Self-training AI and the Rise of the Application Layer

Source: Startup Insider – Original Episode (episode in German)

Recursive Superintelligence raises $650 million at a $4.6 billion valuation—without a product on the market. The thesis: AI improves itself, while humans become a hindrance in the loop. At the same time, the application layer is pushing into the market with full force—Anthropic is building integrations for small and medium-sized businesses, OpenAI is closing in on the major consultancies, and Microsoft and Apple are finding themselves in an unusually vulnerable position. Björn Rieckhoff and Jan Thomas analyze the pre-product mega-funding rounds in Europe and the U.S., discuss the growing dependence on AI tools and the price decline in the consumer segment, and take a look at strategic M&A activity in the AI market—from the Stainless deal to the long-term bet that Apple will eventually acquire OpenAI.

Key themes in this episode

Transcript

This transcript has been edited for readability. The content and statements have not been changed. The original conversation was in German; this is an English translation.

Jan Thomas: Björn, hi! Nice to meet you. It’s great to have you here in the studio today.

Björn Rieckhoff: In person for the first time.

Jan Thomas: These are crazy times—times that justify us sitting across from each other. And yeah, it’s pretty wild; I don’t know how you see it. There are two things going on right now: you get the feeling that a lot is happening in the pre-product phase—huge funding rounds—and they’re happening in Europe, too.

Björn Rieckhoff: Yeah, there’s a crazy amount of momentum right now. Pre-Product is the topic I brought up today: Recursive Superintelligence was actually announced a few days ago. I always have this thankless task on Tuesdays of having to bring it up.

Jan Thomas: But that sounds like you’re not happy with your topic. I think it’s a really exciting topic.

Björn Rieckhoff: It’s a really exciting topic. Recursive Superintelligence has raised $650 million. At a $4.6 billion valuation—without having a product on the market. Or rather, they probably already have quite a bit under the table. They do want to go live this year, after all. But outwardly, that’s not public yet. Google Ventures has invested, Radical has invested, and Nvidia is on board, as is usually the case.

Jan Thomas: I don’t think you can do without that. Yeah, yeah, yeah.

Björn Rieckhoff: A chip manufacturer is also on board. That’s a clear signal in itself.

Jan Thomas: Okay. But I think the team—we’ve seen this a few times in the past—DeepMind seems to be becoming a breeding ground, I believe, for the new super-startups.

Björn Rieckhoff: It’s crazy when you look at the founding team’s résumés: they’ve got it all. Research Director at Meta, a professor from UCL London, Chief Scientist at Salesforce—it’s insane how much brainpower is packed into that group. And I always find that so impressive because it’s not about the founders themselves standing at the whiteboard running the day-to-day operations. It’s more about the second tier—but the foundation that this second tier can build upon is simply enormous.

---

Jan Thomas: We talked about this here recently—there was that other crazy funding round. I can’t quite remember the name right now—I think it was Inflection AI. A London-based company that raised a 1.1 billion round, also pre-product. I think the vision is always clear. Yann LeCun as well. You get the feeling that people are betting on the person and on the vision. But now DeepMind—many people know about that too, and I just want to briefly mention it again: watch the Netflix documentary or on YouTube—I think there’s one about AlphaGo. Super exciting. It’s fun to watch how, for the very first time, an AI won a game and defeated the world champion. I’ll just spoil it now, since it’s been eight years. It defeated the world champion in a game where people thought it was impossible for a machine to ever outperform a human, because there are so many infinite moves. And that’s just DeepMind. And now, one by one, individual startups are emerging. Probably not the last one today.

Björn Rieckhoff: Probably not. The approach is very impressive. Briefly on what Recursive is building: Ultimately, they’re banking on AI improving itself—without human intervention. The vision behind this is that the “human in the loop” is really just a hindrance. And if you pit two models against each other, they’ll challenge each other so intensely that the result will simply be a better model. That’s essentially the reasoning behind it. Systems analyze their own weaknesses, redesign themselves, then test themselves again, and improve over time. Of course, that’s also a very bold claim. From a German and European perspective—and you always have to mention this—there need to be limits somewhere. They can’t just run wild to the point where you no longer know what will happen. But that’s always something for others to worry about.

