This Week in European Tech: Apple rents AI. What should Europe build?

0:00What some researchers are starting to say is, wait, wait, wait a minute. Where is this ending? Are we now letting the models pontificate on their own? Come up with their own language, do their own things. And not sharing that chain of thought. Apple has made the most rational decision in the industry and it's decided now to rent the models. And it means that Apple competes the layer that it's strongest in, which is hardware and distribution. and this matters way beyond Apple. Every big company is arguing about whether to build its own AI or pay for someone else's. Apple just made renting more respectable.

0:36If they don't build their own model, then the other players don't have to. I've for a long time on this pod and elsewhere been whinging that there are too many grants out there and it motivates the wrong behaviours within start-ups and the government needs to write contracts to solve problems that ultimately will help the taxpayer. Hello and welcome to Upside, where every week we dig behind the headlines that are going to affect European venture. Today it is Andrew, Pri, Mads and myself. And what are we talking about? Well, we've got AI prices are up. Has the free lunch stopped? Bonds, why so hot right now?

1:10And what does this mean to us? Circular AI financing moves from equity to debt. Is this panic time? Predictions, the week ahead and deal of the week.

1:32This show is not investment advice, and the hosts of this episode may be invested in the funds and companies featured. Boys and girls, how are you, Andrew? How are you doing, my man? I'm very well. Facing down the onslaught of September as we go back into the final quarter of the year. Friday the 4th at time of pod, So we are now all looking forward to back to school and off the yachts and back into the offices and doing amazing stuff. Pri, tell everyone a little bit about you because you've just joined us at SuperSeed. You are ex-founder, investor, now VC. Welcome, welcome. Tell everyone a little bit about you and what you've done.

2:13Yes, great. Thank you for the intro. So hi, everyone. I'm Priyanka. I've recently just joined SuperSeed. and prior to that i was a founder of a cpg company where we were selling healthy snacks into major retailers into sainsbury's boots wh smith and then into the food service channel just to interject you brought some into the office super delish yes so i was running that and now looking at of course investing in the ai in the physical world in manufacturing that the world that i knew well we ran two factories we ran warehousing distribution and now looking AI products built exactly in this space.

2:51Because you saw what we invest in, not as a consumer per se, but on the other side of the desk, which is great because you know exactly how you want manufacturing to run and supplies to run. Exactly. Thank you for joining us. It's a pleasure to have you here. We've got a pretty full docket, but Mads, my man, what is cooking with you? What is new? How are you? We're back in London. It's September. Summer's over. It's back to school. Business is humming. It's good, isn't it? And do you know what? It's been raining cats and dogs this morning. So it's proper London now. Yeah. Not this wishy-washy summery thing.

3:31Yeah, yeah, yeah. Well, we're going to be bouncing around some quick news. Then we're going to get into the meat and potatoes of today. But I wanted to just start with Broadcom because they missed estimates. Not by much, but the market kind of punished them. Stock fell 3%. AI chips tripled and it still wasn't enough. Google just signed 120 billion custom chip deal with Marvel and took a$12 billion stake in them to do it. So their customers, Broadcom's customers, are buying the competition. The hyperscalers are understandably hedging. Mads, you chatted about this and you keep an eye on these things.

4:06What's happening with Broadcom? We spoke about Broadcom last week saying that I thought they would have a good quarter. And actually, they did have a good quarter. Results were very good. in terms of the Q4 guidance, that came in a little bit below what the street was expecting. So, and the stock was down, although not much. I think for the week, it's down about three, 4%. I think their forecasts are really interesting. So this year, they're expected to do 58 billion of AI business. So kind of custom AI chips for various providers. They're guiding two years forward, right? So a lot of confidence about what they're going to do next year and the year after.

4:43And they're expecting to double that$58 billion for next year and then double again the year after. So the pipeline they have from everybody who is now piling in and wanting to build their own and make their own silicon is looking really good. And I actually think they're encouraging results. I mean, just like NVIDIA that was guiding forward in the results we saw last week, everybody is really, really optimistic about the direction of travel. I mean, missing by 200 mil on 35 bills sounds negligible, but more to come is what you're saying. Watch out, watch Broadcom. A lot more. One thing that caught my eye this week, and I don't know if anyone has any opinions on this, but it did, I don't know what the mood is here, but it kind of made me smile a bit.

