to, like, they care about whatever they care about to vote for the left, but what they care about more is redemptions to like Bali or Fiji or the Maldives or whatever.

31:44Oh, they won't redeem to Maldives, they'll book it with luxury schemes. But I think they care about that more than about like climate change probably. Oh, totally. And so I think that's the end of those votes. Yeah. Are you going to keep the card? Oh, I'm actively looking for alternatives. Yeah. I want to get rid of it. Yeah. Like, it's kind of useless. I'm principled a bit. Like, the fact that they've constantly made it remember it used to be one point now it's two points that's ridiculous basically a point now worth and then they smoke that's what you're saying is they smoke screened it yeah they basically said what we're going to do is we're going to give you double the number of points for every dollar you spend oh that's amazing but we're just going to increase the redemption by more than two weeks oh but you won't know because it's a smoke screen i think if you look at from is there any worse will yeah i can tell you the bad will is there any credit card with worse will towards its high spending customers than Amex.

32:33Well, I think they're all bad. Amex is no worse anyway. They're all bad. And that's why Amex has gotten worse because everybody else has gotten worse. It's been a devaluation across the whole ecosystem. Should we start a card? How hard would it be to start a competing card? Luxury Escapes is starting a card. We'll hopefully have a card in market in about six months. Very generous card. Surely you won't be able to compete. Very generous card. More generous than Amex. But redemptions are with Luxury Escapes only. Yeah, Luxury Escapes. But what else would you want to redeem? The best redemption in the world.

32:57I'm all in favour. But ultimately, lots of people want to redeem with airlines, right? Does anyone redeem with hotels? Yeah, I redeem with Bonvoy and Hilton all the time. Are they good rates? It's a really interesting question because it depends how you value business class because business class you can argue the ticket price of$15 ,000 or whatever they charge. A, you can get it cheaper if you buy early birds and B, a lot of people wouldn't pay that. So people who are redeeming would never pay the$8 ,000. You've got to pay the hotels. That's a great point. Whereas hotels it's a real redemption.

33:27No, I think that you're absolutely right. Like the real beauty of these airline frequent flyer programs is they are aspirational in the sense that you can do things that you otherwise couldn't afford to do. You're absolutely right about that. And the rest is corporates. Annika Wells, who's paying$34 ,000. But for normal people, most people for leisure travel, I know a lot of wealthy people who travel for leisure, who fly economy because they're not spending$40 ,000 flying their, or$60 ,000 flying their family 12 hours each ride, which is ridiculous. So you know Jay Kel? Of course. Jason Calacari.

33:57So I recently saw that he talked about where the founders should use the points accrued running their startup for personal travel or whether they should pump them back into the business to use them for business travel. His summary was typical American Silicon Valley summary. Why are they having personal holidays anyway? They're a founder of a business. But putting that to the side, what are you prepared to say? Your view on that? For all my flights of electric escape, I use personal points. which I've earned through the luxury of its games. So I think that founders, it's totally within their rights to use the points they've earned running their business for personal travel.

34:38No, I don't think the two, like, sorry, is it a question they can use for personal travel or they can use it for both? Like, I think it should be for both. Yeah, I think that the argument was, someone posted this something which said, I would never invest in a founder that uses points that they earned in their startup for personal travel. I totally disagree. I disagree with that. I think they can use it for business for sure, like they should, right? But I don't think there's anything wrong with using it for personal travel. I think the point I'm making is if you're flying business class on the points and then making the company pay for business class, I don't agree with that.

35:07I think you should use the points for all your work travel and personal if you have left over. So what if I set up a cooperative where I'm not trying to make any money as a credit card business and everyone who joins is just a member of the cooperative and every cent that we get from all of interchange fees or whatever else there might be, it's a charge card, not a credit card. so you have to pay it back every month and every cent of profit goes back into maximising points given to the members. That's a good use of a cooperative. I'd sign up to that. What you should do is create an affiliate program and instead of getting cash back you get points back.

35:43Yes. Because it's more tax effective. So I can tell you like pulling back the curtain that back in the day it's still the same today if I went for like a hotels.com the cash back that was the highest like you can get high cash back but for big ticket items that's a big ticket item that was the highest maybe five percent you get back and so you can buy a lot of points with five percent absolutely i mean that because like a credit card's giving you one percent yeah of spend and so if if that yeah so yeah that's interesting just funnel it straight into points yeah that's interesting shop shop that's an interesting shop back option i told you i use cashback non-stop yeah non-stop king yeah non-stop we'll go to a super quick break but back with some great stories just in a moment

36:36and we're back and have you heard of a company called bending spoons i can't talk about this in detail but i know bending do you know where they're based um i do know exactly a hot spot of i do know where they're based i've actually forgotten where they're based italy that's right Bizarrely. That's right. Europe's not exactly a hotbed for... Their name is unforgettable. It's a great name. It reminds me of Uri Gela. Yeah, absolutely. So this is a business that buys and revitalizes stagnating tech companies. And it was actually valued at$11 billion recently after announcing a$270 million funding round.

37:11So they've done an unbelievable job. They basically buy businesses, intend to hold them forever. Almost like a Berkshire Hathaway for tech in many ways. Why don't you say it's like a... Constellation Brands. They're basically based on Constellation. Constellation buys profitable businesses. These guys kind of buy... Not only. Yeah. They tend to be more... Constellation's a bit more Charlie Munger and these guys a bit more Warren Buffett. Like it's a bit more sort of... They'll buy unloved businesses whereas Constellation tends to buy small, like decent... They've got the advantage though that if you're one of their CEOs that are running their business, you get to go and meet them in Italy.

37:48Yeah. I mean, that feels good. So, actually, back in the news last week after agreeing to buy Eventbrite, remember Eventbrite? Of course. Started by Kevin and Julia Hartz, Kevin was a really early investor in Airbnb, and early BC, back in there. What do you mean, do I remember it? Because I still get Eventbrite links for stuff. Yeah, but there's a lot of competition in that market. So, they've agreed to pay$500 million for Eventbrite, a far cry from the$1.76 billion it was worth when it was in public seven years ago. So that's not been a great story there. Do you know its numbers? Are you about to tell me its numbers?

38:23Or your annual revenue was flat at$325 million for 24 and 23. Well, that's a lot of revenue. I mean, it's not in the billion-dollar league, but it's not a minnow. Yeah, and that sold for$500 million, so it kind of indicates just how far it's fallen. One times revenue, but one a bit times revenue. Is it public? It's public, yeah. Does it make any money on the$325? I think it's just getting profitable now, like barely. I wish I would have bought this business. Yeah. I mean, can you imagine what you can do with this business? Oh, these guys will kill it. They'll do really well, not kill the business.

38:56So Eventbrite, it's a similar low multiple, but Eventbrite shareholders really$4.50 cash per share, 81 % premium. This business was basically trading at under one times revenue. So the$500 million is an 80 % premium? Yes. It's crazy. Because basically what's happened with this stock price, we can say what's happened with this stock price is because that means what was they're trying a 300 mil or something right before the acquisition yeah and so yeah less than one times revenue what you just said so this is what's happened investors have looked at this and said there is never going to be a catalyst for a re-rating of this stock yeah they are never going to figure out how to make money it's a lot of competition they are never going to grow yeah and so what i've said over and over again and i'll say it again is that with especially in public markets with institutional investors.

39:44They are waiting to put their money in to get in just before a catalyst comes and re-rates because every month that they have their money in a stock that does nothing is problematic for them because they're reporting, right, to their investors. And so here there was never any FOMO to drive the share price up and to attract people's attention. Maybe originally it was listed, but then it was... No, but in the last two years. It's just too much. It's just a commoditized product now. There's literally no... There's 325 mil of revenue. That's a lot of sales. Yeah. Global business though. And a lot, but$3.25, I mean like, I can't even imagine what you pay for with Eventbrite.

40:19Oh, you're paying for like, some events are free but if you're having a, like it's like Paperless Post which is actually a really good business. Yeah. Paperless Post has actually smashed it. Is Eventbrite, when I get an invitation to do anything, is that on Eventbrite? Yes, that's pretty much still a dominant platform. The digital stuff I send is Paperless Post, which is much better for sort of call it gatherings, like parties, gatherings, where Eventbrite was a bit more, they do ticket, they do actually big ticket events as well. or something presumably. But 325 mil, I don't know, like I feel like, shouldn't they be able to make 100 mil of profit off that 325?

40:50It feels to me, it's software. I mean, I don't really understand how these businesses don't make$100 million of$325 million of software businesses. And betting spooge itself has a really interesting story. It was born out of a business called Evertail, a Copenhagen-based startup that raised seed funding, had a photo sharing app called Wink. That business failed. Investors exited, but the founders and a couple of employees kept working together on some apps. In 2022, it started acquiring a bunch of businesses after that. In 2022, it acquired Filmic, a video and photo editing app, and laid off its entire staff a year later.

41:23And then it bought Evernote. Remember Evernote, which is a really popular note-taking app, which was worth a billion dollars, and they bought it for a song. That kind of had a spike up and down, that thing, really. I mean, it was big in 2013, 2014. But really big. I remember, were the founders maybe Eastern Europeans of Evernote? It was really big. It was worth a billion dollars. It just plummeted. And they paid a song for it. And then they sacked everyone. You see my theme here. And in 2024, they acquired Meetup, Mosaic Group, and StreamYard. And then they bought WeTransfer, which I use a bit.

