Anthropic Goes Full Stack and the AI Power Shift Begins

0:00Hello and welcome to Upside, where every week we dig behind the headlines affecting European venture. Today it is Mads and myself. We're going solo. What are we talking about this week? We've got Viktor Orban, who's been ousted from Hungary. What does this mean for European startups and our ecosystem? Anthropics app builder. Who's going to get crushed? GPT Cyber versus Mythos. The solo founder unicorn is here or is it? Compute scarcity and what that means for markets Musk in space Deal of the week and our predictions

0:46This show is not investment advice and the hosts of this episode may be invested in the funds and companies featured Now Matt's my good man, how are you? You've been away on holiday, haven't you? I'm terrific, yeah. Just back from Italy. Such a beautiful country. Pasta and wine, pasta and wine by the sea. What was that? Oh, and plenty of gelato. Copious amounts. Love it. Did the girls have fun? They had a blast. It was so good. Oh, lovely. Good for you. It was an extended Easter as well. Good for you. You didn't miss much back here. I think we had our summer in London. We had our one day of summer while you were away, so you didn't miss much.

1:19What's up with Lomaxius? Are you not joining us? No, he's underwater today. Andrew's underwater. So it's just me and you. But it's lovely. We get a little super seed loving, me and you talking, shooting the stuff. I won't swear, shooting the breeze. And we're talking about something that doesn't feel necessarily all that tech centric or all that investor centric, which is Viktor Orban being ousted from Hungary and Peter Magyar coming in as the new PM. So something I didn't know was that Hungary is or was hopefully, hopefully in the past tense, was the most corrupt country in the EU, according to Transparency International.

1:58Orbán is now out. On the surface, this looks just like a European politics story. But for us, we want to talk about capital flow, it's about investing, startups, founding, and how all of this is going to affect us in Europe, because I think it is. So Viktor Orbán, when I was looking at his legacy, he started out as a very progressive, smart, kind of cool dude, but spent the last decade looking after Russia as China's circuit breaker in the EU, blocking Ukraine's financing deal, sanctions packages and rearmament loans. It became quite difficult. I don't know why. Obviously, maybe we never will.

2:36But Peter Magyar is now in. He takes over midsummer, hopefully unblocking 90 billion in euros of EU funding, 20 new sanction packages. the Hungarian defence spending bill will get greenlit and some Russian financial assets might come through and hit the system. So real money moving through the network is what we're hopefully looking for. Before I ask you to tee this one up for us, Mads, what are some things that are on my mind? Maybe you'll cover these, but who is Peter Magyar? Do we know anything about him? Will he be the unlock that we hope for or will power corrupt him too? Are we going to see more capital flows?

3:10If so, will they be meaningful? How quickly will they hit if they do come? And it looks like at first inception of defence spending only plan or story. But Peter Maggio has also been talking about turning Hungary into a knowledge-based economy. I'm not sure if he can do that or how that's going to work. But where would you take? What's your take? There's a lot to talk about here, Dan. I think you're absolutely right. This is not kind of square centre of venture capital and technology, but it's very much about Europe. And it's an important component of what's going on in Europe. Now, Magyar, he is not kind of a lovey-dovey liberal internationalist, but he is much more pro-European and pro-EU than Orban was.

3:55He comes out of Orban's party. He's a renegade who was under Orban's wings and left the party to set up his own movement. And he's now trounced Orban, his former leader in this election. It was decisive, right? I forget the numbers, but I I remember seeing the seated chart, the parliament seated chart. It was like a proper trouncing. Very much so. And it's interesting to try and dissect some of the backdrop. Now, obviously coming from the land of the Brexit, which we are in here, you can sometimes be forgiven for thinking that there isn't, isn't there a lot of criticism of the EU everywhere and all the bureaucrat bashing and what's Brussels done now and all this stuff.

4:36Macchiar, a big reason for him winning the election was that the Hungarians were so sick and tired of Orbán's stance in the European Union and the fact that he was siding with Russia and China rather than with friends and partners here in Europe. And people wanted to be closer to the Union, and that's why they threw Orbán out. And that's really interesting. Ten years ago, at the time of the referendum here in the UK, people were saying, oh, the next is France and Italy and Greece, everybody else. Well, it turns out that actually the EU is really, really popular across Europe, in many places more popular than local governments.