5:26So the Dutch have moved their gold from the US to London, 86 tons out of North America for crisis preparedness, I think was the official reasoning. so they sold in New York and then bought in London rather than in my head I saw them shipping like hundreds of tons of gold across the ocean that's not how they did it so they um they own 600 tons now they're ninth on the list the US being first with 8 ,000 tons which is the same as Germany Italy and France combined and the UK is 18th with 300 tons and obviously this is an inflation hedge everyone's going to gold for the safety but I don't know if anyone a bit of an odd one to ask on this pod, but I don't know if anyone's got any thoughts on the Dutchies moving gold from North America.

6:10The other thing that caught my eye was Matt Clifford joining Anthropic as Managing Director of International Affairs. Pree, did you see this? Yeah, I did. So I guess Matt Clifford, the person who wrote, I guess, a large part of Britain's AI industrial strategy, had concluded that the most effective place to shape AI policy is inside a US-lad international affairs function, which is where he's moved to. And I think that's a clear verdict on where leverage actually sits. It's at the US labs. Yeah, he seems to have kind of jumped around a little bit, but it's great to have him in the fray still. It's a shame that he's not in the UK government.

6:49Mads Andrew, you got any thoughts on Mac Clifford at Anthropic? Yeah, I mean, look, he was in number 10 when the Starmer government kicked off and he left after not too long. I think he's pulled together a good plan and a good strategy, but obviously he's concluded that he can have more impact and more effects working for Anthropik than working in the UK government. And I do think it's a shame. I think he was an asset and a resource here. He'll be able to do good things wherever he is. And I think he follows in the footsteps of other ex-officious Brits that are going to work for big US tech, which maybe tells you something about where the power balance lies.

7:28And I can imagine him doing, he's very diplomatic, I can imagine him doing great things as managing director of international affairs. It's a shame that the UK government can't keep hold of entrepreneurs. Andrew, did you do anything back in the day with UKGov? I've done a bit of informal stuff here and there, and I've done sort of via the MED and that sort of thing, but I've had no glamorous formal position. No international matter mystery of international affairs. No nine to five at number 10 in the future. Well, Matt has been in there a number of times. I think we're star more, isn't it, Mads?

8:01I'd like to see more entrepreneurs. I saw Alex DePledge is back in, which is great. I think she's wonderful. And I'd like to see more entrepreneurs and more founders and more business people in government. So if you were to move on, we'll take note. Another thing on the quick news list. So John Turner's Apple. It's his first week. He's got a huge launch next week on the 9th of September. One of my happy places is the new Apple launch, being an Apple fanboy. The foldable iPhone, new Siri. and they've got a product guy who's the CEO, which is music to me. Pri, have you got any thoughts on Apple here or John?

8:35Yeah, so Apple has made the most rational decision in the industry and it's decided now to rent the model. So Siri now runs on Google Gemini, which is costing them about a billion dollars a year. And this is a very small cost against their$4.6 trillion market cap. And it means that Apple competes the layer that it's strongest in, which is hardware and distribution. And this matters way beyond Apple. Every big company is arguing about whether to build its own AI or pay for someone else's. And building is always the easier thing to defend to your board. Apple just made renting more respectable. If they don't build their own model, then the other players don't have to.

9:16What they tend to do is buy and then build around or build on top, which I guess is an old tried and tested method. But this time, I think they've just not come up with the beans for AI. So I think they've had to. I think this is they've tried and tried and they've gone, OK, all right, well, let's rent it in. But I wonder where the line will start to move over time as they start building their own stuff. Also in the news end scale. I don't think they've really tried, have they, Dan? Well, who knows what goes on behind closed doors? All you see is the people shuffling, right? Yeah, but you know, because when Meta tries, I mean, Zuckerberg, he'll go out and he'll splash the cash and everybody will know about it.

9:52But I mean, this is, it takes so much resource to really try here that you will know about it. Silicon Valley will know about it. And I think, look, I think at the end of the day, you're absolutely right. They've been sitting in this one out. It does make Apple, I think, a slightly dull company. I mean, the products are beautiful, but they're leveraging things that were invented 20 years ago. And it's been great for cash flow, but I don't think it's maybe so good for innovation. And I worry that, you know, whether they can attract the talent that will create the things that they need 10 years from now.

