This Week in European Tech: Europe’s dependency problem runs from rare earths to AI

28:48So you have a massive now price differential, and that tells us something about the pricing power of Anthropic and of the Frontier Labs. We go back to June. we were talking about kind of the signal that comes out of what Anthropic did at the time, which was put Fable in at a very, very high price. And as we said, look, it's really an experiment they're running to test the pricing power of frontier intelligence. And it seems that they have run this experiment, and it has largely not succeeded. They are being pressured by OpenAI, which has released formidable models, and of course the whole open source panoply and plethora of models.

29:27Well, everyone will have seen, they'll have seen the reports from Hugging Face, from Vercel, all of these. And I know that they have a particular market segment that they're highlighting. But I mean, as in they're the hobbyists, they're the independents, they're not necessarily all of the enterprise. But some of these reports, actually, there are a chunk of big corpse enterprise in the figures where you can see open source really, really taking hold. So I wonder on the open source front, if there is more of a story on that side with more people becoming aware of what is available for on-prem or for full control or other, is that fair?

30:03Yeah. I mean, I think if you're in technology and you were not aware of open source six months ago, you were sleeping under a rock, right? If you're at the coalface and you're doing development, but I think you're probably right. It is probably more at the enterprise where things are moving because, you know, people would have initially been more comfortable purchasing from a kind of US vendor that would show up and have everything nice and shrink wrap for you. And open source might feel a little bit more experimental. And then because the cost, because what people were spending on the models just exploded the way it did.

30:37Right. You know, Anthropic going from 10 billion run rate to 65 billion run rate within seven months. The money is coming from somewhere. companies are spending, people are paying and looking at their bills and going, oh, gosh, we better see if we can get this a bit under control. And that's been our reaction, right, to our bills, which they're lower now than they were last month and the month before, because we are, like many others, we are migrating out using different models for different things, etc. So yeah, I think that's the way the industry has reacted. I read that the cost of GPT 3.5 level performance has dropped 280 fold in two years to get performance.

31:17So it just seems very early to make reliable predictions about how the pricing is going to unfold. but i think what is obvious is that as enterprises embed this stuff into their their systems um usage will go up and then you're locked in and there'll be a huge cost for unraveling that integration and when they mitigate some of these when they when they mitigate that by having this middle layer where effectively you're a you're able to constantly flick and change the models that you're using i've always i'm speaking slightly at the top of my head but I've always seen it that way. Yeah, but historically it's not been that simple.

31:57I always think this sort of plug-in, plug-out analogy, actually in practical terms is a bit of a myth because you've got everything from different legal contract considerations to nuances around the models and how the models respond, but different API integrations. In reality, I don't think it's ever going to be, especially for an enterprise, flipping one in to replace another. So my guess is that as enterprises get locked in, even if the price goes down two things will happen one is that although the cost per token or the cost per use will drop because it's integrated into your company the usage will go up so you'll still end up spending more but secondly once you're logged in what do people do they raise the prices well so they may claw back later going with your i mean i think we're on a race to zero personally But going on your Jevons piece, Mads, have you got any notes on Jev?

32:51I do. I'll come back to it in a second. But just picking up on Andrew's point there first, I think the evidence of what you're saying is, and I think we have a chart we can bring up for those who are viewing. What's happened over the summer is that if you look at the Pareto frontier, so what does that mean? Well, you know, all model labs, they've got models of different capability. You know, we know Anthropic, they have the fable, the opus, the sonnet, the haiku, and open AI, they've got the Sol, the Terra, the Luna, the Astra, the top-of-the-line model. And you can run these in different levels of reasoning and different levels of effort to get different levels of capability.

33:36and if you map all of that out, you can map on a sort of classical XY diagram, how much intelligence do you get per dollar you're spending or how much output you get from these models. And you can draw a frontier where you can say, at any given amount of spend, how much intelligence do I get and what is the model that is on that frontier? So what is the best model I can get depending on my budget, depending on how important that task is and how much I want to invest in it. And Anthropic used to lead a lot of that. But actually, over the summer, OpenAI has taken almost all of the Pareto frontier all the way up to the most powerful Anthropic models.