Jan Thomas: Yeah, exactly, let others worry about it. I’d say I get a bit uneasy about a lot of things. Maybe I’ve quoted this here before: Recently, the first robot developed entirely by AI was unveiled. And when you look at that thing—it’s something that no human could have come up with. And I find that so spooky. We’re just entering a world like that right now. And also what Inflection is doing—Olaf Jacobi analyzed this recently and said: So far, all LLMs are essentially based on human knowledge. The new generation of world models is being trained on things that aren’t based on human knowledge. I think that’s why this uneasy or spooky feeling is actually pretty intense. We also had “Emergent World” as the top story in our newsletter. They set LLMs loose on each other, and they started forming societies. That’s a super exciting point, too. I don’t think that’s entirely accurate—they didn’t set different ones against each other; rather, Anthropic has its own society, Gemini has its own, and so on with Grok and others. And then they analyzed what emerged from that. How many virtual people fell by the wayside, how many were eliminated. All the way to Armageddon—they ran through every scenario. It’s pretty wild. That’s why I get this uneasy feeling. Others are welcome to handle it, but it’s definitely important to check in every now and then to see what’s going on.

Björn Rieckhoff: Yeah, yeah, yeah.

Björn Rieckhoff: To simplify a very complex issue and make it more tangible: I have this feeling that I’m actually off the mark here. And the way products are designed—a robot is a good example of that.

Jan Thomas: One that really scares you.

Björn Rieckhoff: We always have this human perspective in mind; we have thoughts, we have standards of quality within which we operate, and within which we then also develop the machine. And the machine doesn’t care about that. Suddenly, a joint configuration pops up on the screen that you’ve never seen in your life. And you’re surprised that it can even work, because you aren’t building on those preconceptions. That’s actually the interesting part. I’m curious to see what comes out of it in the end. The vision is so grand. The founders regularly talk about curing cancer, fighting viruses, etc.

Jan Thomas: Exactly, saving humanity. That’s one part of it.

Björn Rieckhoff: You’re right, of course; it could also go the other way.

---

Jan Thomas: And we’re all familiar with the debate here involving the Pentagon and Dario Amodei. I find this so fascinating: People are paying much more attention to the founders, to the founder types, and trying to figure out who holds which values and who actually stands for what. Elon Musk used to be a poster-boy entrepreneur; now he’s persona non grata in certain circles. Sam Altman jumped right in when Anthropic launched. Or Peter Thiel. You start to develop a certain image of people because it’s becoming so important right now. We just had Helsing here—they’re probably going to raise a massive funding round; I don’t think it’s finalized yet, but they’re the most valuable startup in Germany. You’d really like to know: Who are the founders, and how close are they actually to our values? Are they more in the “we want to advance humanity” camp, or more like Palmer Luckey with Anduril and his Hawaiian shirt, where I keep thinking, that’s a pretty good costume to maybe mislead us a bit. So you know what I mean?

Björn Rieckhoff: Yes.

Jan Thomas: I’d like to disagree there, Björn, because I think you’re coming from a VC background—we shouldn’t forget that. Exactly. And I think you’ve also got quite a few—I’m relying on them a bit—of these impact-driven, purpose-driven founders who say, “I get up in the morning because I want to fight world hunger” and stuff like that. You have those too, the ideological ones. But they’re maybe a bit removed from VC, right? I just don’t want us to paint a bleak picture here—I think there’s a mix of both, and there’s still hope. We don’t know much about the founders of Recursive yet, but maybe let’s get back to the logic. I find it pretty mind-blowing that an AI is actually starting to train itself. In my mind’s eye, I just see some kind of exponential curve. And you might know that a few years ago at Meta, they eventually shut down an internal AI system because they could no longer understand what it was saying. It had started communicating among itself, and it was no longer comprehensible from the outside. That’s kind of my concern—that this might happen here at some point, too.