10:21Obviously, now they're still milking the cash from all the smart decisions that were made 15, 20 years ago. Well, their model has always been, isn't it? Build things that others have done better and do it later, as in wait, let the market kind of create. Let Samsung and Google do the innovative stuff, copy it, make it Apple's version, and then roll that out for the next two, three years. Andrew, what were you going to say? Is that true? I mean, the iPhone was revolutionary. I mean, the iPhone wasn't just a copy, a slightly better version of the Nokia N95, for those that are old enough in the tooth to remember.

10:56I wonder if this is the first early death knell. It's the canary in the coal mine of sort of Apple's demise of the company that builds products that revolutionize an entire space. Because why are they launching a foldable iPhone? I mean, it feels like an also-ran. And it feels like, you know, when... How very dare you. It feels like when he left and they started launching hundreds of products and, you know, it was just another tech gadget company. I really worry about this because it's not something that is going to transform the foldable phone market. It's going to be an Apple version, something Samsung's been doing for a very long time.

11:37So I'm skeptical that this is a good move. Foldables have been out for, I mean, six, seven years now. And I think I'm pretty sure Samsung have built the screen technology that Apple have adopted, which is why they brought out their foldable before, because obviously they want to take a march. But I think this is exactly what they do. Well, they've got the distribution there. I mean, they're going to roll on the ecosystem play. So they are going to keep bringing out these bits of kit later and better than anyone else. But they'll own the ecosystem. They will allow you to run local AI. I mean, this is probably a pod for itself.

12:15So let's move on. But I want to have this conversation with you again. I think there's, I couldn't agree less. I think they're going to end up owning it after all the CapEx crazy over there, where Apple hasn't stepped in. But talking about CapEx crazy, N-Scale listing, London-based, New York listing, 103 in contracted revenue coming through. I don't know, Pri, if you've got any notes on N-Scale? Cool. So this announcement has been made ahead of an IPO offering that we think could launch was as early as it's sort of around now so early September and the underlying context that's quite critical for anyone who's looking at the upcoming listing is what does this figure mean it's it's a backlog it's not their current turnover this metric represents future revenue tied to multi-year lease agreements for data center capacity much of which has yet to be built and end scale's estimated actual revenue for q2 2026 just gone was 100 million up to 37 million in q1 um secondly they have high customer concentration the figure recently doubled from previous base sign of 51 billion primarily due to a single monumental deal which was of course the um 45 billion dollar contract with anthropic to rent compute capacity at ncl's west virginia campus and then the third factor is the annualized breakdown the signed contracts carry an average duration of around 5.7 years which breaks down to an annualized average of roughly 18 billion dollars So that's breaking down, like, where's this thing coming from?

13:42It's not revenue. Yeah, and we've got an even bigger listing that we're going to talk about a bit later. Wave is on the streets. Another thing that caught my eye this week, we've started seeing driver in, so not completely human-free, but we've seen wave on the streets. Uber cut 3 ,300 people. Pri, any thoughts on getting into a self-driving Mustang? I don't know if anyone, has anyone been in one yet? I don't know. Been in a Waymo quite a few times and I love it. First time very scary in San Francisco. And after that, you love it. It's a great experience. Because robo-taxies and robo-cabs, sorry, all over Austin.

14:24That was all over the social web. All these super sexy gold driverless cars now all over Austin. I just don't know how they're going to work in London. Did anyone see the Sky News reporter who got into, they were doing the promo with the lady from Wave and they got caught in a stock oil protest and the car didn't know what to do. And he threw a hissy fit. And I was like, well, stuff is going to happen, right? I'm sure had it turned into the street, it would have worked out that there were people glued to the road and stopped. But that was not a great entry point for them. Yeah, they're not actually driverless, are they, Dan?

15:02They're not driverless. No, no, they've got bodies in for now. They're going to have bodies in for another six to 12 months, I think. so yeah they've got to get training well i think there's also um it's hard in london because of the way that our streets are made they're sort of they're very windy it's not the sort of grid system of the u.s that makes it a lot easier to navigate but and waves bet is on the licensing model and they're building the software not the car so they've got they've got a system um called like an ai driver that they license to car makers and fleet operators and this is the bet that makes them different to Waymo.

15:34They don't have HD maps. Waymo pre-maps the city centimeter by centimeter before it can operate there, which is why it expands city by city. And it's slow to our dismay, waiting for it to come to London. And Wave system learns to drive the way that the person does, the way that you or I do from a camera input and experience, which is what we've now seen on the roads of London. So in principle, it drops into a new city without having to do the mapping step by step and that's the picture of the business which is why they can claim testing across hundreds of cities at a time and this is smart they're able to deliver a software margin quickly on on someone else's capex the car makers will buy the cars and wave sells them effectively that the driver brain and that's exactly where you want to be in robotics when the money is expensive waymo has the right i think i'm not sure i want i'm not sure i want a funny feeling I've got a funny feeling Tesla, with their simplified camera-based technology, is going to win out here.