34:19The most powerful Anthropic models are still ahead, but the long tail, the cheaper models are far below now. What does that mean? That means with Anthropic, unless you go for frontier intelligence, you're not getting all that much bang for your buck. and how can they do that? Well, Andrew, I think that's back to your point, is because a lot of this stuff has now been embedded inside workflows where it's a little bit sticky. And so there's an incentive on Anthropic to not give away too much for too little, right? They have some pricing power. They can do some of those cheaper workloads that are kind of run rate, that are sitting embedded in different places.

34:57They can just sort of keep that being less performance so people have to spend more money to get the same amount of output. What's the example? Haiku, which is the cheapest Anthropic model, is so far behind comparable models now for the same price point. And it's not because Anthropic can't make a better model. I promise you they can. I promise you they have one. They're just holding it back. So I think, Andrew, that's exactly to your point, there's pricing power. I think then to your point, the question is how long does it take and how sticky is it? At the frontier, we found out people are moving and they're shopping a lot because it's very expensive.

35:33But for those tasks further down the stack, the most simple tasks, how long will it take for people to switch from one to another? We're seeing a lot of people switch in-house. I'm hearing anecdotally a lot of large orgs now bringing in-house and not from the big frontier. Exactly. But even us, we're a small org. I think we've got 50 different paths. And we have systems set up to route things on different paths and to measure different things. But if you imagine a large enterprise, it's not just, oh, I'm going to finish my models today. There are thousands of small decisions you have to make. So that's not going to be something you do every 24 hours, but it's probably not going to take you two years either.

36:15I think it's going to be built in the infrastructure. I think it has to be built into the infrastructure. There's no way you're going to work out how this model war is going to end. So you have to have that flexibility. Also, even if you use open models for the menial task, those open models still need integration. They still need uptime support, service level agreements. And as the complexity around safety and regs and legal sort of intensifies and becomes more complex, it becomes even harder than to switch a new model in and out. But doesn't that work both ways, Andrew? It doesn't almost like the Jevons version of that on the engineering side mean that you're going to have the ability and the tech toolkits to be able to do that relatively quickly.

36:56I just don't buy. I always feel that, you know, the dream is never delivered on that front. Mads, any notes on Jev or any notes on the IPOs? Anything else that you'd want to touch on? So, Jev is this exciting new thing that was launched this week. So, there's a group called TypeSafe AI, where the founder, Diogo Almeida, he was one of the original co-authors of the OpenAI paper that was the foundation of ChatGPT. And he then left OpenAI, he set up his own company, and they've launched this week a model called Jev, which is a tribute to Jevons from Jevons Paradox, which we hope you all know. Why is it interesting?

37:33Because it's not a classical LLM. It's a completely new model family or model type that specifically is targeting classification. So one of the issues we have with models is that there's LLMs is that they're stochastic, which means they're not deterministic, which means that they're not giving us a predictable answer necessarily every time. And that has meant that if we want to be quite certain about the answer we're getting, we'll sometimes throw much more intelligence at a problem that it warrants because we want to be quite sure it gives us the right answer. So what could that be? It could be a simple thing.

38:10I'm looking at kind of a customer lead coming in. Is this a good lead or a bad lead? Is this something I should pay attention to or not? And that's not necessarily a super complex question, but I want to make sure that my answer is not too random. I want to make sure it's quite deterministic. Now, what JEV is, it's a way to, instead of having a model like an LLM that produces text and sort of tries to reason its way, via text to, is this a customer lead I should pay attention to? Well, yes, I think this is a good customer lead. You should consider it. You should pursue it and sort of construct that sort of stochastically or, you know, use probabilistically sort of a, you know, what it does is it actually says, look, here are some specific parameters.