41:55Remember WeTransfer, the transferring, if you want to transfer a big file, which is really good business. I've never used it. And they obviously get rid of all the free tiers, which it seems like their model is get rid of the really generous freemium and make people pay. And the big one was they announced in September they paid$1.38 billion for Vimeo. That's right. which was a 91 % premium to its share price. So these guys are the masters at finding these unloved, hated fallen angels, and that's an amazing job. They're worth$11 billion now. Yeah, it's very interesting. It's a great story. Moving on, that was a really interesting story.

42:29And between them and Constellation, I know Constellation's share price has dropped, but it's still been an incredible story. So there's plenty of money to be made in buying profitable, turning businesses profitable, heaven forbid. Well, I'll show you the other extreme version of that. at the other end of the spectrum. I don't know if you saw this week, but Quadrant finally got out of Adore Beauty. I did, yeah. They sold their last equity. Adore's doing well. They're turning around. Well, yeah, listen, their share price is above its low. No, but they're actually making money now. Tamalyn did an amazing job turning it around and she obviously had to move on for personal reasons, but I forget the new guy's name.

43:01He seems to be a pretty steady operator. So they're doing a great job. The thing is, well, let's... Off where they were. I think they're really working as hard as they possibly can and they're smart to try and do stuff with it. Like, I still do not think... I wish I could... I wish Mecca was in the public markets and I could just do a pair and go long on Mecca and short on the door beauty. Mecca is just an incredible business. It is, but I don't think they're going to... They're not going to get any ground off Mecca, but the point is this... But they're making money now. They've turned it around.

43:29So the Quadrant, they bought half that business for nothing, like 50 million bucks. For next to nothing. And then COVID came. Yeah. And then they floated this thing for 650 million. and I think they might have sold a quarter into the IPO. And so that's 150 mil thereabouts. And so this last 30 mil doesn't really make that much of a difference. And the thing is, like, and then Kate Morris went away with Justin Ryan to start Glow Capital. So Justin obviously was the guy, I think, that did the investment at Quadrant. Like, you know, timing is everything, basically. And so everyone is a winner, except for the investors are bought into the IPO at$6.50 and the share price at$1.50.

44:07No sympathy for those speculators. There were smart people, like genuinely smart people, that said to me at the time, nah, this is a great business. I think what you just said is a complete oxymoron. No, but they're otherwise - You could not be a smart person. They are, they're otherwise - Sorry, the business is fine. The valuation was crazy. Do you think the business is fine? That's interesting that you say that. Fine, I don't say great. I say fine. It's not mecca. I don't want to re-prosecute a door beauty on this episode, but like - Oh, it's okay. Yeah, it's fine. But like, I don't think it's a fine business.

44:31I think it's - It makes money. It's a fine, but it doesn't lose money. Is it Templer Webster or Adore? I have to look more closely. I'll take Templer Webster because of scale, right? And Templer's a great brand and they've got a lot of stuff going for it. Yeah. We just don't like the valuation. Yeah. So anyway, I think it was very interesting to say. I wonder why they just decided to sell it. Now, I mean, the share price has risen off its lows. So I have this recollection. I think I wrote about this. So I have this recollection that they got a takeover offer at$1.20 and the founders rejected it.

45:03and so did Quadrant, and so they didn't want that deal. And then it plummeted, and now it's managed to climb its way back up, and as of recording this, it's a bit under$1.30. Yep. And so it just goes to show... But you want to have floated it in the first place. Yeah, but I thought they should have sold it at$1.20. Yeah. Like, take the deal. Yeah. It just seemed weird to me. It was conditional. There was a few things on that deal. Well, we didn't. It was like an NBIO, right? Non-binding indicative offer. Yeah. But don't dismiss it. I don't know if they thought somebody else was going to come and offer more.

45:35I'm not. Take the deal. Yeah, probably, but it's not like it's gone to 50 cents. How many times have we seen businesses knock back a deal and sell them half? It did, it won't. It plummeted. But it's back to, it's above the deal now. All right, but it went down. So the decision was right. It actually did go down. It might not have gone down to 50 cents, but it would have gone down to 60. Would have halved. I know, but it's done well now. So the point is that they made the right call. Do you understand the idea of the time cost of money? But it's only been a year. It hasn't been 10 years. No, it has been more than that.

45:59It's just rates of 4 % at the time cost. I don't know. Well, it's definitely rocketed since July because since July it's doubled. They're making money now. I was critical of them but I think they've actually done a good job. I'll just tell you the money. Credit where credit's due. Credit where credit's due. I don't know. Have you ever bought shares in this or something? No, not at all. I think it's unfair to compare them to Mecca which is one of the greatest businesses raised they've produced. I'll compare them to this. Somebody comes in and they say I'm going to offer$1.20 and you say nup too cheap and then you wait for a couple of years ride the roller coaster.

46:30That's two bucks. No, but it's not two dollars. But it could be two bucks in six months. It could be lots of things. There could be lots of things. Everything could be anything, okay? The world is an uncertain place. But what you know is this. The same party that didn't want the takeover offer at$1.20 just sold their entire stake at about$1.20 a few years later. What does that tell you? We don't know. Take the deal. We don't know if that was Quadrant who didn't want to take it. Who knows who didn't want that deal? Well, Quadrant would have had enough of the equity that if they wanted to take it, there would have been a lot of pressure.

47:00possibly but they couldn't force the deal through 25 % of the stock or whatever well they could have said we want to take the deal they even have 20 I think they did have 20 anyway anyway I'm surprised you're so dismissive about this I'm not dismissive I'm going to go you know what I'm going to get a private investigator onto you that's going to follow you around and see what you're doing inside a door beauty you're a bigger shopper obviously friends with Tamaline and O 'Hare we all like people I mean I think I like people even more than you like people and I think that yes she did a great job to turn that around and Kate did a great job to build a very strong brand right but I felt like this business has always struggled to make a reasonable return and it's never going to be the category leader in its space and they should have taken the deal at$1.20 that is all I'm saying yeah incredibly hard it's in the temple realm of really hard business to run like easier in a sense that you're not shipping massive stuff around but harder in a sense that it's so commoditized so if you own a business and I offer you $500 million for that business and you say, no, it's undervalued.

48:00I understand the time value of money concept here. It's not undervalued and then it goes to 250 mil and then it goes back to 500 mil and you say, oh, now I'll take your offer. Should you have taken it two years before? I'm familiar with the concept of time value of money but I'm also familiar with the concept of the business being worth optionality value as well. Yes, that's true. The only reason you shouldn't take it is if you're like, nah, we've got a really great throw of the dice coming up. We think we can make this amazing. I don't think they had that when the offer was$1.20. Maybe. In other takeover news, what's the big takeover news happening around the world that you'd love to talk about?

48:32The non-takeover takeover that won't get approved, that takeover? Well, somebody else could buy it, yeah. So that's the Netflix purchase or Netflix potential purchase of Warner Brothers. The Netflix no chance of closing purchase of Warner Brothers? It depends who's running the... Is it FDA? I think you mean depends who's running America. Oh, who's running America, exactly. Well, I can give you a tip on who's running America, Donald Trump. Thank you. Any other tough questions? Come to me. I'll answer them. So Netflix is offered $72 billion US after the company splits off its studios. After Netflix said, we're under no circumstances are we going to overpay for this business.

49:11Yeah. And then all of a sudden they push all their chips into the middle of the table. So Netflix has outbid rivals Paramount and Comcast. Paramount, of course, now owned by the Ellisons and Comcast owned by the Roberts family. The deal stunned Hollywood where many assumed that Paramount will be the most likely suit up. Well, they are the most likely suit up. Yeah, still are. So seven weeks ago, Netflix... They offered more money. Yeah, well, but Netflix is for... Well, I think there was... Netflix said, we'll give you X dollars. But not for the whole thing. Yeah, but like... Split that off. I know, but Netflix came to the table and said, we'll offer X dollars.

49:44And look, here's two bags of cash we're bringing into the room. And we've got debt finalised. Yeah, and if we don't do it, we'll pay you. You know what the break fee is? 5.8 billion. So crazy. I would be praying the deal doesn't close. unless I was, who's the dude that runs it again? David Zaslov, who's the most highly paid. Zaslov. Is it? No, Zaslov. His name's Zaslov. The most highly paid hack executive ever to live. He could get a billion dollars, this guy, and he's completely stuffed the business. He won't get a billion. He'll be worth a billion. Yeah. He won't get a billion for this. But is Prime not bidding for this Amazon?

50:17I just know they can't get it. Why? Because they had trouble with MGM. They had trouble as in financing it. No, to get clearance for MGM was really hard. There's no way they're going to get something after that on top of that. But isn't it just whoever's nicest to Trump gets the asset? There's no other way. Bezos is a good mate with Trump nowadays, isn't he? He gives you money and stuff. He's better than he was. Can you believe the conversation we're having about America nowadays? Yeah. It's actually crazy. As you said, kleptocracy. So Netflix co-chief executive Greg Peters played down the idea like several weeks ago, as he said, saying, big media mergers don't have amazing track record over the history of time.

50:51And never true of a word. to speak and almost everyone's been terrible uh on the history on the heels of uh paramount's merger with skydance david ellison submitted several unsolicited unsolicited bids for warner discovery including its cable channels they were all rebuffed and warner of course started the sales process because your argument which i didn't really pick up on this but you're basically saying although the ellison offer was higher yeah number one it was they had some finance questions about it but number two it was for more of the assets i see so this is the so Warner's had a history of like so many terrible deals.