5:14So that's an interesting context, I think, for this. Orban, as you said, he gave Russia and China a veto around a lot of things in the EU. That's not very helpful. It's almost sort of like a fifth column that undermines things from the inside. Hopefully, that's over now. I don't think we should expect kind of a total 180 from Hungary. I think they'll still have the reservations. There's still a lot of business they're doing with China. And in many ways, that's fine. We want to do business with China, right? We just don't want to give China a veto over our foreign policy. And I think those are important things to separate.

5:52So what does it mean for us? Huge unlock around defence. It means that Hungary, they can now unlock their own safe application, kind of access to the European funds for defense investment. It means, as you say, some of these vetoes are unlocked around what happens in Ukraine. So there are lots of good things there that will help things move better and faster. I think if you look at downstream, again, defense is going to be a big component of this. One of the things Macchio has been talking about is turning Hungary from what it is today, which is still very much an assembly place where you assemble things from kind of Chinese manufacturers and even BMW, I think, as some of us cast there and others, to more high-value-add.

6:34And what is that? It's a bit of a slogan, but I think people are looking to Poland and looking to the way that economy has transformed and say, we would like some of that here as well. And, of course, you can't talk high-value-add without talking about tech and software. So there is a hope and expectation that Hungary could become the next sort of tech hub or tech growth leader in Central and Eastern Europe. And hopefully they'll find a way to get it done. It'll take a while though. I mean, Poland was on the rise for, what, 15 years maybe, you could argue. So I think it's going to take a little bit of a time before his knowledge-based economy and his new infrastructure, and also how you get beyond all the corruption and reinstate all the institutions.

7:16That's the big risk. And arguably, Hungary starts with worse institutions than Poland did. So there is a hill to climb. Well, good luck. Break legs, wallets, hearts, whatever you need. Peter Magyar, I hope you make it a roaring success. And we do see, because my sense is we'll see some fairly quick flips in the EU with regards to all of the things that were blocked by Orban. So hopefully by the end of this summer, we'll start to see some movement and some action, which I think will help the European story and the positivity. Well, I think one of the questions you asked was around, you know, can it unlock the 28th regime?

7:53game, what can you do for startups? I don't actually think Orbán was the biggest problem there. I think we have plenty of other kind of vested interests around Europe that are blocking those things in itself. I think it was actually more about defense and foreign policy. And those are also important components. Well, that's all right, because I mean, there's so much of a defense push across the VC industry, across governments, across startups, that there will be some capital flowing through those good hills. So I kind of put it all in the same bucket but we'll see let's go further afield let's talk about anthropic we did do a bit of a a mega deep dive into anthropics business model last week but there's more as in there's another product it's anthropics app builder let's ship something great which was leaked you could argue maybe this week um but anyway whether it was on purpose or not we will we will find out i'm sure but six months ago on this show you called it you said frontier labs will either become application companies or consolidate into hyperscalers and you were right so this is just one of their latest moves anthropic is now building the full stack app builder they've hired workday cto and from what i can tell they're now hoovering talent from all the incumbents no longer focus just purely on clawed they're building product so this also means that the margin story has now collapsed.

9:19So that's going to be a challenge for other startups, obviously. If Anthropic owns the model, the interface, the design, the sales motion, the enterprise value, everything, they keep the full value stack. B2B SaaS founders, remember them? Well, they used to be able to differentiate by building on top of these frontier models. That is no more, I would argue. It also does raise the specter of another thing that should bother us insofar as Europe has no equivalent move. OpenAI is doing it. Anthropic is doing it. XAI is doing it. We don't have a European frontier lab. Now, I put in brackets in my notes here, Mistral and Amy.

9:57Yes, but we don't have the capital. We don't have the operator talent. We don't have this vertical integration piece. So the US labs, it looks like they're going to own the whole stack. So is that fair? Where would you start? What would you pick up on this piece? Well, there's a lot to unpack. And let's take some of those bits. But I just thought it was worth, before we go there, to talk a little bit about Mike Krieger, who is a fabulous operator. He built Instagram as a co-founder, sold that, built a phenomenal business there. And then he left Meta in 2018. He created another company called Artifact, which failed.

10:35And then he joined Anthropic two years ago as chief product officer. and he has just been absolutely cooking there. He shipped the model context protocol, which everybody rushed to integrate. He didn't design Cloud Code, but he was part of driving the revenue ramp around that. He also joined Figma's board last summer and Dylan Feldover from Figma was really excited about having him on board. But what happened now in the last week is he's left the board again, just as it looks like Anthropic now, is creating a design challenger to Figma with all the AI goodness inside. So good or bad, I mean, you certainly hear somebody who is moving and shaking and making things happen.