38:54Yes, it's a good customer lead, yes or no. And here are the probabilities of each. And it'll assign probabilities. And it'll do that extremely fast, extremely accurately, and at an extremely low cost. So a 50th of Sonnet in terms of cost. And so that just means you can run classification and things much, much cheaper. And that opens up a whole host of new things we can do. Interesting. Have you got any notes on the IPO? I'm going to lay out a prediction. I reckon Anthropic end of October. They've just pushed it to November, right? I think that just came out yesterday. Oh, well, that blows that prediction out of the water.

39:35Do you know what they've said? Do you know what the latest is? The latest I've seen is that they've pushed it by a month. Okay. But who knows whether there will be more pushes. It's obvious that if you go back three, four months, the whole trajectory was this march towards the IPO that we all knew was going to come because everything was just up and to the right. and the market is more complex now. So, but what do you do then? Do you go earlier? Do you get the roadshow going now or do you push it to Q1? What do you do? I think you have to get it done. Who knows where the world will be Q1? I think so.

40:11I think so. And especially if you've got Trump going, hoax, hoax, hoax, we're going to push, we're going to go. At least you've got that as a bit of a tailwind. I don't know. I mean, are your numbers ever going to look much better than they are now? the other thing i wanted to talk about was this the notion of this ai money window narrowing we've got the data center ipos which are stalling and this is a bit of a marker i would argue sb energy has been delayed whole tech has been paused nscale is probably going out they've got an october listing haven't they so there'll be the there'll be the litmus test we're watching nvidia trading at cheapest multiples in a decade how are they doing that um anthropic as we've just discussed reported slipping to November.

40:54So is the wind coming out of sails? What's the deal? Mads, can you kick this one off? What's the deal with the money window, the AI money, the cash flowing? Well, I think we've seen SB Energy being delayed. But SB Energy, as far as I know, it's a company with no data centers running and they have one main tenant that's been signed up. And so I think if you don't have any revenue and it's just kind of hope and expectation and maybe a signed contract but nothing in the ground yet, then it's probably reasonable that you're not going to have too easy a time IPO-ing. Whereas Nscale has revenue. They have got lots of contracts.

41:33We know Microsoft and Anthropic are customers. And they look like that listing in October. Yeah. Yeah. So, yeah, it doesn't seem too unusual or unreasonable here that investors are asking for something tangible before they underwrite. Okay. So this is the anti-bubble trade then, isn't it? No, we haven't got enough. There's not enough bluff and guff to push. We're going to be more realistic, which is fine. The SP Energy backlog, I think, rests on one lease from OpenAI. Yeah. And OpenAI has itself said that they're going to delay their IPO. Okay. NVIDIA trading, what's happening there? Well, so when we say it's the lowest multiples, it's forward multiple.

42:17And forward multiple means that we are going to forecast what the company is going to earn next year. And then we're going to look at that kind of the company valuation as a multiple of that today. and because the growth is so phenomenal, forecasting another 70 % next year, that means that either the stock has to trade a lot higher than it does right now or the multiple will look low. And I think investors are looking at, gosh, there's another 70 % coming next year and then what's coming after that and what's coming after that and saying hold your horses a little bit. So I wouldn't say that the company is cheap necessarily.

43:01but certainly if you believe that they can keep up 70 % forever, then it is cheap. But then again, nobody's done that in history. Andrew, what does this all mean for European IPOs? You got any thoughts on that? Not really. I think you've still got a cultural blocker with the money markets. If I was a founder wanting to IPO, I'd want to go where the most enthusiasm is. Well, we're still looking to at least beat last year's record. So I'm just wondering, obviously, when wind is out of sails, it's a bit of a global phenomenon. But I wonder if there's anything specific. I imagine AI tech will be a specific marker that investors will be keeping in mind.

43:41I'm cautious to be optimistic in that there seems to be certainly in the London markets an acceptance of many of the challenges that have been talked about for a long time. And some of those changes have already been rolled out. So if we can get to this inflection point where, first of all, technology companies become a more attractive stock and more fashionable, and if we can get over that cultural hump, because I think it is a cultural hump, then that could have a big impact, along with regulatory changes that make the London markets more attractive. Don't you think it's going to take some time, Andrew?