16:32And I'm not a massive Elon fan, but I think I've got a funny feeling that Tesla is going to take the biscuit. Andrew, what were you going to say? I was just going to say, is robot drivers driving like humans a good idea? I thought AI was supposed to advance humanity, not drive like all the idiots I have to put up with when I get behind the wheel. Well, you just don't have to smell the armpits. It's fine. Andrew, one for you. So 100 million sovereign AI competition, government as first customer. The thing that caught my eye that I thought was great was the fact that this isn't a grant. This is the UK government putting 100 million to a pot and saying, look, compete and win and see if we can solve some public services.

17:11It doesn't feel like a lot of cash, but it feels like a great place to start and a great thing to do. Andrew, I don't know if you've got any thoughts on this one. A hundred percent. You'll be delighted to know, Dan, I'm all positive about this one. It's a great step in the right direction. For a long time on this pod and elsewhere, I've been whinging that there are too many grants out there and it motivates the wrong behaviours within startups. And the government needs to write contracts to solve problems that ultimately will help the taxpayer. And so I think this is all thumbs up. Yes, it's only 100 million, but if it goes well, if we see technology deployed from particularly UK companies into UK government departments or public entities, It's all positive as far as I'm concerned.

17:54So it's a great, great step. And I'd like to see more of it. No, I agree. I totally agree. And yeah, it feels like the beast has been awoken. And hopefully we'll see more of these. German, semi-random, but bear with me. German foreign domestic investment is up 50%. US investment is down 44 % into Germany. And the UK is now the largest investor. So what does this mean? Europe is going alone. Capital is leaving the Atlantic both ways. German investment into the US fell 65%. So we're starting to see that Trump dislocation of the global trade is now coming out and the numbers in a big way. Another one for you, Andrew.

18:34So the Lords wants a kill switch in the UK on AI systems. Can this even happen? I always think of the Lords as a crusty bunch of useless so-and-sos not really having their finger on the pulse of tech. I don't know if you have an opinion on Lords wanting an AI kill switch. So this is about having an ability in an emergency scenario to sort of kill an entire data center. And it would be mandated that data centers can be essentially unplugged. The big problem is that it's wrapped up with legislation which would potentially pause or stall or prevent development of AI, particularly towards sort of AGI.

19:14Now, you guys all know I'm pretty bearish on the chances of us extrapolating current LN technology directly to sort of superhuman AI in the truest sense. But I think that that's a bit of concern because anything, as far as I'm concerned, which prevents us keeping up with the Joneses, keeping up with the US and China on development of very advanced AI is a problem. So I'm worried about that. The kill switch I'm less worried about. But obviously, you've got to make sure that that kill switch is secure. Otherwise, you can tell someone else I'm plugging your... Well, it's not Eric Schmidt's kill switch, which was basically going to bomb the bastards, I think.

19:52He was like, listen, if it gets too much, we're just sending some B5s over and take them out. So I don't know if that was in your mind. I think the story here for me is not so much whether you need to kill Robocop or not at the edge when we've got this super AI. it's more the fact that even AI, which is not, you know, super AI intelligence, whatever you want to call it, can still cause very serious problems. And we have seen these situations, one of the most serious ones, I think, where sort of a developer was trying to change the code base and the agent went off and started to discredit the developer online because he didn't like what the developer was doing.

20:31So you don't need to have, in the same way that, you know, people who aren't the smartest tool in the room can cause you big problems in life, The not smartest LLM could still cause us big problems in life. It's a growing issue. And I think we're going to see a lot more problems, but also a lot more pressure and policy lobbying for the sort of safety. It felt like this week we saw a lot more kind of that risk threshold being hit. I don't know what is, you know, PR and the noise machine, but it feels like there are a lot more stories this week. And we've seen more recently where AI is starting to create that level of risk that people are starting to pay attention, that sovereign level of risk.