Björn Rieckhoff: I actually believe that these kinds of issues are very often present behind the scenes. We’re hearing about it directly now at Recursive during the funding round. But in reality, it’ll happen at Anthropic, and it’ll happen at OpenAI. That’s taking root in people’s minds, and the question is being asked very strongly: what do we release, what don’t we release, where do we draw the line? I don’t know if that has reached the Americans yet, to the point where they say: Free-for-all. The one who’s fastest will win. Where we’ll end up—I have a bit of hope that it will regulate itself.

---

Jan Thomas: Yeah, I’ve been—that’s why I’m actually pretty excited that so much is happening in Europe right now. These are all such promising developments. On the one hand, seeing how much capital is being allocated there, but also that it’s coming from Europe at all. All three of the ones we just mentioned are major European funding rounds. Because I’m thinking ahead about these dependencies—when you really talk about AI, and maybe, as you’ve noticed, we also wanted to briefly touch on what’s happening at Anthropic. I get the feeling they have a roadmap; I’ve never seen a company release society- or market-changing features at such a rapid pace. That’s totally mind-blowing. And not being able to keep up with that—I’m totally dependent on these tools. Even now. Pip Klöckner said recently that if Anthropic decided overnight that they now cost 500 euros a month, people would probably still pay it. Because you’ve gotten into that kind of dependency by now. It’s become such an addictive behavior by now. I think that’s almost worse than those early Facebook moments when you had that feeling of, wow, something new—I have to endlessly scroll through this now. Here, because you’re in “build mode,” it’s even cooler. But the dependencies are so strong.

Björn Rieckhoff: And I find it extreme because even if you’re informed and knowledgeable—whether as a private individual or as a CEO—you can’t keep up. When you do your research, you have a certain expectation of what you’re ultimately building. And a week later, your announcement goes online and is already outdated.

Jan Thomas: That’s crazy, isn’t it? You’re constantly on the go, right?

Björn Rieckhoff: Yes.

Jan Thomas: You still look great, though.

Björn Rieckhoff: No, but that’s definitely a dynamic to keep in mind. To take the story a bit further: Of course, you still have this dynamic—recursive superintelligence, where the strongest model will win out in the end. That’s also important; it would be the vision behind the funding round. At the same time, we’re seeing that the models are becoming almost interchangeable because they’ve reached a point where they can handle everyday tasks very well. It wouldn’t really matter whether I work with Claude AI or ChatGPT. Whether Driver A or Driver B is driving—the result will be more or less the same. And naturally, this also leads to a certain drop in prices. Just in the consumer sector with subscriptions, you can see a sharp price drop from last year to this year. I took a look—60, 70 percent. I believe Google has lowered prices by about 70 percent from last year to this year. Which simply speaks to the fact that competition is so fierce and it’s difficult to differentiate based on product quality.

Jan Thomas: I hadn’t noticed that. I kind of get the impression that I’ve never spent this much money on tokens before, especially with Anthropic. Because I feel like with all these new features—first there’s Opus, then there’s Claude Design, and stuff like that—everything just eats up tokens all the time. But you also want to try everything out. I’ve had phases where I was just buying more the whole time. That was really bad.

Björn Rieckhoff: Yes, but now I’m asking you as a power user.

Jan Thomas: Yeah, yeah, yeah. Of course, yeah!

Björn Rieckhoff: And then there are the tokens to get out of that. I’m talking more about the end consumer right now. But yeah, I’m with you.

Jan Thomas: But far too many people still aren’t paying. I think that’s the problem. If you look at ChatGPT right now—I think only five or ten percent, something in that range, are actually paying. And a lot of people only log in once, twice, or three times a week. That’s the complete opposite of a power user. Not even casual users, but people who know it exists, and if I ever want to know whether I can still drink milk after the expiration date, I’ll ask ChatGPT. But even that is cool. It’s just that, of course, all these companies want to collect a lot of data. User behavior is important for building stickiness. And these are extremely expensive models. We’ll probably end up with everything that’s currently being planned for data centers. At some point, all of that will need to be translated into some kind of meaningful equation.