44:13Oh, for sure. I mean, look, I was trying to spin something positive on something I'm not hugely positive about. I was looking fast, trying to be optimistic on a Friday morning. I mean, I just, I'm trying to think of the world where an N scale would be better placed IPO-ing in Europe. I can't quite, I can't quite see it. Let's talk about staying with AI. We have AI agents loose in the hoose. There are, as I've mentioned, rumors of agents all over the web in stealth. I'm not sure I buy into that, but we don't really know. We talked earlier about this open AI agent getting to the Australian government.

44:47um we also saw i love this story about amazon blocking metas muse so there's some agency e-com i think is going to be a big a big piece here of what e big e-com allows agentically to interface with their product set i don't know there's there's going to be if andrew let's you i know you feel quite strongly about this one how do you think this is going to play out or any thoughts on or insights on this kind of agentic e-com piece? I just think it's so early. I have no nuggets of wisdom beyond everything's a bit of a mess and in flux. I just think we're going to have agents running around, running amok all over the place.

45:34But will they be allowed to? I don't think, I mean, if you've got an agent that can avoid the nonsense that an e-com, like an Amazon site, tries to play with their pricing and packaging and all the games that they play to guide customers to a specific price for a specific thing. I just think they're going to be blocked. I just wonder, maybe it's not such a big deal. I mean, you're already saying that... You want plane tickets. Exactly. You're already saying that some sites are blocking some of these bots and some of these browsers or the chat GPT automated browser in order to prevent that traffic skewing analytics or preventing customers go direct.

46:19We all pay different ticket prices for airlines depending upon our web usage and profile, which is slightly frustrating. But ultimately, it's whether the prevention can keep up with the bots looking more like a human. It's the same cycle. Well, I guess this is where you'll end up. You'll end up with your agent behaving like a human so they don't get caught. Not just like a human, but like you, personally. Yeah, for sure. Mads, bringing it up a level, I mean, obviously, OpenAI getting into the Australian government. I don't know if you've got any more thoughts on this. You've kind of covered it already.

46:55But anything more broadly on the risks and the liability, perhaps? Lots of thoughts on it. I thought this agent showdown here between Muse and Amazon is really interesting. I think Meta, they again released a good product. Zaki's had a couple of good months. And Amazon, I think, are right to block it, not in the sense that I want them to block it, but I think it is probably short-term in their interest to keep these types of agents away from their site because it messes with their model. I think what will happen is that some vendors, some providers will say, look, I've got a model where I have a bit of a walled garden, and they're going to want to try and keep that hold off.

47:36And others will care less. Like, if I'm running my small shop on Shopify, maybe I'm happy for people to come and shop using an agent. And so I think things are going to bifurcate. And I think this is, at least based on what we've seen today, this is what the industry needs to sort out itself, right? I mean, as a customer, I want my agents to be able to go where I want them to go. But as a business, I also want to be able to regulate how people behave in my store. If I'm running a shop on the high street and somebody's behaving in a way I don't like, I want to be able to show them the door. If somebody's coming into my digital store and behaving in a way I don't want, it's my decision, what I want to have happen in my store.

48:13And then it's also my consequences. And if that means people don't want to shop on Amazon, I'll pay the price. So I think the business has to be left to sort this stuff out. I think the bigger question, I should say, is around these breakouts and what happens when there is an agent that breaks into somebody's system and it's not been allowed. And, you know, we talked about it in the earlier show, early part of the show. We've seen that 22 countries have now asked for a sort of an IAEA style AI body. And Finland and Norway and Germany have also, look, we support having some global regulator of AI.

48:56But we also know that the US has got no interest in this and China's got no interest in this. and so that means that neither is the uk by the way interesting neither is the uk supported it either but we're playing the political game aren't we i do yes that um but i but i think the truth is all the people that don't matter in this they want it and the two that really do matter they don't want it so it's not going to happen it's it's fiction well the winners set the rules don't they i also just don't know how it would operate in reality i mean the standards There's a difference between obviously adhering to standards, having a standards body, and then the difference between a safety body.