21:14I'm talking about frontiers and thinking things and AI thinking machines is looking to raise another billion at 40. They wanted 50 last year. Secondaries marked them at 13 last month. So it's a slight, not down round, but slightly maybe less than what they wanted. Mads, have you been tracking in your AI corner anything on thinking machines? I think Miramirati, obviously extremely talented, but I think is really struggling to position the company. They released their Inkling model earlier this year. I love the naming conventions, yeah. A lot of talk about it for about five minutes, and then people forgot about it again.

21:53And I think they're really struggling to break through and position themselves because what does the company stand for? Well, it is sort of looking like it could be the open source or open weight American model. But, and I think that's one of the next thing you have coming up, NVIDIA is now leaning heavily into open source with all the resources they have at the back. We talked about the Hugging Face deal last week. It's now been confirmed. They are really looking to pay$13 billion for Hugging Face, which will allow them to really put their arms around the whole open source movement. On top of that, they're looking to spend, I think it was at least$6 billion a year are training their open source NemoTron model, which is, you know, when you think about just the technology and the skills they have access to, I think it's just going to be really, really hard for somebody like Thinking Machines to come out under that weight.

22:40And so I just, I think she's going to have to come up with a different way to position the companies and explain to us, what is it really about? And what is it going to give us that all these other great open source models don't give us to justify the kinds of valuations she's looking for? Yeah, I wonder where they'll find a position. Obviously, it's a very busy space and there are lots of people with lots of money doing amazing things. Well, let's talk about it. Well, 100%. I mean, let's bring it back to Mistral, right? Let's bring it back to our own European frontier lab. And it could be just great things in the open weight space concluded that, you know, we probably can't compete at the frontier.

23:16And instead, what they're doing now is a lot of taking the open-weight models out of China, sometimes reshaping them, retraining them on some European data sets, and then packaging them and then wrapping a lot of services around them and working with lots of European corporates to implement AI technology, which is a credible position, right? These companies all need AI and they need the best partner to implement and make sense of it. So I think they've found a position and that seems to be working. And I think that's what Thinking Machines is still working for. On the Hugging Face deal, what do you think this is going to mean for open source?

23:49How will this play forward? More. This is more, right? I mean, so obviously Hugging Face wasn't struggling to fund themselves, but they've raised, I think, about$400 million in funding, have built a great business, phenomenal what they've done for the open source ecosystem, open weight ecosystem, but now they're going to have all the weight of NVIDIA behind them. Now, question mark, will the deal be approved? well it's not on the european docket though is it it's not for the aren't they don't they slide underneath the revenue threshold at 150 mil or something crazy like that who knows who knows how people will look at this i think uh if one wanted to take a very stern view one could argue that there are you know competition issues here now i i you know take a step back i think lots of m &a activity is good it's good for innovation because it recycles capital into the ecosystem And so, yes, if the deal goes through, I think it will mean more open source.

24:44It will mean open source with lots of resource behind it. I think the irony here is that since Brexit, I think probably the UK is the only European, inverted commas, country that could probably have a popper stopping it. But anyway, I want to talk more about AI. So two big model releases this week from Anthropic and OpenAI. We've seen Fable 5.1 and Astra. We've seen a number of stories about limiting bioweapons risk. I think, like I said before, we've seen more on this kind of risk, this sovereign risk stories in the news cycle this week. Meta's putting prices up. Is the free lunch over? The other fact that caught my eye, which I thought was quite interesting, according to ramp.com, 80 % of open AI and Anthropics enterprise revenues come from about 1 % of their customer base.

25:28I kind of look at stats through slightly blurry lenses because stats, stats and damn stats. But Mads, AI Corner, what's caught your eye this week and where would you start? So the models are fascinating. And I think just picking up on your comments around Fable 5.1 and Astra, OpenAI, they've been pushing, pushing, pushing, and they've been cooking all year. And they've been keeping saying, look, we are going to catch up and we are going to deliver something formidable. And with Astra, it looks like they've done something or pushed something into the world that certainly competes with Fable and Anthropic and maybe even is ahead on some benchmarks and some scores.

26:05It's not a clear-cut victory. I think there's still areas where you'd say Fable is the preferred model, but they're neck and neck, and I love the competition. Now, you're touching on spending and revenue, and obviously we're seeing spending explode. This has become really, really expensive, and I think we've seen innovation from a side this week that I hadn't expected. I think Meta has pulled an absolute blinder because what they've done with the new Muse Spark 1.3 that's been released is they've released two pricing models. And so they've leaned into something they've done better than anybody else on the planet, which is this real world of monetizing your data.