---

Björn Rieckhoff: In one of the last episodes, we discussed just how large Big Tech’s CapEx is. If you keep that in mind, those will likely be huge sums in three years. At the same time: What you’re bringing up—if you’re not able to differentiate based on models, you have to ensure you secure access to customers. And that brings us to the second part of this topic. On the one hand, you still have these foundational models that are being heavily funded. On the other hand, you’ve seen some very strong announcements in recent days, whether from OpenAI or Anthropic, which suggest that right now the leading players in the market are pushing very hard toward the application layer to ensure end-user access. I’m thinking specifically of Anthropic’s announcement a few days ago: “Claude for Small Businesses.” There they said: A large portion of U.S. GDP is generated by small and medium-sized businesses, and they’ve now built a platform offering all kinds of integrations in a single package—QuickBooks, PayPal, HubSpot, and the like. These are tools that are already widely used within SMEs. And they’re now integrating much more deeply with Claude, which is essentially what everyone knows from PowerPoint, Excel, and the like. It works incredibly well and boosts productivity. You can see this strong alignment right now to tap into a new customer group that wasn’t fully there before. And there are so many indicators of this. I think OpenAI is starting to buy up IT consultancies. The big announcements with partners, with the top consultancies out there—the TCS of the world—which are also incredibly revenue-dependent. Simply to ensure access to these companies. That’s, of course, a completely different play than what I’m working on in my basic model. But it’s all the more important for long-term monetization and for the prospective—yes, hopefully someday—profitability of these businesses.

Jan Thomas: I think this is an interesting question for end consumers as well—but the real question is about these dependencies down the line. Are you essentially training a system right now that, through more and more training, ensures that you become increasingly dependent on it? That’s kind of the logic behind Anthropic. That’s why the question—going back to pricing—makes total sense, of course, that you might even offer it for free. There’s that old saying: If it’s free, then you’re the product. And maybe that’s a bit the case here too, where you’re saying: We’re now taking the knowledge from the McKinseys of this world—no matter what industry—to get even better, so that eventually we can achieve maximum prices again in a hockey-stick fashion. And I do believe that you’ll see more and more specializations with these tools. I think that at some point—just like a media group—you might say that Anthropic is better in the SME sector or the retail sector, and then you’ll have something like OpenAI, which might be more at home with consultants. But I think that at some point, maybe one or two others will start to enter the market, and the markets will split up a bit. And that the fees charged for these services will become really expensive—I mean, really, really expensive.

Björn Rieckhoff: Yeah, that could very well be the case. There’s also the question of using the individual models on proprietary data. That’s also part of the gamble many investors take—who are, at most, providing a bit more funding—when they say: There’s value in the information that’s currently being utilized. That’s exactly the logic. It’s certainly intriguing when you look at it that way. If you’re working with the integration or with Co-Work, all the data is protected and the models aren’t trained on it. That’s interesting too. These are policies where you can definitely adjust the parameters: how aggressively you play the game, and what you do with the data.

---

Jan Thomas: Yeah, I think looking ahead—I’ve told you this before—I’m a bit skeptical about Microsoft, too. So maybe you could say—I just checked again—Microsoft is currently worth 3.15 trillion on the stock market. What do you think they’ll be worth in two years? I would have said half. Maybe not half, but I said the other day that if I could, I’d go long-short on them. Because I do believe that Microsoft is actually the tool of choice—the tool stack of choice—in the very area we’re discussing right now. You either have the Google ecosystem or the Microsoft ecosystem. And it’s an operating system, but if you look at Word or Excel, those are terribly dumb programs. They have no memory. And I think that now, suddenly, with Anthropic and OpenAI, a new dimension is being added: memory. The tool understands you. It understands what you typed into Excel just a week ago. It remembers and asks, “Should we just pick up where we left off? Should we adapt the same thing again?” And because Microsoft can’t do that, Microsoft will eventually just be a little window to the customer. Like a frontend. But the backend, I believe, will be provided by whoever ends up holding the data. And it wouldn’t surprise me at all—at OpenAI, they’ve been saying all along that one or more devices are coming. I can even imagine that Anthropic will eventually—I mean, why wouldn’t they? You’re also getting Apple’s Siri on Apple—that’s, I think, the biggest blunder in AI history. Nothing happens there. It’s a bit like Alexa’s first steps. That’s not AI; it’s more of an early showcase. My feeling is: Why would Anthropic or OpenAI ever say, “I’ll just use Apple’s or Microsoft’s dumb operating system”—we’ll just quickly build our own. And that would be my—well, not in two years, two years is an exaggeration now—but I think, looking ahead, Microsoft has absolutely nothing to counter that with.