51:25Like even going back to the 80s when there was run, like Steve Warner was running it. Then they bought Atari. What a disaster. Almost bankrupted the business in the 80s. I think somebody came and - Good logo though, Atari. Somebody came and say, it was like, remember Atari went, they bought it in like 1976 from the founder who was Nolan Bushnell who also found Chuck E. Cheese. And it was incredible for a couple - I love that you know that. Yeah, it was incredible. And then just absolutely died. It had all its stock in like 1983. It was a disaster. But don't you think, have you seen the Atari logo?

51:51That's one of the coolest logos of all time. It's like a Space Invader. It's almost worth just buying it for that. Yeah. And then they, so they had the Atari debacle and then they bought Time, became Time Warner. Then of course, the greatest disaster of all time. Do you know what the greatest disaster merger was of all time? Well, I thought that was the greatest disaster. Not Time Warner. AOL Time Warner. AOL. Oh, yeah. People talk about AOL being a disaster. AOL were geniuses. AOL sold this disaster. So AOL was, quick background on that, AOL was flying as a business. It was doing really well, but it was like, it was like basically OpenAI now.

52:20I mean, all these circular deals, all these terrible dot-coms were about to go bust who were paying it. As soon as the dot-com crash happened, suddenly AOL's revenue just completely evaporated. They managed to con Time Warner into buying it into like this sort of 50-50, roughly 50-50 merger. So it was a great deal for AOL shareholders and horrendous deal for Time Warner shareholders. So basically AOL became worth nothing, essentially. So it gave away half the business for nothing. Then they had this AOL, eventually got rid of AOL. And they did this discovery deal like three or about five years ago where they bought Discovery all these crappy cable assets like Discovery Channel all this kind of stuff they merged with these premium HBO assets so they'd stuffed up again and Zaslov just conned them into this deal top of the market deal and now so they literally had a history of bad deal bad deal bad deal bad deal now they've been acquired and another bad deal from Netflix by the looks of it I don't think they were very nice the Warner Brothers like this kind of is the karma the original Warner yeah yeah the actual brothers Jack Warner I think this might be the karma that comes back to you know I think they were known certainly Steve Ross who was running in the 80s, was known as being this really talent-friendly exec.

53:24So all the big stars want to work with you. He just massaged them, massaged our egos really well. He was known as being this sort of much-loved, sort of tyrannical much-loved. Is that Steve Ross, the property developer in New York, same guy? Possibly. I think he got his start in like garages or something like that, didn't he? He started from nothing. His business was Kinney National Company and then he bought Warner Communications that became Time Warner. So Kinney was essentially, I think it was his relative-owned part of it. His father-in-law, sorry. And there was underworld figures who were involved.

53:56There was a whole thing. But yeah, it sort of effectively became somewhat of a property guy. He bought Warner Communications in 1971, which was pretty early days. And you know what their first popular TV stations were that really made? Was. They had the movie studio, but they launched MTV and Nickelodeon, which I think eventually brought by Paramount. But then 76 bought Atari, obviously then crashed. He survived by selling 20 % to Chris Craft Industries, which is a big conglomerate, and then became a big cable business and made a fortune in cable business and merged in Time Inc. in what was a$14 billion deal back then.

54:35So everyone talks about this deal now, right? And obviously Netflix wants the deal because they think they want the back catalogue of content presumably. Well, back catalogue and forward content as well. And also Netflix. A bit of defensiveness as well. Well, but also I suspect this. So I think this about the Amazon acquisition and I think this about Netflix trying to. Amazon buying MGM, you mean? Yeah. I think there is some emotional moment where there's new media people that grew up fighting these giants end up owning the giants. Yeah, absolutely. The old giants. And that's the changing of the guard.

55:10And I think Netflix might have hard eyes for one of these old school studios that really started the whole movie industry, right? Like Warner Brothers was one of the originals. Now it's about HBO, really. I know, but I think that there is some glory in, you know, everyone knows what the Warner Brothers lot looks like. Yeah. And I think there is some glory in being connected to that. You know, we talk a lot about how retail brands love heritage. Yeah. But I think these guys have hard eyes for the movie heritage, right? But they won't get it. I think the issue is the price. You can forget about it.

55:44Warner was$30 billion not long ago. So had they paid$30 billion, I think that was a great deal. It was assuming they could get it through. But at$72 billion, this is a business that only makes like$10 billion. So what's Netflix worth nowadays? It's about$400 to$500. Right. So this is going to be significant dilution for them. I mean, it's like close to 15%. They're using debt. All right, so 10 % dilution in debt. Now, Netflix is a great business. I mean, it's a terrific business, but I don't know. I mean, you'd want a lot of content for$72 billion. So Netflix market cap is$425 billion. Wow. It's dropped off.

56:21It's dropped about 30 % from its peak. This is almost 20 % of their market cap. Now, I know they're not doing it all with debt. Now, they'll use some cash, use some debt. Yeah. They're not doing it all, sorry, with equity. It's a very big thing for them to bite off and swallow. Yeah, it makes about 10. It's a multiple of 41. It makes about$10 billion a year. So this is seven years of earnings, technically. Yeah. At$30 billion, no brainer. I'd love to know. At$70 billion, it's a lot of debt. So I'd love to know how much value they put on the back catalogue. Because presumably, you know, this is how these deals work.

56:52When they go to the investors, they say, if we buy this back catalogue, we don't have to spend as much money on reducing new content, right? Back catalogue's worth$10 billion, mate. It wouldn't be that valuable. But they'll say, oh, we're going to get all this content and we would have had to make that content ourselves and whatever. But they're not going to reduce their spend on content. Like that's going to be a non-real argument, right? Well, I run two brands here essentially. So what do you think Netflix can do to get this deal across the line? So everybody hates it. The whole industry hates it.

57:18Yeah, the industry hates it. The talent all hated. And so I don't think Trump hates Netflix. I think he just loves - Well, I think Reed hates it. He loves Ellison, right? I think Reed hates it as a Democrat as well. So there's probably a bit of that. Right. But also Ellison is like one of Trump's. Yeah, absolutely. He loves him. Yeah, of course. And so do you think there's any chance that Netflix can convince Trump that this deal should happen? I think it's going to be hard. I don't see how it can happen. Yeah, it's going to be really hard. Ultimately, just go on for age. I think a big part of the reason Netflix is doing this, they don't want Ellison, Paramount to get HBO because it becomes a real competitor.

57:57Disney is the only real competitor and then there's a bunch of smaller ones, so Peacock and HBO Max and Paramount. They're all too small to count. I think if I'm Netflix, and this is the I don't agree with the 72 billion but this is the rationale and the rationale makes sense so the beauty of Netflix now is compulsory you have to get Netflix everything else is optional but Netflix you can't argue about and if you've got kids Disney you've got to get but Netflix is what as in you mean if you go and buy any streaming service it's going to be Netflix I see is that right and because it's the default streamer I don't really use it so I don't know because it's the default streamer you kind of just leave it on like other people you come in and out of Paramount you come in and out of Stan you come in and out of everything but Stan actually got a bit of stickiness here because of the Australian content but Netflix is really the default and I think the worry is if it didn't get Warner, maybe Paramount, HBO becomes good enough to be a default as well and Netflix lose a bit of that kind of clear number one.

58:50I said to you that Eden is sitting behind me. Do you watch Netflix? You can talk into the microphone if you like. Do you watch Stranger Things, for example? I watch Wednesday. Is that on Netflix? K-pop Demon Hunters? No, I don't know the hype about that. would that be your would Netflix be the number one platform that you watch shows on streaming platform Netflix and Paramount probably interesting there you go well I can tell you I don't watch anything on anything and so I have the highest dollar to watching ratio paid for streaming you pay for everything no because I pay for everything for my kids and I watch nothing I actually watch nothing I think that's a great line Netflix and Disney are probably the ones that get the biggest Star Marvel Star Wars I guess you can watch a bit of yeah Yeah, and then everything else, yeah, in and out of, like when something's there you want to watch, like, oh, go there and then hop off.

59:41Do you do the Netflix with the ads? No, but my family pays, so I'm not too. Oh, it's not your problem. I think it's ludicrous that they're running ads when you're paying for the service, whatever price it is. Oh, the ads, so if you pay for non-ads, do you still get ads? No, no. Oh, I see. You pay a lower price for Netflix with ads. You know what I find weird? for like 50, 60 years, Free to Air TV worked where you didn't pay a cent and you got ads and the subs of it. Cable, you had ads and you paid. So it's not as if it's unheard of. Cable, you had ads. Cable had ads. Oh, that's true. That's true.

1:00:13Like Foxtel or something. Yeah, absolutely. Oh, that's true. But less intrusive ads, but it has ads. Why, how come Free to Air could just work with ads and you didn't have to pay? It had more ads, I guess. It had more watches. It had more scale. It had more scale. It's so much scale that when you get to scale, you can charge more for ads. because there was like three channels. So one of the channels might have half the Australian viewing audience at any point in time. In the 80s it was nine. Yeah, it was massive. And even 10 had the demos. That's a good point. So yeah, I think you're probably right that it's going to be hard to get.

1:00:44And there's obviously a 5.8 billion breakup, which is, so Netflix, so I'm presuming... So they pay half of one year's earnings, 60 % of one year's earnings 60 % on a breakup fee, which is incredible. Cue that song. Who sings that song Desperado, that old song? That should be the background song for this episode, sorry, for this segment because Netflix is very desperate here, right? When you're paying 60 % of one year's earnings as a break fee, that is desperado. It's so un-Netflix-like. Netflix has never paid more than 700 million bucks for a business ever. Would they buy for 700 million dollars?