11:16Another thing to mention for all us Anthropic and Claude fanboys and gals out there is it looks like Opus 4.7 is coming soon. Ooh, you're going to be a happy man, aren't you? Oh, I mean, it's like Christmas coming early. It's just so good. It's so powerful. I mean, 4.6, I mean, I've told you this before, but the kids and I will ask it questions in the car and then ask it to give us a quiz at the end of the quiz. My kids are five and seven. So it's just, even on that basic, basic level, it's just like talking to a buddy in the car. It's just so smart. I love it. Yeah, it is really powerful. But you asked a bunch of other things, I think, about what happens with SAS and what happens with European Frontier Lab.

12:00Do you remember SAS? There was a thing called SAS. Do you remember that? What's that? What's that? No, I think European Frontier Labs, I actually think Mistral is doing much better than we feared they might a year ago. They're growing. They're growing rapidly. And the approach they're taking is very much that of being the AI integrator and the AI enabler for European enterprise. I think it's a solid strategy. A solid, solid strategy. Rock solid. And they won't be the only game in town. We know there are others coming through the system. Anthropic, I mean, you look at them right now and you just sort of say it's hard to cache them.

12:39But we also know that this space is moving so fast. And actually, if you look at what plot code is, it's beautiful, but it's also something others will be able to replicate in 12 months. So what is the probability that Anthropic forever and ever will be able to have the best model and the best coding scaffolding around it? But look, I think some of these things, some of the margins will narrow over time. And that is, of course, why they keep expanding their surface area to go after other places because they know they can't stand still. Everybody's just moving super fast. We've got open source waiting in the wings.

13:16So much happening. And it's just such an exciting time to be in the tech industry. Well, let's talk about because we saw this week, we saw GPT 5.4 Cyber going up against Mythos. We've got this cat and mouse game, haven't we, where when one of the big, big teams releases a thing, the other big teams release the same thing. We've just spent a little bit of time talking about Anthropic eating the application layer. And rightly so. Here's another round in the chamber. Their cyber tool, Mythos. So I'm going to ask you a bit more in a second as to why chat GPT 5.4 cyber could be released into the wilds.

13:56but Mythos still can't be. So I think that's a really interesting aspect to all this. With Mythos not being released and not being distilled by the Chinese yet, I want to talk to you about, is there enough of a time gap? Is that going to be enough of a little bit of a Moti gap away from the Chinese open sources? But let me hand over to you. Give us a bit more of the background and the benchmarks and how this is all happening and how these two tools compare. Super interesting to see what's going on here. We know these benchmark games are just making everybody's heads spin because there are so many benchmarks now and everybody's got a benchmark where they score phenomenally well done.

14:38So Mythos was kind of not quite released, but certainly talked about last week. And they were talking about this exploit-to-vulnerability ratio, which is around, you know, can you find the vulnerability and then exploit it, create some kind of hack using that vulnerability. and they scored 72.4 on that benchmark. And then OpenAI, they released their GPT 5.4 cyber tool where they had another benchmark called CTF or Capture the Flag and they scored 76 % on that. When it sounds like 74, 76, 76 is higher than 74, does that mean that the OpenAI tool is better? The truth is they're measuring completely different things.

15:20that capture the flag is a test-solving approach that asks the model to solve known pattern puzzles in controlled environments. So it's the equivalent of what sort of a mid-level pen tester might do if they try to break into a system. Say there is a token, there's some text, there's a flag hidden inside this application or inside this infrastructure. Find a way to retrieve it. And that's kind of a useful way to see whether you can break in and you can find the thing you're trying to find with some known vulnerabilities. Now, the exploit to vulnerability thing Mythos is testing is different. That is just saying, here is source code.

16:04We don't know whether it works or not. You figure out, A, whether there are vulnerabilities, and B, you prove that you can create code that exploits those vulnerabilities to get access. Completely different level of benchmark. Completely different challenge. So this is just a prompt, right? This is just a very, anybody, you don't have to be a security engineer, anybody can just throw a prompt into Mythos and it will work stuff out. That's the smart piece, isn't it? Yes. And they're not challenges that are engineered. It's not a benchmark that says, oh, here is some known problems, some patterns I have to solve.

