This Week in European Tech: Anthropic’s rise, China’s AI push and Europe’s tech dilemma

0:00And so there was a ratio between the number of CPUs you needed and GPUs that used to be one to eight. So one CPU could drive sort of an average eight GPUs. Well, that's shifting because the new reasoning models and agentic workflows, they have all these tool calls. They're essentially quietly admitting the frontier AI is no longer just a tech product. It's strategic infrastructure, sort of alongside kind of nuclear, alongside biotech, more than just software. So given the size of the market and the demands and China's strategic imperative to catch up and overtake the US, I think it's only a matter of time.

0:36It's not if it's when. Well, you know, if they keep going at this rate and you project that out over the next two years, well, then Anthropoc will be a 200, 250, 300 billion revenue company. And the free cash flow on that is going to be pretty juicy. Welcome to Upside, the weekly podcast about all things affecting European venture. Now, the smart among you will notice that I am not Dan Boyer. Our good man Dan is out climbing mountains this week. We told him to stay safe and return in one piece, as we definitely need him back to moderate next week.

1:21This show is not investment advice, and the hosts of this episode may be invested in the funds and companies featured. With me today is our good Lomax Ward, our very own Life Sciences Luminary. Looks like he's back from the Lisbon beaches there. Looking good, Lomaxius. How are you doing, buddy? Yeah, that's the Parmatan you develop when you live in Southern Europe. Sweet, sweet. Good to hear. And we are today also graced by the presence of Dr. Andrew Scott. Our deep tech done. God, my parents wish it was Dr. Frotting adventures. Andrew, how are you doing, my friend? Very well. Great, great. Good to see you here.

2:01Well, what are we about to talk to today? First, we'll have a little roundup of the week of earnings releases. We've got some great new earnings out from Arm and Spotify to great European tech companies. We're also going to talk about all things Anthropic, the big behemoth on the AI block. After that, a little quick pivot back to the new fight that's looming between the EU and meta. Get out the popcorn, guys. Then we'll have a look at what's happening at ASML. They may not be Anthropic, but no ASML, no GPUs, no GPUs, no AI. So they are pretty important. Let's look at what their CEO has to share this week.

2:41Finally, we're rounding out the news about a new Fusion Consortium right here in the UK. Get a note of that. Now, before we do, look forward to what's ahead and a deal of the week. So let's get cracking. And to get started, Arm and Spotify, two very important European tech companies. They've both been announcing earnings this week and both had pretty good numbers. But markets interpreted them quite differently. Now, Arm announced lots of goodness around customer demand from AI, and the share price really has been great. It's been up 84 % year-to-date and took another pop-up this week. Spotify, on the other hand, well, they also had revenue numbers that were good, and they're growing, but the stock is down 5 % this week, 25 % for the year.

3:30And in many ways, it looks like this could be stories that both are AI-related but maybe pushing things in a different direction. Lomax, let's start off with you here. What's the read? Is this the AI economy bifurcating? Is this about cat picks or consumer spending or threatened opportunity from AI? What's your read? So, I mean, firstly, we don't often get to talk about big listed European tech companies. In fact, Spotify and Armour are basically the two preeminent examples apart from Adyen. And it's somewhat fitting or depressing, depending on how you look at it, when they actually post good numbers and the market doesn't really love it.

4:10In fact, the market kind of hates it. I think that's because a lot of the performance, even though these are record numbers for these companies, is actually baked into the price. And actually going forwards, there are some challenges that the companies are going to face. So in the case of Spotify, while the increase in the premium subscription numbers was well received by customers and didn't see a lot of churn, I think they did disclose that cost of acquisition of customers is getting more and ad spend is getting more expensive. So actually to acquire new users going forward, it's going to be more expensive.

4:48So that will put constraints on growth. So while the subscriber numbers have been growing nicely, like will that continue into the future? Arm is a little bit more nuanced and ties a little bit more into the AI story, of course. I don't necessarily see this as something to extrapolate more broadly into the market other than, you know, Arm has announced that they have. So maybe stepping back, Arm has historically been a license only business, right? So they license this architecture, which is pretty much in every single mobile phone in the world. They announced what the end of last year or beginning of this year, I think it is their move into actual AGI CPUs actually developing chips now.