1:01:16I don't know. Who knows? Irrelevant. Yeah. It's worth 400 billion. What's 700 mil between friends? They really should have done this earlier. They had the chance to buy it a year ago. What if I postulate this to you? They don't want to buy it. Yeah. They didn't want to buy it. They never wanted to buy it. They even think it was overvalued at 30 mil. They didn't want to buy it. But they feel like defensively they have to buy it. That feels the only reason to buy it, is to stop a competitor challenging them. Well, I've got good news for them. They're not going to get it. So they're not going to have to spend the money.

1:01:44There you go. I might be totally wrong, but it doesn't feel like there's any road for them. Yeah, I think it's hard. Let's make a massive donation to Trump. And you know what they should do? A hundred million to Trump. Who's the dude that runs FIFA? They should go to FIFA. And they should say - and now there's a lot of cooks who run FIFA. They need to create a peace prize with a gold trophy and give it to Trump and they still might get done. Also, actually, there's a lot of overlap between Netflix and Paramount and stuff. A few Nickelodeon shows are streaming on Netflix and Paramount. Like what?

1:02:18Victorious, iCarly. The same show as an older series. Some of us share the series, like early series on one. These are remakes or something, right? Well, there's a remake coming next year. Well, spin-off coming next year, Victorious. That's going to be on Nickelodeon, Netflix and Paramount all at the same time. They all have the rights. Yeah, so it's weird, right? I don't really understand how this stuff works. On that note, we've got a super quick break back with our great deep dive in just a few moments.

1:02:48Thank you. over decades. Thinking of selling? Discuss how it might work at terim.capital slash contrarians. And we can't go past the absolute dumpster fire at OpenAI. And of course, last week, Sam Altman, the founder of OpenAI and chief executive, told employees that the company was declaring a code red, which is pretty common in Silicon Valley, which is basically an all hands on deck focusing on one thing to improve the quality of CHAPT, delaying other products as a result. According to an internal memo viewed by the Wall Street Journal, Altman said OpenAI had more work to do on the day-to-day experience of the chatbot, including improving personalization for users, increasing speed and reliability, and allowing it to answer a wider range of questions.

1:04:00Of course, OpenAI's main competitor now, in the consumer space anyway, is Gemini, whose user base has been climbing since the August release of an image generator, Nano Banana, and of course, a couple of weeks ago, its new version of Gemini. Google said monthly active users grew from 450 million to 650 million, even though that includes, I think, people searching on Google itself. So it's a bit of a misnomer. Altman said, OpenAI, interestingly, pushing back on other initiatives. And this is really critical, including advertising, AI agents for health and shopping, and a personal assistant called Pulse.

1:04:31I do hear views on just what's happening in OpenAI. Well, I want to tell you, I said last week, I want to tell you the AI stack. So I think OpenAI is doomed. I think it's going to end up to be a subsidiary of Microsoft in three years' time. Or even less than that. at a very cheap price. It basically sits on the Microsoft stack anyway. I think Microsoft own a huge chunk of it. They own a chunk of it. Effectively own a huge chunk of it. But also Microsoft is incorporating all their stuff into Microsoft and it sits on the Microsoft stack. It sits on Azure. It's somewhat reliant on Bing, as many things are.

1:05:01Yeah, by the way, your perplexity that you love so much, that's also using Bing stuff. I thought it uses heaps. Like it has to index web pages. And so it has its own indexing, but also it has to use someone's index. Yeah, it's good, but it also lies a lot. So I want to tell you this stack, which I think you'll find fascinating. I'll try and do it efficiently. There's five steps in the stack. I've cut some out. You'll be familiar with many of them. So I'm starting from the interface that the consumer touches, the consumer business, the large language model, all the way up. Oh, it's in reverse. Remember a few weeks ago I talked about the other way around.

1:05:41I think it's in a way easier to understand in reverse. So... This is the GPT, essentially. Yeah, the bit that you touch, that we touch, is a piece of software. Yeah. It's a large language model, which can do generative AI, which is like when it writes stuff. Well, isn't the large language model one level below and the GPT is the top top? It is, but I'm going to bundle this all into the software that you're interacting with that's either helping you write stuff, like generative AI, or diffusion models, which are doing the images that you're using. Yeah. That is just a software layer. And so the way to think about that is lots of different companies are training models and then running models for you to interact with.

1:06:23And so the way to think about it. Less everything from OpenAI to Anthropic to DeepSeq. Anthropic to all of them. They are just software. It's just software. And so there are two bits to this, and this is the important bit for the next step up. One is training the model and one is running the model. Now, when you train the model, you have to teach on millions, hundreds of millions of bits of labeled data. Can you explain to us the difference between training and inference? Yeah, so inference is the model running, you talking to it, you asking it to do stuff. Training is hard because training involves hundreds of millions of pieces of labeled data.

1:07:04They run through this matrix of numbers at the bottom. So you say, we're going to show you something. What do you think this is? and then it does all these random calculations. And at the bottom it says, a boat. And you say, nah, it's an apple. And so then the model has to run in reverse to try to recalibrate the maths so that the output would be an apple. So it's running in both directions. And you've got to do that like hundreds of millions of times. That takes a crazy amount of processing power. So that's training, okay? And at the end, the random maths should be such that when you show it an apple, at the bottom it says, that's an apple.

1:07:39I'm talking about images. But it does something similar for Gen AI, just does it with words and concepts. And the way that it stores all of this information is it links things together with patterns. You're basically teaching it patterns, and that pattern linkage is called vector space. That's what vector space is. So that's hard to teach. Then you come along and you say, what do you think is the best flavor of these macarons that are sitting in front of me? That's actually pretty easy. You're asking one question, like it's a one prompt thing. It's one output. It's happening only down the model, not in reverse.

1:08:13And it's pretty easy for it to do that. That's inference. And so that does not need a lot of processing power at all. And so you have to remember those two things. Training, lots of processing power. Running the model, not lots of processing power. So that's what I'm going to say about models, okay? But my understanding is when you're – but a lot more cost has been consumed in inference now because so many people are using the models. Well, we're going to talk about why so much cost is being consumed in inference and whether that's going to continue in the future because it's not. Inference, you can run on a Raspberry Pi.

1:08:47You can probably run it on your electric toothbrush. I mean, like it doesn't need much power and much processing power to run inference. Running the models is not where the cost is. The cost is in training the models. That's where it should be. It is there for a reason we'll come back to in a second. So in order to use these models, like I'm going to go one step up and you're going to be maybe the unusual place to go next, but next I'm going to go to the data center. So when you go and type into your ChatGPT, you're actually sending a message into a data center. So what is a data center? The easiest way to think about it is it's a big box, a big empty building, and it just gets tenanted out.

1:09:26A bunch of computers in there. Well, there are no computers in there. There's nothing. It's an empty building with racks. and you might say, well, why is it just a warehouse? That should be pretty straightforward and that's like the next DC or the air trunk and so what's so hard about an empty building? Because actually it really is an empty building with racks and stuff into it and then Amazon might come along and say, we're going to rent this whole chunk of a building. It's like an apartment building but every tenant gets to design and furnish their own apartment, right? So why is that? But I'm going to say to you, there is a big barrier to entry that's a genuine moat in these data centres.

1:10:04And the reason there's a moat is because the data centres have got two big problems that they need to solve. Power? Power and water because they need cooling. This all runs very hot. I'll give you an idea of how crazy power is. So in 2018, AirTrunk was trumpeting a data centre that it built and it plugged 20 megawatts of power into the data centre and it was going to increase dramatically with another 10 megawatts of power. The data centre that Air Trunk is trying to build now, how many megawatts of power do you think that one is? Eight years later. So 1 ,000 megawatts would be a gigawatt. So they do want to build a gigawatt data centre, but that's a bit aspirational.

1:10:43Well, not super aspirational, they're trying 400 megawatts. Now, that means in the space of eight years, the data centre has increased by a fact as 20. I'm supposed it's only been 20. Well, let me talk to you about what 400 megawatts is. Firstly, they couldn't get 400 megawatts. The government, or not the government, the supplier said... How much power is Colossus using, or the new Colossus? Let's come back to all of these, but let me tell you what 400 megawatts is. So the government said, we can give you 300 megawatts. Australian-based. Australian-based. This one is in Sydney. We can give you 300 megawatts.

1:11:19The last 100, we can give that to you, but we might need it back a few times a year, like when it's very hot and everyone's wearing their air conditioners and obviously you can't do that because you can't say to AWS, yeah, we can run your servers but like when it's hot, a few days a year we have to turn them off. So the problem is 400 megawatts. How many homes, Mike, do you think 400 megawatts powers? In how long? For a year or for? 40. 40 for how long? The question's only half asked. Well, that's a good question. So 400 megawatts, the way I think about it is that's the maximum power that it can draw at any one time.

1:11:53So like how many houses could you run at any one time drawing 400 megawatts at a typical moment of houses? 100 ,000. What do you think, Mike? 50 ,000. Joel? 50. Okay. That's a pretty big house. 50. So you're all wrong. 200 ,000 houses. That's 400 megawatts. One data center, the instantaneous power it's drawing can run 200 ,000 houses. So I don't know if you saw, but in Queensland, they just announced this new gas-fired power station. And so guess how big that gas – they trumpeted it. Yay, new power station. Guess how many megawatts at any time? 100 megawatts? 400 megawatts. Same as that one data center, instantaneous power.