49:32And I just don't know. Again, I don't know how you monitor and implement it. It might be useful to have a talking shop which thinks about these problems, but you can't look at satellite imagery to see how many GPUs somebody has or what a model is doing. So the analogy doesn't apply very well. It has to be liability, Andrew, surely. It has to be the liability piece. It has to be how you can manage that. Is that fair, Mads? Yes, we could look at, you know, what are some of the people that have experienced the hacks of the breaches? What are they saying? Well, we know that one of the biggest that have happened is this hugging face breach.

50:09And their CEO, Clement DeLang, he came out and he said, look, I don't, I want us to accelerate, right? I just want to have access to the capabilities. The problem they had was that at the time they were suffering the breach and they tried to use Fable to defend, Fable's own classifier, Fable's own guardrails stopped them from using it and they had to resort to using the open source models. He's not happy about that and I can totally see where he's coming from. So he's saying, look, accelerate model development, let's get the best stuff and give us access to use it to defend ourselves. That's important.

50:44Yeah, fair enough. I think what's interesting about the Australian hack though is that that incident alone doesn't prove extraordinary hacking ability. I mean, my understanding is that it was an ordinary security weakness, exploited automatically and repeatedly and on a legacy system, public system. And so I think we're in a phase now where just the sheer capacity of an automation of AI is going to reveal a bunch of security flaws, which if they were looked at closely anyway, would be discovered and are well known. So I think the evidence here supports concern about autonomous persistence, words I can't get out without my coffee, but it doesn't yet establish sort of breakthrough hacking ability, although I think that will inevitably come.

51:34So this is a wake up call for governments and for enterprises and you and I personally, that if you're using the same password for everything, or you don't have two-factor authentication, or if you have legacy public systems attached to your back office, which have not been updated and the security patches have not been put in, or you have not done a security audit, you really do need to think about that now because they are going to be found and exploited. And that's today's problem. I think the cleverness of hacks or the technical ability or thinking a little bit ahead to quantum utility, where we will be able to crack RSA is definitely coming down the line.

52:14But today it's a bigger problem of just existing security flaws, which in many cases may be quite simple, being revealed and exploited. I agree. I think I can hear the phone ringing in IT support service call centres. It's a good time to be in cybersecurity, right? All over Australia. Invest in those cybersecurity stocks. Yeah, yeah, yeah. I think there's definitely even more cat and mouse in this space. Do you want to circle back around liability? Because you asked the question. I do. I do. My take is that if you just if there is if there is a strong liability anchor here, I think there is that that surely that is that is an endpoint as in it's your model.

52:56It's your liability, I think is going to be the the creator of the of the of the right outcomes. And, of course, there are lots of complexities underneath this. But we have legislation now, and we have a framework, at least in the EU, around it. There was a plan for specific AI liability, but that's been dropped now. And so instead what the EU is doing is it's saying we're going to use product liability and software counts as a product. So anything shipped after the 9th of December that has an AI model will be seen as a product and will be subject to product liability. And the view of the liability is that it's strict in the sense that if you can prove that there's a defect, you don't have to prove that the maker was careless, just that there is a defect and then there's liability.

53:49But hang on, just to clarify, is this, if I'm building a wrapper, whatever, using an open AI model, is it because I'm being paid, because I've got the contract with the client, is it my liability? That's my understanding. But this is some of the complexity that will have to be worked out. And some of it will probably be down to case law. Importantly, only individuals can claim, not businesses. Oh. And it covers injury, damaged property, and lost personal data. So it's not in the business realm yet. It's not in the business realm, and it does not cover financial loss. So if an agent somehow finds a way to put your coat on fire, there is damage and there's liability.

54:43But if the agent makes a bad financial decision with your money and invests it wrongly, there is no liability. So this is interesting, and I'm sure this space will evolve a lot from now on, but we are starting to see a new legal landscape here. What's this mean then, Mads? Because traditionally, software has been provided as is. In other words, if there's a bug that crashes your computer or loses your data, it's like, we are liable. You signed our end-user license agreement. It's not our problem. So how is this going to pan out? From what I can see from the 9th of December, all of that is going to change if you operate in the EU.