5:28and that's been a big pivot to their business model. There was concern that that would cannibalize some of their existing business because they would be competing with people who are actually their customers. And actually the demand on that has been really, really strong. In fact, it's so strong that they now have massive supply chain constraints, which I think is one of the reasons why they do have a relatively lukewarm reaction on the market is because they've now moved into this supply chain bottleneck. But ultimately, I think it's a bold move that Arm have made. If you look at the revenue that the business projects by 2030, the CPU numbers are going to way dwarf the licensing revenue.

6:10So it's been a kind of bet the farm company existential bet by the management team. But I would suggest one is going to pay off in the long term, But there are going to be challenges such as this supply chain bottleneck they're finding at the moment. Interesting. I mean, Andrew, I mean, we're seeing Spotify sort of under threat. Is that the market thinking that it could be kind of new generative AI music that could displace traditional streaming? Or what do you think is happening here? Yeah, I mean, where would I be to predict the wave is the market and what people are thinking? but I don't think AI music is going to displace existing apps in the same way that I don't think AI movies are going to replace traditionally made movies.

7:05Normally the market expands so AI music will be used all over the place in different places but people are still going to want real people to fall over, real people to admire, to fall in love with from a distance. I don't think it's a threat. I don't think AI is a threat to the music industry itself, but I think it may replace some revenue streams where it's far easier to press a button and generate the perfect music for your TV commercial or something else where perhaps licensing would normally be the way to go. But even with that example, people use certain music in TV advertising because it evokes certain emotion from the audience because the audience associate the music with certain feelings or a certain vision or certain aspirations.

7:55Any thoughts on the expanded ARM strategy? Yeah, I mean, I think it's keeping up with the Joesiness, right? ARM was a licensing business for a very long time. They've got to continue to expand. And I think the current stack isn't going away in terms of AI and the need and the requirements. I think it's just reflecting the expansion of the market. I think, I mean, if you look at the ARM business, they're basically growing, you know, they're forecasting by 2031, 15 billion of revenue from this AGI CPU revenue, 15. So they've gone from 0 to 15 in five years. And their core licensing revenue, they're only forecasting it at 10 billion, right, by 2031.

8:37So what you're seeing is clearly they're basically monumentally moving the business in a relatively short amount of time. but as Andrew says it's probably something they need to do or they felt they need to do both as a defensive and an offensive move I would say just going back to Spotify I think the interesting thing which may tie into the broader AI narrative rather than focusing on distribution rather than actual content creation is actually if you look at the Spotify numbers they're pretty good you know for a mature listed tech company you know they're growing premium revenue was up 10 % in Q1.

9:18Gross margin was up 33%. Actually, the numbers are not bad, but the ad-supported revenue was down 5%. And that may be what people are interested in because that may be being disrupted by AI. So I think there's a question mark as to distribution for Spotify. Are the channels changing and what does that look like in terms of customer acquisition going forwards? Interesting. I suppose on the ARM side, I mean, there is another part around kind of the GPU, CPU shift. It's a while since we stopped thinking about CPU companies are being cool and it's been NVIDIA and the GPU businesses that have been driving the industry, but actually as AI shifting from sort of legacy Gen 1 generative AI models to agentic models, you're seeing a shift in the workloads where, of course, every computer needs a CPU at the heart to make sure it runs.

10:19And so there was a ratio between the number of CPUs you needed and GPUs that used to be one to eight. So one CPU could drive sort of an average eight GPUs. Well, that's shifting because the new reasoning models and agentic workflows, they have all these tool calls. And every time you have a tool call, every time the reasoning model of the agent goes down and calls the file system or fetches data from the internet or does some operation, that is not a GPU operation. Well, it's a CPU operation. And actually the expectations that the ratio could shift so that rather than it being one to eight, it becomes one to one, which means that either the amount of GPUs we're selling is going to decrease massively or the number of CPUs we're selling is going to increase massively.