1:12:39That's how – so – and like the water is – I don't know. It's like millions of liters a day to cool these things. so there is a so definitely there is a moat can i reuse the water or has to i presume that there's some recycling but i don't think it's that efficient at the moment isn't that all of the curtis's thermos business has something to do with the cooling well we'll get to that because that's not a data center that's a neo cloud we're going to come to that so exactly it's like core wave so we're going to come to what that is but these data centers have a competitive moat the existing ones it's hard to get regulation it's hard to find the land it's really hard to get power like approval to run that much power and water into them.

1:13:16But one thing you might think about is, is it possible that we are going to be able to keep increasing data centre capacity on the exponential curve that we've had to date? I would say no. Well, there's been real issues in the US about this because nobody wants a data centre in their backyard. Yeah, it uses too much power. It uses too much power. It's not coincidental that all the data centres being built in Australia are in New South Wales and Queensland because even though there is a shared grid in Australia, Victoria is yet another example of the government destroying energy in the state is that nobody will build data centres in Victoria because the power is too unpredictable, basically.

1:13:48And, like, can you imagine trying to build a gas power station and you run in Victoria? I mean, it would never happen, right? So the industry is basically a New South Wales industry with some of Queensland. So that's the data centre. Then there's a neocloud. A neocloud is basically, instead of, like, the data centre being a building and everyone coming and building their own apartments, And what NeoCloud does is they say, we're going to build a fully furnished apartment and then rent that apartment out to other providers. We'll come back to that because that is what you might call a circular economy.

1:14:22But the NeoCloud is firmness, which is being talked a lot about. And CoreWave, which has dropped about 50%. Right. And you might've heard them advertising on the Acquired podcast, Crusoe in the US. Oh, Crusoe, yeah. That's another one that's moved into NeoClouds. What does Neo mean? Neo, new cloud cloud. cloud so it's a nonsense term right but they so they might own a data center they might not they might just use space in somebody else but the point is whether they own it or not they're renting it out fully furnished yeah so that's the data center now we move up one level the the cards because inside these data centers you put these cards inside gpus so gpu stands for graphical processing unit whatever these cards do they do not process graphics yeah that's not what they do Can you also have TPUs in there as well?

1:15:09We'll get to what TPUs are, okay? TPUs are Google thing. Google ones. So let's start with GPUs. So I'm going to do one snippet of history. So there's something called a CPU, Central Processing Unit. That's this Intel thing. You know, the first one was the 8088, and then they had 386s and Pentiums, and they got i9s. So what these are very good at doing is these are very good at doing complicated, very sophisticated calculations. calculations and so when they made games to begin with and so i'll tell you what a complicated calculation might be i'm using my computer and i want to print in the background in excel whilst continuing to use word that is a very complicated set of um orders to process and so you need a cpu to do that okay and then you see amd cpus coming as well right and so um and so that's what a cpu is and so when they made the first games they got the cpus to do this and cpus are good at handling like all of the calculations in games like physics for example but then what games need to do is they need to do like a simple task lots of times for example a whole screen has to be covered in different colored pixels and so what you're saying is it's not very hard to figure out what to do paint a pixel yeah you just have to do a lot of it and it would be even better if you could do it simultaneously and cpus are not good for that yeah and so then these guys came along like one of which was nvidia and said how about we create a card and the card can't process very much like complicated stuff but it can do lots of simple parallel stuff simultaneously in the 90s they started this right yeah i think in the 80s maybe yeah maybe early i think nvidia was i think founded 93 and so i think um around that time and so they thought nvidia was 80s because i thought jensen went from denny straight to nvidia oh i thought it was you might be right you might be right but i thought it was 93 so basically they create this thing called 93 could be the first gpu i think i had other stuff before that right and so they basically say instead of a central processing unit let's just have a graphical processing unit and they do this thing called rasterization just bear with me on this which basically means painting i'm going to simplify it and say it's painting lots of pixels very quickly it's all about images okay and so they go and they build these cards and they take off for games and then at some point in the early 2000s jensen huang has this very smart idea and he says in fact around this time a little bit before that he says what if we did some processing on the cards as well and not just this drawing so what if we said how about the card actually helps you again i'm oversimplifying how about the card actually helps you figure out what you should do with the pixels and where you should put them and speeds it up and maybe we can do a few of them at the same time and and we can build this stuff in more effectively because remember this is not just code this rasters this rasterization it's hard coded into the chip it's some of the hardware and so he builds this thing called a tensor core and the tensor core does some processing on the card to help the rasterization then we fast forward to the last step where in the early 2000s i think this was 2003 he said you know what these tensors they're really good at doing lots of simple tasks very quickly and so we got some of them on the graphics cards with the rasterization but what if we created some software because i think that like research research organizations they might benefit from the idea of doing lots of little transactions very quickly and so he builds cuda this piece of software and so when people think about nvidia and they talk all about these cards really they should think about the software much more that's really what It's defensibility.

1:18:48That's right. Because you've got to use Kuda to program on Blackwell or LNVIDIA chips. Right, and so we'll get to Blackwell. It's like its own language. It's its own programming language to optimize these cards. So the rise of AI, I was a bit harsh on NVIDIA and I'm like, well, I just stumbled into the rise of AI. But it's unfair. Like Jensen Huang actually saw that lots of parallel processing of simple tasks would be really good for scientific research and created this code to do it. Then as we move forward, But what starts happening is then there's this rise of… So even more remarkable is during this time, NVIDIA has been having these 80 % drawdowns in stock price.

1:19:25The stock price gets high and then bang, drops 80%. Like three different occasions. That's right. So the hair-raisingness of this… Because their business is still predominantly selling graphics cards for video games, right? And so then at some point, they basically figure out… So then the AI revolution takes off. And it turns out the training of these large language models, the training of them because it's a big matrix with maths but the maths is simple yeah but there's a lot of maths so what do you need cards that can do a simple task but lots and lots of times simultaneously so that's perfect for these training these models and it eventually gets to a point where they start making cards where they say we don't even need this rasterization on cards because we're not drawing anything yeah we're just going to keep the um the tensor part of the cards with in in video's case the cuda part the programming and so these cards basically i'm going to simplify but like all the rasterization goes they call them graphics cards they can't draw anything and so when i go and buy a card let's say i'm going by 5080 an rtx 5080 for my video game system that's a blackwell card it's called that's the generation but it's not like well chip yeah like chip but that is not the same card as a b100 also a blackwell yeah architecture that goes into ai because the b100 has none of the ability to draw anything it is purely cuda the software cores and tensor cores the processing cores and so those are the chips that are on sitting on the cards that predominantly have been going into these data centers but let me say a few things about this these cards are great for training maybe you would say almost essential for training and they can do lots of other things they're fully programmable so it's not just about large language models they can do anything that good cards but they are massive overkill for inference for running the model these are expensive they run hot they i'm going to talk in a second about all the problems of building these cards but they are just massive overkill for running models but open ai they are using these cards to run the models and most of what open ai does so expensive yeah and hot and so hot and uses so much water it's all unnecessary yeah and um and so the problem is that all of these companies that run the software what they've been paying for is these very expensive high-end cards when most of what they're doing doesn't need them okay and so well let's say isn't deep Seat came with like a tenth of the running costs of GBD.

1:22:01They found a bit of a way to reverse engineer some of the training. Yeah. And running it does not need these cards. Remember that. The last thing I'll say about NVIDIA, the last two things. One is everybody that uses NVIDIA cards wishes they didn't use NVIDIA cards because they're so expensive and NVIDIA makes a 70 % margin. So all their customers hate them. Yeah. And the other thing I'll say about it. They hate them. They buy them hand over fist. Because they can't avoid it, right? They can't avoid it. But they don't want to. Okay, question for you, and this is what I've wondered for a long time, is why can't anyone – I know that NVIDIA's got Kouda, but why can't somebody – like TSMC manufacturers it.

1:22:35So NVIDIA don't manufacture it. Well, we'll get to TSMC next, right? That's the second to last step. So it's not as if they build it. They design it. Like Apple. Apple, they design it, and obviously Apple – Well, they're very good at designing cards. But surely you can recruit some really good people out of here and just start designing cards. Well, you might be able to, but why would they move? And like it will take your time to ramp up. So NVIDIA does have intellectual property inside their business. No one denies that. They've got CUDA, which is the language to grade. And that is the biggest obstacle.

1:23:02Like if you do not have CUDA, then whatever you do designing one of these cars. But why can't somebody build another CUDA as well? Well, because it's had 20 years. We're talking$5 trillion of value here. All right, so we'll talk about Google in a second. But the bottom line is this. Don't underestimate what 20 years of development, not just CUDA, it's a whole ecosystem of tools designed to optimize the usage of the card. But I would say that is NVIDIA's single biggest competitive moat, not the card. Let me ask you this. Would you call that a cornered resource? That I would say at the moment it's a cornered resource.

1:23:33It won't be permanently cornered. You argue there's a heap of process power there as well. I'm going to say TSMC has process power. Let's get to that in a second. But let me ask you this question. TSMC is the ultimate process power. Well, we'll see about that. But when NVIDIA says their cards are four nanometer cards, this is my question to you. a little trivia quiz. What does the four nanometer, it's obviously a measurement of length, what does it refer to? I thought they were less than four nanometers. Well, there's one now that says it's three nanometers. What does it refer to? What length is four nanometers or three nanometers?