16:39Like you can imagine kind of for humans, you know, do an IQ test, right? Where you say, here's some specific problems I've created for you to solve. No, no. These are vulnerabilities in the wild that you're trying to find. So it's a completely different level of complexity. And Mythos is just, it's a bar above anything OpenAI can produce right now. And that's also why they've chosen to gatekeep it because it's been too dangerous to release in the wild. So that's in Project Glasswing, isn't it? I think they put 100 million in tokens behind the project to give to all the people testing. So it's quite a chunk of cash to say, come on, here you go.

17:15Fix all your stuff, guys. It's amazing PR. I mean, obviously, right now, I don't know if you have heard, they're talking about Anthropic almost becoming accidentally profitable right now. That's a great phrase. Because they can't acquire enough compute quickly enough. And there has been outages on the platform. Yeah. I mean, it just went so fast. I had a couple this week. They can't get enough compute. And clearly revenue has wrapped so fast that they've swung into accidental profitability at a point where they definitely weren't planning on it. So very interesting dynamic. I also saw that they are actually doubling down in London.

17:56Did you see that? They're taking a massive space somewhere near King's Cross, I believe, like another 12 ,000 square feet. I mean, it's a wonderful thing and kind of shows the role that London has gotten in the AI economy. Clearly, DeepMind has been kind of the anchor here for a long time. But the amount of innovation we're seeing, the quality of companies we're seeing, just what's happening is phenomenal. It's great to see. Do you have an opinion on the Chinese distillation piece? And obviously the OpenAI model now being out, the cyber model being out. It's funny how they also, incidentally, just to raise it, OpenAI have labeled that model cyber, whereas Mythos is a general model.

18:40It just happens to be really good at the cyber piece. So I don't know if there's anything in that. But what about this distillation piece? Will this give the Americans a bit more of a longer stretch away from the Chinese open source? or is it incidental? Well, you know, the world will only have it if the world has the models. Right now, the world doesn't really have the models. It's very few labs that have access to the model. You could look at it two ways. You could say, we've got a global internet, we're running on lots of open source software, there are lots of vulnerabilities there, likably still.

19:12And if they're not patched, then nefarious actors could use that to gain access and, you know, whatever we think about, China is a trade, it's a rival, it's also a trading partner, right? We've got lots of relationships. But would you want somebody that is a fierce rival to have access to basically all your systems? And you're probably going to try and contain that. The other side, the slightly more nefarious view is we've got the tech right now. We as in the West has got the tech. China doesn't. What are we doing? What are we doing? Yeah. Yeah, how is that going to be used in the next six months while this stuff is not widely released?

19:52So you could kind of look at it, glass half full, glass half empty. It reminds me of that Pete Show sketch where two guys, I can't remember their names, they're in the Nazi uniforms and one turns to the other and says, are we the baddies? It's like, I remember my history teacher was telling us, obviously, all these kind of conflicts throughout the world. And he was like, who are the terrorists, Dan? Who are the terrorists? Maybe we're doing nefarious things behind the scenes against the Chinese. Who knows? Let's talk about Sam Altman's prophecy. Maybe it has been fulfilled. Have we seen a one person unicorn?

20:30So I'm sure many people have seen the story of Matthew Gallagher from Medvi in the New York Times or other press. Now, the business Medvi sells GLP-1 weight loss supplement. So this business being able to grow this fast is of no great surprise with peptides doing what they do and GLP-1 absolutely blowing it out of the water with regards to everybody wanting to lose weight and yada, yada, yada. He has allegedly, and I'm going to come on to why I'm going to use these slightly mealy-mouthed words in a second, he has allegedly built a business from nothing to 400 million in a year and by his own words on track for 1.8 billion this year.

21:12So why am I using cagey words like maybe allegedly? Well, the whole thing looks a little bit sketchy, a little bit shady, and I don't think I'm being a total cynic. Some nasty marketing staff, several class action lawsuits. If you look behind some of the numbers and the reviews on Trustpilot, the traffic and the numbers don't quite add up, but it's all great PR. However, I don't want to talk about the shadiness. People can go and do their own research, make up their own mind. What I wanted to talk to you about, Mads, was are we now at the threshold of a single person unicorn? Are we there? Will we ever get there?