11:02And so the CPU market is going to explode. And so I think some of the optimism is at the heart of this, of just how many more CPUs we might actually be needing in this industry. So that's potentially an aspect as well. Okay, very good. Well, let's stay on the AI lane. Let's move over to Anthropic because there's no week without major Anthropic updates. And the company was in the news this week announcing plans for$200 billion spend with the Google Cloud platform over the five years from 27 onwards. That's now 40 % of Google's nearly$500 billion backlog. And if you take the Anthropic and OpenAI promised spend with the hyperscalers in combination, that's now more than half of the$2 trillion cloud backlog of the hyperscaler.

11:55So that's obviously a massive number. Just this morning, the Financial Times reported that Anthropic has passed 45 billion in ARR. And listeners will recall that this is up from around nine to 10 at the start of the year, and a 45x on the start of 2025. So, you know, a staggering amount of growth that continues. The company is now working on a new round of funding, reportedly a$50 billion raise at a valuation that could exceed$1 trillion. So we've discussed an Anthropic IPO on the pod here, but the company seems to have so much money thrown at it that it might not happen in the near term. Now, at the same time, we've seen Anthropic and OpenEye both announce new joint ventures this week.

12:40Anthropic, they announced that they've launched a new JV with Blackstone, Hellman and Friedman and Goldman Sachs to build a services business that will implement AI at big enterprise firms. And at the same time, OpenAI has announced a similar initiative with TPG, Bain Capital, and Brookfield to implement AI at enterprise firms. Now, the idea seems to be that AI can unlock an ungodly amount of productivity in enterprises, but it's hard to get right, so we need consultants and engineers to make it work. Lomax, let's start with you here. I mean, what are your thoughts on the continued rise of Anthropic?

13:14I mean, when they finally do IPO, will they be the only company left worth paying any attention to? Yeah, it does seem to be going that way. I mean, I'm conscious that these things can change from month to month, week to week, day to day almost. But Anthropic definitely is sort of, you know, to steal from Zoolander, like so hot right now. You know, definitely it's the company that has a really strong positioning for a few reasons that we may talk about in the context of the other things that you mentioned, including, you know, AI safety and national security, etc. I think Dario has managed to position the company incredibly well in the way that actually Sam at OpenAI has well-publicized, had their struggles in terms of the perception in the market.

13:56And actually, if you listen in the corridors of family offices, the secondary market, the kind of private market, secondary market, there's definitely… Whose corridors have you been lurking in? Well, it tends to be digital corridors, right? so whatsapp groups generally or the the lobbies at conferences you find um that or maybe the washrooms at certain you know members clubs but you find that there is a lot of um appetite for anthropic in the in the in the private investor community and actually open ai is is nowhere near that and the fact that there'll be quite a lot more sellers than buyers potentially so i think that per your point now you know anthropic has the ability to raise 50 billion potentially at north of a trillion which is mental because the last round was a only very recently at the end of last year or beginning of this year was at 380 and actually at one point it was sitting at less than half the valuation of open ai and looks like it will surpass that quite quickly i mean that's fairly reflected not only in the perception obviously in venture we talk about you know narrative and fundamentals anthropic seems to be nailing both at the moment right so you know we talked about narrative when he moved to fundamentals um as you alluded to mads the the revenue ramp is insane you know now 45 billion of arr and and you know the having only been nine billion what nine billion at the end of last year is that right um it's yeah the growth is just we're only in beginning of may the trajectory you extrapolate that out one thing that we'd been and then that also ties into the capex where you're saying well look you know between open air and anthropic they're now accounting for um over half of the two trillion of capex projected in in the industry which has been a big concern for people was like well actually how are you going to generate the revenues and the free cash flow to generate to justify that capex well you know if they keep going at this rate and you you know you project that out over the next two years well then you know anthropic will be a 200 250 300 billion revenue company and the free cash flow on that's going to be pretty juicy and you know let's imagine that sort of 50 50 gross margin let's imagine they're spitting out at least i'm doing very high level numbers here i don't know 100 billion of free cash flow every year potentially i don't know like maybe you guys have a different view on that but if they can get to that position then suddenly the capex starts to look justified um that's a pretty good start what else what have i missed i think you also you know you mentioned the Google deal.