1:24:04Isn't it like an atom per nanometer? So this is how you know we might be in a bubble. Nothing is four nanometers. Nothing is three nanometers. Nothing was seven nanometers. They started at 30 nanometers. Yeah, everything is some small number of tens of nanometers. and so a 5 nanometer card had 1.7 times the density of transistors on the chip as a 7 nanometer card. So if you did the maths, actually that should have been called a 4 nanometer card. These are just marketing terms for generations of chips. Yes, exactly like that. So this is very confusing. They're not, you know, on a 4 nanometer card, like on one of these B100s, 100s do you know how many um transistors which is the little switch that goes on and off and the less efficient the transistor is the hotter it gets because it's wasting energy and so do you know how many transistors are on a like a b100 chip what order of magnitude do you think a million two billion okay two billion can you i mean that is mind-blowing like i just want to say like basically i don't know like 10 000 years ago yeah we were like making stone tools and whacking each other on the head with bones although that was maybe a hundred thousand years ago that's obviously a 2001 a space odyssey reference yeah and now we're putting two billion transistors on a chip we'll get about how hard that is to do yeah but what's more amazing is they're not the only ones who can do it so let's go up a level okay so that's nvidia i think nvidia is a great business with real competitive moats i think their main moat is the software not the hardware yeah i think the hardware is at risk in my view guys did their episode that three series episode on nvidia like two or three years ago and they were talking about CUDA being their main computer.

1:25:44Well I agree with that. This is when NVIDIA was a$4 billion business. Yeah I agree but I want to say this in order for AI to grow at the rate that all these data centres are being built NVIDIA would have to do something between 3 and 10x the number of chips that they're producing like over the course of the next three years I don't see how it's possible. Like I think that's why this is also a bubble. Like even if there was demand which I don't think there will be the scaling up of this so we'll go up one level higher the second last level so did you want to talk about tpus now before you get to uh i'll talk about yes i'll talk about tpus so google so google does a lot of stuff with a big announcement last week on tpus has really rocketed google share price yeah but they say it's a big announcement but it's just another announcement they hint so let me say what tpus are so the t stands for tensor so at least someone has stopped calling it a graphical processing unit and starts calling it a tensor processing unit so google said like the nvidia cards they do lots of different things but we only need to do two things.

1:26:42We need to train a model and we need to run a model in France. So we don't need all this other stuff that they do. So what if we try to design our own cards? Now, most of what Google does with AI, people know Google with Gemini but that's not most of their AI. DeepMind was original. That's where AI started, right? Yeah, absolutely. But most of what Google does with its AI. It's an advertising business. Yeah, it's running its advertising business and working out what ads to serve and its own search algorithms. Most of Google's AI work is internal. And so what they have the advantage of being able to do is to say, what if we design some cards that really only did one thing?

1:27:20These are called application-specific integrated circuits, ASICs. And what if we only designed them to do very specific things in our own data center? That would be good. They wouldn't be as good as the NVIDIA chips. They wouldn't be as powerful. They measure power in flops, okay? They wouldn't be as powerful, but they would be a hell of a lot cheaper and we could get out of the NVIDIA ecosystem and be dependent on them. And so they started designing these TPUs. Amazon is designing their own version of them as well. I forgot what they're called for a second, but they're doing their own version of them.

1:27:50Like everybody wants to design their own cards and Google is rolling them out. They are in their own data centers, these TPUs. Because Google's full stack, right? They own almost pretty much other than TSMC. They own almost the entire stack. So I think Alphabet is the best company in the world today. We'll get to this at the very end. but yes they own their own cards now in their own data centers and they are trying to move away from NVIDIA but it's worth understanding they've got Google Cloud they've got Google Cloud they've got the chips that's right they've got the customer they've got the biggest search engine that's right yeah they've got everything they make the chips but just to be crystal clear about these TPUs like they do not have the power of a B100 or B200 from NVIDIA they're not designed to do that they perform like one function or something they perform that's it they're application specific right But they can be used for training.

1:28:38And you remember I said to you that running models on these very expensive NVIDIA cards is massive overkill? So Google is trying to overcome that problem to begin with. It's going to be much, much, much cheaper for them. It's already cheaper for them to run their AI than for other people that don't have their own cards and data centers. For example, OpenAI, which is just a software business. And then what that then happens, and this is what the all-in guys were saying this week, is that Google then gives away Gemini for free, most certainly, or gives it away if you're a subscriber or whatnot. It gives away a lot of people for free.

1:29:12And OpenAI has got no revenue because our only revenue comes from these idiots paying 20 bucks a month or 200 bucks a month. Idiots including me, right? And you're going to switch to Gemini because Gemini is A, better, and B is going to be cheaper. Well, at the moment, I've still found, for what I need, OpenAI has still got an edge. But it's going to... In the last update, Gemini is going to be considered better. I tested it like a week ago. But it's going to disappear, right? Basically, I think the models will be commoditized. That's where I think it's going. Yeah, well, they largely have been already.

1:29:36And so that is to understand Google, Amazon, meta, you need to understand people want to make their own chips and put them on their own cards, okay? And so that's what's happening. And they've been working on this for a while. They didn't start this like a week ago. They have. And the announcement they made, which is we're going to sell them to third parties that rocketed, and we're going to sell the TPS to third base, that's great, but it'll be hard to imagine how they're going to sit outside the Google Cloud. They're really designed for sitting inside the Google Cloud. Yeah, but they sell the Google Cloud to third parties.

1:30:04I think that's what they're going to sell, right? If you want to go and do this processing, you can sign up to our cloud and they'll have these TPUs and it'll be much cheaper. Yeah, absolutely. And A &B is probably doing the same thing. And hopefully use much less energy. Yeah. That's a big factor, okay? Absolutely. Don't underestimate the energy. Totally. Now we go up one level because the one level is we get into the... Because people think NVIDIA is the bottleneck. It is just because today everyone's using these NVIDIA cards. The real bottleneck you mentioned, who makes NVIDIA's cards? TSMC.

1:30:28Who makes the TPUs? Probably TSMC as well. And where do they get their stuff from? We'll get to that at the end. Who makes Tesla's cards? Oh, Nvidia makes... TSMC. Yeah, that's right. But they get from Nvidia, right? Basically, there are only a few foundries in the world that can make these, I'm going to say, four and three nanometer cards. But now I'm winking when I say that. But these very specific... So TSMC, Taiwan Semiconductor... Taiwan Semiconductor Manufacturing Company. In Taiwan, one of the reasons such a geopolitical problem right now. Yeah, exactly. This company, basically, this foundry, or a fab, whatever you want to call it.

1:31:07What's the founder of the SMC again? Morris Chang. Morris. Yeah. So he basically... Is he Steve Morris still? He's like 90-something, right? Yeah, he's still going. They make everything. Now, Samsung also is trying to ramp up, and there's one in China trying to ramp up. And like Intel... By the way, I think Intel... You could probably bet on Intel. I think they'll work it out. Isn't Japan got some stuff going on as well? Government funding some stuff? Yeah, but it's just the same companies. with foundries in different places. And so then you might say, well, how is it? How is it that they're not TSMC's cards and the machines which we'll get to, they're not from TSMC.

1:31:41So how could TSMC have a competitive moat? And it turns out, I'm going to tell you, the two real bottlenecks in this industry, which are not widely talked about. Oh, by the way, just to go back to data centers, if you want to bet on something with data centers, go find out who does all the cooling in data centers because those stocks might still be undervalued, right? And so, all right, so anyway, But back to TSMC. So what are the bottlenecks? Bottleneck one, there's a special kind of memory that needs to be used on these cards because it gets stacked up into a three-dimensional stack. When I say a stack, it's probably a millimeter high, okay?

1:32:11But it's a stack. And this is called high bandwidth memory, HBM. There are three companies in the world that make this. One is Samsung. One is a company called Micron. Now, Micron made all this memory until yesterday for lots of consumer applications. They turned it all off because HBM is so valuable to make. They just want to make HBM. And there's a third company. I can't remember who they are. So there is a real bottleneck in getting this HBM memory. You cannot make these advanced cards without this HBM memory. So that's a real bottleneck. They're trying to increase 2 to 3x supply over the next few years.

1:32:46It's bloody hard. And like the yield rate, do you know what the yield rate means? No. When you make a chip, a lot of them, it's a very complicated manufacturing and a lot of them don't work. Yeah. And so the yield is how many work. The yield rate for this industry at these high-end cards is really low. Lots of stuff goes to junk. So one of the advantages that some of these simpler cards have is that they don't need to have this super advanced manufacturing. They don't need to use the most up-to-date machines. We'll get to the machines. So that is why it is better to have less sophisticated cards if you can get away with it, right?

1:33:21And you don't have to get in the queue of TSMC. It's hard to get in the queue that demand massively outstrips supply. So that's one bottleneck. This is the other bottleneck. So we talk about chips, but actually you make the chip and then you have to put it on a wafer and then you put all this other stuff on the wafer like memory and then you have to put it on a substrate which is the final thing. That whole process is called packaging. Packaging is not I put it in wrapping and put it in a box. It's what do you do with the chip. Now, it's bloody hard to do this packaging for advanced chips. So I said a chip goes on a wafer, goes on a substrate.