Björn Rieckhoff: I think that would also be an argument, so to speak, that works for Microsoft, just as it does for Apple. It doesn’t have its own foundation at all, but simply takes what’s already there...

Jan Thomas: They tried that with Siri. Siri was the attempt, but they completely failed in that area.

Björn Rieckhoff: Yeah, but in hindsight, it was a very deliberate strategy to say: I’m simply not going to play that game, because my expertise lies elsewhere. The value lies—unlike so often with large corporations—in the distribution channel, the sheer number of end devices we have, and access to customers.

Jan Thomas: We thought of BMW, we thought of VW, we thought of Mercedes. And now China is coming. I think it would be exactly the same for me. This is just a theory, so feel free to ignore it. But what I’m saying is: I believe that because you have tools that understand you, users’ dependence on these kinds of tools—that’s my point—won’t just grow; it will become second nature to every user. It’s a bit like it used to be. There used to be this joke where people would walk into a newsstand and say, “Hey, there isn’t even a search function here.” It’s like two eras colliding. And I think the next era is coming, and you’ll just be used to the fact that your tool, your device, understands you completely.

Björn Rieckhoff: Yeah, the way I use certain programs has changed. For example, I use Visual Studio Code for my code. I don’t use the native Claude app on the desktop; instead, I do it through Visual Studio Code because I prefer the interface.

Jan Thomas: But your approach to Excel has probably changed too, right?

Björn Rieckhoff: The way I work with Excel has changed, too. It’s not that I use VS because I’m writing the code. Rather, because I like that the code is visible. I feel like I’m not looking at a black box; I can see what’s happening. It’s the same with Excel. So, you’re right, of course—the way we use these tools is changing. At the same time: I find it really exciting—I actually ask the AI regularly how I should approach things.

Jan Thomas: Mhm, exactly. Recursive logic actually, right?

Björn Rieckhoff: A bit of that Recursive logic, yes. When I start a new product or project, I throw certain things into the chat and notice: “Oh wait, this discussion is actually much better suited to Claude Code, which suggests the tech stack to me.” For other conceptual decisions that are more feature-driven, I go back to the chat. So you’re essentially handing off between the two sides all the time.

Jan Thomas: But that’s still the case, right?

Björn Rieckhoff: Yeah, super interesting that there’s no real bridge yet, isn’t it? You think to yourself: “Why?!”

Jan Thomas: That’s still a bridge technology at the moment. But it’s coming—I’d guess, fast-forward a few years into the future, it has to come. When these data centers and things like that arrive, then memory will probably get even bigger, and so will these devices. At OpenAI, they’re speculating that some kind of daily companion is coming that records everything you do. It’s exciting, after all. Data protection aside—I’d think it’d be pretty cool if, a year from now, I could look back and remember the conversations we had today. Or if I can’t remember which conference I was at or who I spoke with about this or that topic. That’s just how the human mind works, unfortunately—you don’t document everything. Being able to look that up somewhere—that part I find exciting. But that creates a dependency that Microsoft simply can’t accommodate. That’s what I’m getting at. That’s why, with this Microsoft case, I was thinking: What even is Word anymore, please? To me, Word is just ancient history—you type in half of it and then maybe rephrase it a bit. But this “I’ll write you a letter” feature—that’s a bit like Faber-Castell.