16:31You know, Anthropica have done a deal, five-year deal with Amazon. They've got a deal with CoreWeave. They're doing deals all over the shop. But, you know, they're feeling confident. The wind's in the sails. And OpenAI... And they've just done a deal with Musk, SpaceX. Yeah, well, there you go. Yeah. In terms of the new joint venture that they've launched and that OpenAI has launched, I mean, maybe that's one for you, Andrew. Is that, do you think that's likely to be sort of old-school Accenture-type consulting or more like a high-margin Palantir style business. Yeah, that's exactly what I was going to say.

17:03It's a sort of, it's a blend, isn't it? Kind of part Palantir, part systems integrator. And, you know, it's also, it suggests the bottleneck has shifted from model capacity to enterprise implementation and, you know, trying to embed your systems inside those companies. And I think it's also, it effectively locks up the securitizer kind of future enterprise cash flow by getting embedded deeply into these companies, right? Rather than waiting for that organic bottom-up adoption that you might see with the Slack or with the SaaS company. So they're building a kind of captive distribution through some of these PE portfolios and through some of these enterprise players.

17:59And I think one of the, I suppose, if you're looking at a centrarian view, one of those might be, does this risk that these companies, these frontier AI companies turn into very large, expensive customer services businesses rather than high margin SaaS businesses? if every single deployment at scale needs a huge number of sales engineers and on-site integration people, et cetera, et cetera. So it'll be interesting to see how it evolves. And certainly when we look at even early stage at our portfolio of companies, we did some AI very early on. And one of the biggest blockers was not that the product or service didn't provide efficacy or didn't provide a good result for the business or wasn't good value.

18:48It's just that the businesses, including enterprise customers, did not have the internal understanding or were not prepared to deploy this technology. So I could imagine that at scale, yes, you can have individuals at a desk running their emails through ChatGVT, but that doesn't really unlock the full power of AI to do that. You've got to be deeply embedded in these corporate verticalities. Yeah, and I suppose that will yield higher quality revenue as well going forward, maybe ahead of a potential IPO. Lomax, do you have a point there? Yeah, I think it definitely ties into the IPO story. I mean, it's no coincidence they're both lining up IPOs because they're, well, they were in theory running out of road on the private markets, but it feels like Anthropic might just be able to squeeze out one more round.

19:32100 % they are. Exactly, but they need to, this is definitely with a view to the IPO, this story i mean it's actually this this um these jvs sit at the confluence of i guess two trends like one is ai obviously and the other one is actually pe that's been going through a kind of down cycle right in terms of depressed returns um everyone feels that which is driven by a few factors like you know pe's got undergone i've undergone a cambrian explosion explosion in the last 20 years so you know if you've got so many companies that are optimized to the nth degree Like how many, how much more can you optimize?

20:11Well, suddenly AI comes along. So the PE shops are thinking, well, you know, our returns profile looks worse. What are we going to sell to LPs now? Well, suddenly we do these JVs with OpenA Anthropic. They can effectively juice the companies that we've invested in. We're also going to invest in the JV. I don't know whether they've done that as GPs or LPs. Do you know what I mean? Whether it's the PE firm itself. They're JVs. They're JVs. They're JVs. e-fund rather than the LPs of the PE firm. That does actually, in theory, create conflicts, but I'm sure they figured that one out. But the interesting point is that, is that there's 7 % of US GDP sits in private equity-backed companies, roughly 8 to 9 trillion in a market cap, depending on how you value it, sits in private-backed companies in the US, private-backed companies.