1:33:59But if I had to do an acronym, I might call that COOS. One company in the world has mastered COOS. Who do you think it is? Not ISM. TSMC. That is why one of the reasons all of this is made as TSMC. It is genuine intellectual property. And even, we're going to talk about the machines. But these machines that they buy to make these chips, which are not theirs, these machines, they don't really work very reliably. and so TSMC has worked out how to maximize yield on operating the machines they obviously have preferential access to the memory and they're really the only party in the world that can nail this COOS which is the packaging you need to turn an NVIDIA or anyone's chip into an actual card and so that is the massive rate limiter COOS is slow like it's days might be longer and it has a low yield and everybody needs it TPUs do you think they need COOS?

1:34:52probably they need co-ops everybody do they need hbm they need hbm so yeah they've circumvented nvidia but they have not circumvented tsmc tsmc is a bigger competitive moat and i think if you have to say who could be replicated nvidia's um nvidia's cuda software is hard to replicate so google for the tpus they don't use cuda obviously they made their own versions but we cannot compare these versions to cuda they are just used to run much simpler applications they do not have the full functionality of something like a CUDA. So even Google cannot do that software, okay? They'll get there eventually, but they're not there today.

1:35:27So now let's go to the last piece of the puzzle. You know what the last piece is. Who makes these machines that TSMC uses? AMSL. ASML. Yeah. Based in? Dutch. Yeah, based in the Netherlands. And so I'm going to tell you what goes on. So these machines, they are the size of like, I don't know, like a double-decker bus. they weigh some 150 tons or something like you know they make like i think they make 60 a year a year and they go and they make them and then they have to install them for you in your foundry because you don't know how to install them and this is what's going on i just got it not with my ikea drawers well i got to tell you this story of what's going on because i'm going to blow your mind okay so the problem is when you're making chips with two billion transistors on them there's no tool small enough to etch in basically the wires are all printed on the boards right and so there's no tool small enough to be able to do that and so how the hell do you go and etch onto these cards the only thing small enough to do it is light yeah light like photons yeah and so it's like well let's use photons to engage in a kind of lithography that is like carving in based on what you shine on and what you don't shine on, photolithography.

1:36:46And I'm going to tell you how crazy this is. So in order to create the right wavelength of light, it has to be small enough to etch on these tiny things. But if the wavelength is too short, then it goes through everything. And the problem is you have to bounce this light with mirrors. If you try to run this light through a lens, like just a straight lens, it will just ruin the lens right and so it turns out the light has to be exactly exactly 13.5 nanometers in wavelength so that is a real measurement and so nothing emits 13.5 nanometers in wavelength except they figured out one way to emit this is what these machines are doing so this is what happens it turns out if you turn tin you know the element tin into plasma which is a fourth phase I would call it a fourth phase of like you got you know like solid liquid gas and then I'd say plasma is like a fourth phase a state of matter it's like a high high high energy gas and so if you can turn tin into plasma then the photons that it emits when it gets turned into plasma happen to be a wavelength of exactly 13.5 nanometers long so this is what goes on in these machines They fire a piece of tin.

1:38:08They fire a laser to flatten the piece of tin. Then they fire another laser to, I don't know, plasmerise the tin. Then this 13.5 nanometre wavelength flies off. Then they bounce it off very specialised mirrors. That's how I'll finish this story. Then it hits this substrate, and wherever it hits, it changes the coating on the substrate. And then what they can do is at the end of that process, they wash the substrate and either where it hit washes off where it hits doesn't wash off depending and then they can use that as like a mold kind of thing or a template and that is how they can have the shape to go and create two billion transistors on a chip it's totally insane what's going on here like and so one of the things that tsmc does is that light kind of goes everywhere and so tsmc has worked out how to maximize the yield using that light.

1:39:03And so one company in the world can make these machines. I think maybe for the history of time, one company in the world will be able to make these machines. It's a crazy technology, right? It's the highest barrier to entry you could ever imagine. And what's interesting is that mirror is made by one company in the world, Carl Zeiss. Oh, Carl Zeiss, yes. private company owned with the same ownership structure as novo nordisk yes it's basically a charitable foundation whose job is to spend all of their profits on innovation german yeah all of their profits are spent on innovation and so one of their biggest competitive advantages in this world is if you think about the whole world of ai going on today every time you type something into gemini or nvidia carl zeiss says thank you very much you've driven our mirrors business so No one else in the world can make these mirrors.

1:39:55And so my view on, you know, where does this AI revolution go? So I said to you. I was going to say just to our listeners, you've just given that sermon without a single note, without a single breath in about probably half an hour. It's extraordinary. I don't say half an hour. I blew it out the episode. TSMC style process happening here. Yeah. Very impressive. It's easy to. You didn't even look down. You just went bang. Yeah. Yeah. Well, I don't know. It all kind of feels logical to me. And so I said I'd touch the NeoCloud. So the NeoCloud is, I told you, furnished apartment. And so they go and correct.

1:40:28It's the CoreWeaves. The CoreWeave. We'll say Firmus because that's known to Australian listeners. And CoreWeave was worth$70 billion briefly. It's now back to$35 billion. And Firmus,$7 billion or something. Well, actually$6 billion, but yeah. Okay. And so what does it mean? It says, we've got this furnished apartment with AI. We need to find tenants. And so at first, NVIDIA comes along and they say, it's going to be pretty expensive to furnish that apartment with like our GPUs. So how about we give you the first$10 or$20 million for free as an investment? Well, once you've got NVIDIA investing in your business, well, they invest in all of these all around the world.

1:40:58It's so easy to raise money because you say NVIDIA is our cornerstone investor. Boom, you get the money. Then you say, well, now we build out the data center. Great. I say, well, that's great, except I think in order to generate revenue, so that we tick the expense side of the equation, how do we tick the revenue side? How are we going to get tenants? So this is what's not well known. For NVIDIA to keep developing their chips and software, they need lots of AI processing power but they don't need it all the time they need it in bursts so what would be a dream for them is if they could find some place where they could just get it in the burst they need and for example they could say how about we become your first customer and take 30 % of your capacity now you've got your first customer NVIDIA is an investor NVIDIA is cards NVIDIA is the cornerstone customer but this is a bit of a kicker you still need to fill the other 30 70 of the data center and you know aws is and and their ilk google they sign 10 year deals with like next dc nvidia signs four year deals and so nvidia is building out this massive supporting this massive ecosystem of capacity around the world my guess is not as educated guess there will not be the capacity to fill these data centers there's too Too much neocloud space available.

1:42:18And presumably things get more, like Moore's Law, things get more efficient. Shaper, faster, et cetera. Smaller, yeah. And so I think the exponential growth curve that we have seen, it's not going to continue. Well, the main reason for that is there's no real demand side. Well, who's paying? No one's paying commercial prices. So we're still in the Robin Hood economics moment of this industry. Uber of 2018. Money goes from the rich VCs to the poor customer. And so the thing I think, so who am I bearish on? I'm most bearish on the NeoClouds. More than OpenAI? More than OpenAI. I think, because I don't see any way that the, oh God, it's competitive.

1:42:56Like, yeah, maybe OpenAI I'm the most bearish on. We're thinking of getting pretty close to zero. Yeah, yeah. The NeoClouds I'm very bearish on. I just don't think that's going to be the demand. And I think NVIDIA, NVIDIA circulating their own money and demand through the NeoClouds, because it suits NVIDIA. It's very smart by NVIDIA. Who am I most bullish on? Like the top of the pyramid I just said. Like ASML, you can't buy it. I'm not sure if it's listed. I think it might be listed actually. You can buy TSMC, I'm very bullish on. NVIDIA a bit less so. But still, NVIDIA is a great business. I think it's just overvalued.

1:43:26It can't continue at this growth speed. The question of NVIDIA is will it go backwards when some of these people start falling over? I think so. I think it needs to go backwards. I think the revenue is not sustainable. The reason why it's multiple, given its growth, its growth is insane. it's like 65 % growth or whatever it is, or 100 % growth, it's a multiple of like 40 or something. That's not a 40. People are thinking it's going to drop. Yes. Otherwise, the multiple doesn't make sense. Yeah, I think it'll drop because demand will drop and also because there'll be alternatives. So the dream is to get these extremely expensive, high-precision cards to stop doing most of the heavy lifting and to fill it with these application-specific cards, ASICs, that are just for inference and dramatically reduce the cost and energy demands.

1:44:11And I think that that's where it's heading. And so if you think about these businesses, Google, local Alphabet, Alphabet, Amazon, Meta, they are all trying to achieve the same thing. Think about the stack. Model, they've also got a business that sits behind the model. So OpenAI has got one business, right? It's Gen AI. but they've got these businesses that sit behind that all this advertising monetization, right? Yeah. And so they've got the model. Sorry, they've got the software. They've got the model that sits behind it and now they've got their own data centers which they might be leasing other people's data centers or building their own.

1:44:53It doesn't matter. They control the data center and now they all have or are building their own cards. And so I think, think about that stack. They're never going to have their own foundries. They're never going to build their own machines. We didn't talk about like, you know, There's obviously also rare earths. So none of this happens if you can't get access to rare earths, which is why everyone's trying to get access to rare earths and the whole battle with China. But I think if you look at those three companies, they are the best positioned companies in the world to benefit from this next wave.