55:25Well, what I'm hearing is really between the lines is lawyers are not losing their jobs anytime soon. So much as you look at, you know, co-counsel or Harvey AI, I just see that, you know, it's going to be a great time to be a lawyer in the next 20 years because all this stuff is sorting out, right? Case law is only produced by humans, right? But really? Isn't there an AI for that? You really want to go up an AI magistrate? I mean, I think there's a good Tom Cruise film about that. AI causes the problem. AI solves the problem. AI legislates the problem. Well, let's get back into D-Love, though, with Euk.

56:03Mads, who are you highlighting this week? Euklid, which is a business in the Netherlands and in Eindhoven. They've just raised a 200 million euro. series a this is the gas it's not it's not it's not your grandfather's series says these it's like decent chunk of change and it was co-led by samsung by eqt scale up funds somerset capital innovation industries and a number of other greats and goods and the chair is peter venink who asml aficionados will recognize as his the former ceo of the business so from chip manufacturing equipment maker to chip maker, which is a nice segue. What are they building?

56:49They're building chips for inference. So it's all about running AI models, not training them. In Europe. Here in Europe, that's exactly the idea. They're claiming they have 100 times better power efficiency and cost per token. We haven't seen any independent tests yet. The first systems will not come until 2028. So from now on, it's all hope and dreams and a lot of capital. but some very, very credible people behind it. And yeah, we wish them lots of success. Yeah, really. And a 200 million Series A, that just makes me laugh. Love it. Good for you, Euclid. I am game on happy. Andrew, do you have a deal of the week?

57:28Yeah, I'm going to stay in the Netherlands, actually. So you're going to like this one, Dan, because this is a positive story about Europe. Is it a foldable iPhone? ABP, which is the Dutch Pension Fund, have committed a billion euros over the next three years to venture and to tech. No way. They're putting quarter of a billion into the European scale-up fund, which is the five billion fund for scale-ups across Europe and money beyond. So this is an incredibly good signal for other pension funds across Europe. Well, Europe has always been okay at it, haven't they? Well, not really. No, I mean, European pension funds put in a hundred times less than US pension funds.

58:06For sure, but compared to the UK. I mean, they're there at a 4 % to 6 % bracket, I think. Well, let me get my stats straight. But they're okay compared to the UK, I guess was my point. They're good. Loving it. Bring it on. Bring on more pension capital. I think, yeah, I think so. The dots connect a little bit here because what we're talking about is a pension fund putting money into the scale-up fund. Now, that scale-up fund is the fund that's just underwritten Euclid. and so what's happening is exactly what we've always said which is that for europe to take some of these big swings you've got to have big pools of capital that can underwrite them and there's incredible technical talent i mean nobody's going to tell me that the people that build asml don't know how to make semiconductors okay and so so we we now have capital to enable some of these people to go after what is going to be you know a huge and hugely important market we finally have the money to do it i just think it's a great thing to see that the dots connect in this way.

59:03Roughly 0.01 % of European pension assets went into European BC in 2024. That's one euro out of 10 ,000 euros. That's Atomico's report, 2025. That was for 24, though. Yeah, I don't think it's... I think it's still a small number, Dan. We have much better to go, but there's directionally... Sorry, man. Sorry. I swear I've seen different figures where we're starting to see... It was in your dreams, Dan. I might well have woken up and had a particular dream of pension funds investing in venture. Lies. Dan lies in statistics. Mine was Standard X. So I saw that Standard X have raised a 10 million seed.