21:02So actually, if you can use AI to squeeze 10, 15 % more efficiency out of that and increase the EBITDA margins, which generally most of those companies get, you know, get, get, um, either increase the EBITDA margin or increase the actual underlying EBITDA, you are going to, um, you know, it's kind of a win-win situation. Anthropoc and OpenAI generate more revenue, maybe using this Palantir model with, maybe you need to have the kind of forward deployed engineer. Um, it will be kind of quasi service, tech service revenue, but that's probably okay at the JV level. But ultimately for the PE, which would be great front topic and open AI.

21:41And then the PE funds underlying portfolios will both ties into both narrative and fundamentals again for the PE firms, because they're like, this is actually going to improve underlying EBITDA. And to all of the LPs that are starting to fall out of love with private equity, like, look at this, AI is going to come to the rescue and help us in our fundraising and show us how we can take the private sector in the U.S. and juice it to the next degree. Everyone's a winner. Isn't it fabulous? Yeah, I think it's great. And actually, in a way, this is quite similar. This is not quite similar to the Project Prometheus thing in a way that Jeff Bezos is working on, which is in the context of robotics, Right.

22:26So Project Prometheus is building, you know, robots for, you know, industry for physical AI, for industrial automation. And then they went in conjunction with that and raised a massive fund, which will buy companies which need which will then, you know, implement the robots to implement the industrial automation. Quite similar to that, actually, in terms of asset owners doing deals with tech companies, whether it's a fund or a joint venture arrangement, to come in at a systematic level and implement that technology in the portfolio. The realization that with product-led growth, you can get so far in an enterprise that perhaps not very far at all.

23:14You need some services bots to make it happen. Let's flip over to the safety and security angle, because in parallel with all these exciting revenue growth numbers and the JVs and the funding announcements, etc., there's been continued discussion around Anthropic's mythos model. Now, we know it's a model so capable at hacking things that Anthropic initially withheld release. Dario Amode, who is the CEO of Anthropic, he recently flew to the White House. Now, you'll recall that the Department of War, They recently labeled Anthropic as supply chain risk due to disagreements over how and where the Anthropic AI could be applied.

23:52It seems that relationships have been patched up between Dario and the Trump administration. And in parallel, the U.S. Center for AI Standards and Innovation, they've recently signed up Google and Microsoft and XAI to some new pre-deployment evals. And from what I know, Anthropic is not in the framework yet. So to some extent, the focus here seems to be on AI safety, which maybe feels like a bit of a hark back to the Biden era. Only Trump doesn't call it AI safety. He calls it national security. So Lomax, one for you, when it comes to AI safety, is this about necessary regulatory oversight or is it regulatory capture?

24:35Or do we even know what to distinguish kind of good oversight from rent seeking in this space? yeah so um a few things a few things to unpack here i mean ai safety is feels like a bit of a dirty word in the tech circles which was you know um lauded by rishi sunak you know three years ago when he launched the ai safety summit which kind of died a bit of a death but now the americans having been quite um well you know let technology run like we need to let things um play out suddenly mythos has got people in the white house to wake up and smell the coffee probably not not um a coincidence that you know david sachs has just finished his role as the whatever he was ai czar in the white house so you know the the tech executives leave um the more political figures uh susie wiles the chief of staff get a little bit more involved hang on a second what's going on here um and look i i think this is not a european style let's go and regulate the hell out of this let's wrap a lot of red tape around this.

25:37But I think it is a signal from the White House. So Trump has a very advanced draft, I understand, of a 15-page executive order, which will effectively put this into practice of basically implementing a bit like the FDA for drugs in the US, like a bit of a review process in the context of new releases of models. Now, clearly, this is being done by the Republican, the Trump White House. I don't think they're going to be trying to put too much of a stopper on this and strangle and suffocate the speed, especially in the context of the broader race with China, right? But I think it's an acknowledgment that, hang on a sec, this stuff can be potentially too hot to handle and we need to have some level of oversight.

26:24So I don't think it's overreach yet. I mean, we haven't seen the details of it. There has been, the name I think is Casey, C-A-I-S-I is the acronym, which does exist already, which is a bit more of a kind of voluntary group of kind of which leads some kind of review process of these models. But that's not, you know, from the not a piece of legislation. So I would say that it's a step in a different direction for Trump. It is a step in a slightly more kind of, you would say, European or like Biden-esque direction. As I said, I think they've been a bit spooked by mythos. I don't think it's a bad thing.