1:45:18There's no doubt in my mind that Alphabet is the single best positioned company on planet earth right now in my view. That's why it's share prices at record levels. No, I think so. I mean, I think I haven't looked at their share price, but I think it would have to be crazy for me not to think that Alphabet continues to be a buy. like and who which company which of the magnificent seven that's relevant because tesla makes chips right and they do some stuff their chips actually are like not these absolute cutting edge chips they get to benefit from the cheaper stuff yeah but which company is missing from this story yes so i think apple's got you know you always post about deep problems that apple's got i mean apple is nowhere in this game they do make their own chips for their machines right you can argue apple is what i'm very quick on apple for a lot of things but apple's incompetence here may have actually helped them.

1:46:04Everybody's going balls deep in this stuff and they're sort of sitting back and can, like what Microsoft's doing, license somebody else's or buy someone else's when they've lost a bunch of money. So that could actually happen. But Apple has demonstrated that they can design chips for their own applications that are vastly superior to off-the-shelf chips. You look at the processing power and battery life of Macs, that's because of their chips, right? And so why they didn't think about designing their own like TPU equivalents, it's hard to know. Maybe it's a secret. I doubt it. last thing um the biggest change with google so i don't know if you i'm sure you remember this at the beginning there was this like you searched on google and it would give you a bit of ai yeah and they said that was gemini and then there was but it was really a search assistant yeah and then that search assistant became a broader like life butler right and you could ask it for the weather or to book things or to do things and at the same time it was developing gemini so it's essentially merged these two things together and to where google this is why i'm so bullish so where they're trying to go is they're going to own the browser experience and then every single thing you do in your entire life is going to be mediated through some version of the browser experience on whatever device you're using yeah using gemini yeah to power everything in your entire life like i don't think it's good for humanity but i do think it's good for alphabet like i think that's they are just streets ahead of everyone amazing people and you got sergi back in there and you dennis in there and they've got as good a team as anyone and they've got the money to employ like you know a little bit of cash yeah in a previous episode basically everybody that started any of these businesses or works in them came from deep mind inside google i mean they can get them back over time right because good deep mind like they're all from there right yeah open ai was yeah it's hard for me to find you know definitely the search monopoly will diminish yeah but i think that this i mean i just think god they're well placed they are so smart i think so if you look at who's well placed i agree with you on google and look you look at open ai who are getting absolutely destroyed by google consumer level and they're getting destroyed by google and by anthropic at the enterprise levels have been absolutely completely squeezed at both ends and not hasn't quite shown in the consumer and obviously this is why sam's doing the code red but why would anybody using open ai in two years when you use a better product and a cheaper product in gemini the code red is you know what they say like the first mover gets arrows in the back yeah And so he actually wasn't the first member.

1:48:28He was the first to popularise the technology to consumers. But I think the code ready is, oh, my God, the moment of commoditisation has come so much faster than we expected. I think that's the challenge. Oh, I forgot to tell you this thing about… Well, the only hope they have is if… So look at Atlassian. Yeah. Average product, great switching costs. So it survives and will survive. Oh, and they bought the browser company. And I finally figured out why. It's because they think that, like, every major SaaS platform is going to need its own browser environment to do its activities. Like Salesforce will have their own one.

1:49:00That's what they're trying to do. So they paid a lot for it. I wouldn't have bought it, but I get what they're trying to do now with it. Anyway, keep going. I'm not sure there's much competitive advantage in that. I think owning the workspace, maybe I'd call it rather than a browser, is ultimately going to be a big battle over the course of the next three years. The question is, can they win, right? Can they win, yeah. so OpenR is getting squeezed from both ends which is really problematic they've got they're losing money 20 billion expenses 13 billion revenue and then in the code red Sam's talked about forgetting forget about revenue even the potential their only real hope was they can get to advertising profitability because 800 million customers but I'm dubious that they will it's hard to see how they're going to it kind of goes against the whole principle of LLM but also I keep saying to you they basically sit on Microsoft Microsoft has effectively taken an option over open ai yeah that when they um run out of cash because you know they've got like 1.2 trillion dollars of commitments or something crazy like that that's ridiculous but half that's just options and it's not formal but i think microsoft is just there and says you don't have to go broke we'll just buy the rest of you when you're already fully integrated into our stack i forgot to tell you this last thing about you'll love this yeah about the neo clouds yeah so the other reason nvidia loves neo clouds is i said to you like you need to try and get supply from TSMC.

1:50:19Yeah. And if you can promise them forward supply, they make sure that you stay high up the order of, and so all these NeoClouds are giving NVIDIA all of this demand for their cards, which means they can go to TSMC and say, make sure you keep us at the top and give us preferential access to the wafers and to the Coors technology. And so it serves NVIDIA's interests so well, these NeoClouds. Yeah. NVIDIA can't lose out of them. Yeah. NVIDIA will only win out of them, but I'm not sure about investors in the NeoClouds. Yeah. And so we talk about OpenAI being probably one of the most obvious shorts, but it's a private business.

1:50:54Yeah, you can't short it. Well, maybe you can. Yeah, go on. Here's my idea on how to short it. And I'm not the first one to pick this up, but which is a business that recently sold their shares in NVIDIA to increase their stake in OpenAI? SoftBank. SoftBank, public company in Japan. Oh, that's interesting. Short SoftBank. How material is it as there? It is, I think, material enough to make an impact. Right. Like it's not a one-to-one, but I think I think it's stake in OpenAI is now like the SoftBank stake in OpenAI is about 11 % so it caught 55 billion I think it's market cap well it reported a 12.8 billion profit in September because it wrote up as OpenAI based on this it's all WeWork the WeWork story playing well you did persuade me a bit I think that Sam Altman has some characteristics of who's the WeWork dude Adam Neumann yeah he does same person no because there's all this other stuff with Adam Neumann I don't want to allege on Sam Altman.

1:51:46The dog smoking. Yeah, and all of that. But the razzle-dazzle P.T. Barnum. But by the way, Jensen Huang at NVIDIA, God, he talks his own book. Like, he is not a modest guy about it. Like, he is spruiking AI like crazy. Yeah, but he sort of walks the walk as well, as talks the talk, whereas Sam's not much walking. Yeah, he's an engineer. Sam's a lot of talking, not much walking. So yeah, SoftBank has about a$55 billion stake in OpenAI, and it's worth 153 billion. It's about a third of its market value. So if OpenAI goes to zero, that's a chunk of SoftBank and it's also credibility. It won't go to zero, but it will go to a fraction.

1:52:23I think my prediction would be it will go to a fraction of its peak valuation. 20 billion or whatever. Yeah, and Microsoft will just gobble it up and that will be the end of it and it will be a great deal for Microsoft. Well, you say that. If it's losing$7 billion a year and continue, like things run out of money, like if it just runs out of money, it's pure hype now. I know, but there'll be something. It'll be like effectively an acqui-hite to give the people some... But how much are you going to pay for an Aqua hire is the question. Well, I don't know. You don't want Aqua hire. For whatever reason...

1:52:49Most of the good people left. No, I agree with you. When I said, I don't know why, I said Google, I said Google, Meta and Amazon, but I should have said Microsoft. Like, my order would be Google, Microsoft, Amazon, Meta as well positioned to take advantage of this. I agree with that. 100 % agree with that. And so, yeah, Microsoft, I mean, I think in a way Microsoft is the most amazing company. I mean, when was it founded? Microsoft was founded in 1975. So, this year... It's their 50th anniversary. What other 50-year-old company is at the absolute global cutting edge of the latest technology? We talk about - They're the most amazing business.

1:53:23They're about multiple acts and it's had three genuine, the initial operating system act, the court, then it had the Microsoft Word, Excel act and now it's had the - Well, what about the - Yeah, well you say internet cloud. Yeah, but the thing is - It kind of missed the internet initially. It missed the whole landscape thing. It's first business, the original business. It missed mobile, yeah. Like, it's still making money for them. Like, it's crazy. 50 years later, he's milking the hell out of the business that first established. Well, the genius of Gates at the time, Gates is probably, if not the best business person ever, one of the best person.

1:53:55Yeah. The genius in him effectively forcing, I'll give you a massive discount, you just put my Microsoft Windows, my DOS thing on your computer, and we'll give it to you for like super cheap, but just preload it. And then everybody had to have it. It was just the, he created this monopoly, killed IBM because IBM was the leader we talked about a couple of weeks ago and they've had act after and they missed the mobile they missed mobile is one thing they missed and they missed that initial internet they came back beautifully with Azure and now obviously where they're at with effectively potentially buying OpenAI so they've somehow had three massive acts and what people don't realise is that Bing which no one really cares about but it's like most of the indexing used by most of these platforms like they're using Bing Google's not giving them their indexing.

1:54:42And so it's super interesting. They're an amazing business. I think Sergey Brin might go down in history as well as one of the greatest businessmen. But yeah, Bill Gates. Google's had two to three acts as well. Well, I think Bill Gates was prepared to ask this question and Google also asks this question. It's not an easy question to ask and I think Zuck asks it and maybe you could say Bezos as well. It's hard to know with him but I'd give it to him as well which is what do we think the reality is going to look like in 10 years? let's think really hard and what would we need to be in order to be the most successful player in that reality in 10 years and let's just start yeah and let's just start doing that today even though people won't know what it is and they'll think we're a bit crazy and wasting money and I think that that I mean there are very few companies that can successfully do that Jobs said this is what people are going to want I'm going to make but Jobs is largely a product guy yeah absolutely a product guy whereas I think those other guys they are strategy guys right they are real deep tech strategy I feel like I've talked for the last four hours it was an excellent thing our listeners will get a lot out of that thank you Joel and Mike for another great episode thank you I do we will see you everybody on Saturday for our Ask Us Anything episode fantastic