59:46Now they're a London-based startup. I think they're doing isotopes for cancer to start with. Alpha therapy. Not that I know what alpha therapy is, but it sounds very smart. uh v squared east x first minute uh uk i2s breven howard macro venture and geometry all investing love it london seed bada bing bada boom more of the above please can i just say how exciting is some of the stuff happening in biotech right now but the stuff i'm reading is just i feel like we're driving towards an inflection point in in treatments for a whole raft of different diseases and i just that's very exciting again i'm a lay person so we should get an expert on the pod about we will we will we're going to start getting some kind of some guru ninja experts to come and share their wisdom on these on these topics in the week but i'm totally with you i think it's so it's a super exciting time to be scary for some but super exciting time to be in tech that's for sure mads what's happening in the week ahead anything that we should keep our eye out for so we've got our great glorious and illustrious leaders in liverpool from sunday to wednesday is the i can hear your tone of voice well they're going to tell us that they're going to solve all the nation's problems with more growth i wonder why nobody's god man you're sounding like me man you sound so yeah you have you have flipped you have flipped flip roles that's for sure uh no hope and dreams baby hope and dreams Look, they're going to be there.

1:01:14They're going to send their messages. I think we're all worried that there's another tax hike coming. But let's see when budget time comes. Where did BADR end up? Was it 18 %? I don't know. But it's in development tax. It's capped at a million. I mean, it's not here nor there. Is it? Is it? Oh, okay. That's okay. That's okay. Ignore me. Carry on. No, no, no. No, no. No, the relief is capped at a million. Okay. Okay. Okay. No, that's not it. But that's been taken to the point where it's not a relevant thing. Good. I'll take it. Which I think is why they call it the bad R. Boom! So going back to next week, we've got one IPO coming up.

1:01:59Aura prices. It's Nasdaq IPO. I think, Dan, you're a fan, aren't you? Smart rings? I was until I went back to my Apple Watch. because you're paying, I resented paying a subscription fee for a 500 quid ring that I bought. And I went, no, do you know what? I'll just go back to using my Apple Watch. And it's just as accurate. Well, I've tested them all. I've tested the Whoop, the Aura, the Apple Watch. And they're all pretty much, as long as you get the flavor of what's happening, you don't need them to be, you know, who will know how accurate they are. So I was a bit, I'm a bit irritated. There's a court case, isn't there?

1:02:36at the moment. Well, there was. There's a podcast against Aura for just basically publishing bullshit. Yeah. Well, I don't think that was the legal terminology, Andrew, but there were questions. I'm no litigator. I'm no litigator. We'll find out in about 10 years' time when that lawsuit's done. When we all die of heart failure because your Aura ring didn't pick up any of your markers, yeah. Anything else, Mads, before we close off? Anything else happening the week ahead? So it seems that you're not a big Aura fan, but they are targeting a$15 billion valuation, Oh, I've got one. I've got one. I mean, I wear it, you know, so I can wake up feeling good, look at the data and then feel terrible because it's a bad night's sleep or wake up terrible and say, oh, I should feel good.

1:03:16Are you sleep maxing? I think the trick is to look at it as a trend, not as point data. I don't look at it. I don't look at any. When I've had a shitty night's sleep, I just don't look at the data. That's my very technical outcomes. Like, no, just don't look. I feel awful. That's fine. Get on with it. Stoicism. Boom. Mads. We keep interrupting you. Sorry, dude. Sorry. So that's Aura. And then we've got, on Wednesday, we've got Micron's results, which is one of the AI CapEx lie detectors. And is demand for memory still rising? We'd expect so because Jensen has said he's buying everything he can get his hands on and that's the only thing that's constraining him.

1:03:53So we would expect good results. The results are one thing. The other thing is the outlook. What is going to happen in the years ahead? What forecast will they give us? And the street will pay close attention to this. And Wednesday, right? That's Wednesday. Yes. lovely there's a big ii conference in san francisco so we might get some news off the back of that and then obviously it's bits and pretzels in in berlin which i'm going to for the first time are you going i might is wednesday is octoberfest day so are you and i gonna gonna have a couple of steins i don't drink i don't drink the alcofrol my man i haven't drank the alcohol you can have zero percent it's all modern over there now you can get zero percent beer okay i'll join you with a zero percent of alcofrolic boys and girls always a pleasure never a chore take care not heroin and for you watching and listening we'll see you all next week see you in the next one not heroin ciao ciao