21:52Will we ever see a single person unicorn? And maybe, Mads, you can go through some of the things that Gallagher said he's done and how Medvi was built, and we can have a discussion about what we think is real and not so real. Yeah. I mean, look, he sort of said, you know, I've spent a couple of months, about$20 ,000, some chat GPT, some Claude, some Grok, some Mid Journey, some Runway. And with that, over a couple of months, I built this new platform. And as you said, he has claimed that it ramped very quickly to hundreds of millions of dollars in revenue. He's also claimed net margins in excess of 15%, which just sounds incredible at that level.

22:352026 projected at$1.8 billion with just two employees. It's him and his brother, right? Yeah, so that's$150 million a month with two guys. Just buy a yacht, dude. You don't need to do anything. Just buy a yacht. You've done it. You won. Well, the question is how durable it is. I'd say distribution is real business, right? Costco, Walgreens, Amazon, they're all distribution companies. The question is not so much the category. I think what's really interesting here is, however you look at it, that you've been able to create and underpin so much economic activity with just a couple of guys and so much automation.

23:14It's clearly heralding what's to come. How much of the value does Gallagher actually keep? I don't think we really know. 16.2 % sounds high. I don't know if I really believe that number. But even if it's 10 % or 5%, it's still a very, very attractive business for him, of course. What changes in the world where you have people able to do that? I think we're probably still not going to fund two people distribution teams. I think I'm not sure it's venture scale at exit and it's really defensible. But I think there is a real lesson for every founder about just how much you can do and how radically different you need to think about how you build companies.

23:55Smaller teams accomplishing much, much more. You know, pharaonic ambition in terms of what you can take on. It used to be you would think, oh, I need to accomplish something big. I need to raise so much money and hire hundreds of people. And it's just not the case anymore. And that's really, really interesting. I think it's really liberating for us as humanity in a phenomenal way. So I'm actually quite excited by it, the specifics of this business put to one side. It does throw some questions for the investment community. And I was speaking to a couple of buddies yesterday about what we know that a lot of the VC dollars, for example, are now going to have to be put into the marketing and the sales and the go to market column and that brand building piece.

24:39Possibly one of the only moats left is that trust and that brand piece. But it also does throw up that people won't need to raise. There will be many more niches, many more speciality businesses that can survive because very small teams have got the best lawyer, the best consultant, the best analyst, the best data scientist, the best coder in their back pockets. And I'm always intrigued as to how that is going to play out, not only in the venture capital world, but maybe down through angel investing and smaller businesses and how that's going to create new economies. There's a lot of excitement to be seen, I have no doubts.

25:20I think you're going to see a barbell, Dan. I think you are going to see these two men, the three men, three women teams that are doing phenomenal things with limited resource and just able to bootstrap. I do think if you're trying to do something big technically going beyond just setting up a distribution business that runs on the rails of other people, are you actually operating in a frontier doing something new, having more money, having more resource will help you. And so, yes, there's going to be a team here that tries to go it alone, two, three, four people, but they're going to be competing with another team that raises some capital and has 30, 40, 50 people.

25:56And all ill sequel, that team will be able to get more done. and so we'll have a better chance of winning. So I don't think you're going to see a complete, nobody will ever need to raise money anymore, but I do think it enables us to go after different things. And we've seen it. The level of ambition we see from teams today is at a completely different scale. Yeah, yeah. I mean, AI has awakened the world in so many positive ways. There's lots of fear, anxiety, and doom and gloom in the headlines because that's what headlines do. but there is a whole world of ambition and drive and excitement I think is uh that it is countering those forces let's talk about compute scarcity we've seen a few stories this week in the press around the lack of compute we've seen some sml stories and tsmc stories we saw the stargate N-Scale Microsoft run-in.

26:55Tom Tongos wrote a piece about compute scarcity market dynamics, specifically what instigated this segment. So we are at a very interesting moment in time where everybody's looking for those canaries in the coal mine, for those indicators as to how AI is doing, how much compute we need, who's going to make the money. I was looking at the TSMC capex spend and they're still spending so to me that's the best lie detector if the people making the chips are spending lots of money to build more chips then that's got to be that's got to be a good test but mads what would you pick up on specifically on this topic where would you start compute scarcity and market dynamics i think you're bang on you know people are looking for signs everywhere that this so-called bubble is going to burst and things are going to collapse And so people were looking at OpenAI's cut in planned infrastructure investment from$1.4 trillion down to$600 billion and saying, oh, gosh, this is mean that the demand really isn't there.