27:05Let's wait for what for the details say. I think if governments want to have some level of oversight on this incredibly important fundamental technology that could be incredibly dangerous in the wrong hands, I don't think it's a bad thing. clearly you know if you went as far as the fda process which costs you know hundreds of millions to bring new drugs to market you know at least a billion these days depending on you know what data you use then you know you end up in a different in a different um ballpark because you will potentially stifle innovation um final thing i would say on this is that before you do before you go there long mike let me bring andrew in on this one because andrew you're normally sort of our pro-free market champion here.

27:47And I mean, is this effectively just kind of the incumbent frontier labs pulling up the ladder and saying, great, we had a completely free run to do whatever we wanted, steal all the IP, develop the models, and now we're going to put some government regulation in place, making it much harder for people to come after and follow us? Or is this necessary regulation? How do you think about it? Well, anybody who's listened to my droning on on this podcast previously will affirmably describe AI as the equivalent of nuclear weapons. If you extrapolate out 10, 20 years, that's the level of importance of this technology.

28:25And I think this probably reflects Washington's alignment with that, and that they're essentially quietly admitting the frontier AI is no longer just a tech product. It's a strategic infrastructure, right? alongside nuclear, alongside biotech, more than just software. So it's interesting that Trump sort of campaigned against regulation, but then he's essentially enabling this push forward of the AI safety that Biden was a big proponent of. but I think it also it's also good for the big incumbents because entrenched incumbents with lots of money can afford to navigate regulation, they get to write the regulation and if you end up with mandatory pre-deployment testing that does favour much larger companies compared to sort of much smaller earlier stage or open source labs, you know, it becomes a moat for those bigger companies.

29:38My speculation is what we're seeing here is a proposal to regulate from an administration that is notoriously anti-regulation. Is it fair to suspect or fair to say that maybe the only reason they are proposing to do this is because even they feel it absolutely is necessary? Yeah, I'm not sure I dare speculate. that's where the fun is you're meant to be a pundit the true conspiracy theory would be that the administration have so much personal assets invested in these companies that they're now happy to pull up the gate as they shut the door as they go through or pull up the drawbridge as you cross the moat once you've crossed it and that basically these companies have had a five year one a four-year free run at this and able to raise hundreds of billions well certainly tens almost hundreds of billions of dollars and now actually when it's become this big everyone's made all their money uh on paper and about something you suddenly pulled the drawbridge up i think that's probably a little bit cynical i think it's probably a combination of um the mythos model like and people actually people starting to get commercial people starting to get quite scared here i think it probably is like a bit of a change of like you know some people like maybe it is internal in the white house maybe it is suzy wiles maybe you know david sacks leaving actually is a big important thing and actually that um that there are people there who are generally skeptical of this technology or scared of it so i you know i i would you know i would distance myself from those conspiracy theories but i would be more i would think it's more something to do with that.

31:23Very good. Well, with that, let's flip back over from Washington to Brussels, where we'll talk about a new fight that's looming between Meta, which used to be Facebook, and the EU Commission. Now, we know that Meta, they've had their issues with data privacy over the years. But actually, this upcoming fight is not so much about privacy, it's about competition. Now, Meta, they've built AI into their apps. We know that Zuck, he has invested billions and hundreds of billions in AI and CapEx and staff and teams, etc., without having a hyperscaler or a commercial model. So for them, it's all been about building the AI into the business itself and benefiting from it.

32:05And one of the things they've done there is they've been including new AI models in WhatsApp, which is used by pretty much every single human being outside the United States. Brussels is fine by that, by the way, but they say that Meta needs to allow users to use other AI models, not just Meta's. Meta has said, fine, we'll agree, but then those other firms will have to pay. At Brussels, they've now rejected Meta's pay-to-access remedy, and there's now speculation that they could force Meta to open to other AI models. And it would be the first time the EU would force a platform to open AI competitors' access at no cost.