Björn Rieckhoff: Yeah, I keep thinking about that paperclip from the ’90s or 2000s—what was it called again? The one down in the bottom right…

Jan Thomas: Clippy, I think it was called.

Björn Rieckhoff: Yes, that’s exactly what Copilot is to me. It’s just hanging around there as a companion. You’re right—the way software is used is going to change dramatically. At the same time: those declared dead live longer—Apple and Microsoft have so many partnerships. They’re positioning themselves pretty well, and I also think Apple is definitely keeping a close eye on things and just has the distribution.

---

Jan Thomas: I have this long-term bet going with Martin Möllmann from HTGF that I actually believe Apple will eventually buy OpenAI. The logic behind it: Because OpenAI will eventually have a funding need that it can no longer meet on the public or private markets. But also the other way around: that Apple will eventually be forced to act. I mean, a few years ago, no one would have thought that a company would come along that could pose a threat to both Google and Apple at the same time. That was out of the question. And now, if another device really does come out—again, this “Memory” thing, that’s what I’ve been getting at the whole time—if OpenAI has a device that understands your entire day and all your conversations: Then Apple won’t stand a chance. At least not with the devices currently available. And that’s perhaps why—we wanted to touch on this briefly again—M&A could become a major area in the future. I don’t know if it will reach that scale, because that’s obviously the biggest gamble of all—that trillion-dollar corporations will suddenly merge or be acquired. But you can already see on a smaller scale what’s happening right now.

Björn Rieckhoff: What you’re describing is very publicly impactful—clear to everyone. And I think it was just announced that Alphabet is Berkshire’s (note: 7th) largest position.

Jan Thomas: Not Amazon anymore. I think they were invested in Amazon, but they’re out now, and now they’re in Alphabet. I find that really exciting, too. But now also after Warren Buffett, right?

Björn Rieckhoff: Yes, after Warren Buffett—for the worse and for the better. We’ll see.

Jan Thomas: Yes, exactly.

Björn Rieckhoff: But as I said, the smaller topics—the supposedly smaller ones—that’s where M&A has already become a battleground. It was just announced that Stainless was acquired. Founded in 2022, the acquisition price was around $300 million. And Stainless makes SDKs for their customers, basically packaging the APIs—yeah, into production-ready SDKs. And they’ve also made these available in the past to OpenAI, Google, Cloudflare, and so on.

Jan Thomas: And that’s quite a bold move, you have to say.

Björn Rieckhoff: So they were acquired, and just like that, the service was shut down. You basically get this message: We’re just turning it off. Crazy, right? Yeah—OpenAI and Google are now just losing access to a tool that helped build their own developer experience. Just think about that! It’s true that they can still use it, but from now on, it’s simply no longer supported by either company. There’s also no workaround anymore. That’s where you see this dependence on certain major players within the entire ecosystem. I find that fascinating. That’s only 300 million. Crazy. But strategic value—I’d say it’s significantly greater than that.

Jan Thomas: Do you think it’ll go through? Or has it already gone through?

Björn Rieckhoff: I don’t know. My understanding was that it was a done deal.

Jan Thomas: Yeah, okay. It’s pretty crazy. I’m surprised—only 300 million. I’m surprised that it was even on the market. That they’d even allow that, you know? Because it’s such a strategically valuable asset.

Björn Rieckhoff: Yes, that’s right. I think you’re also seeing quite a few very small acquihires or similar deals right now. These are all topics of strategic importance. To be honest, I’m much more excited about these levels—a bit of this chess game where I have to anticipate the next pawn move—than I am about necessarily having to look ten years into the future at Microsoft and Apple right now.

Jan Thomas: Nah, that’s just a thought experiment the whole time, right?

Björn Rieckhoff: There’s so much momentum right now—it’s just like the Innovator’s Dilemma — it keeps coming back.

---

Jan Thomas: There are always these nice animated graphics—I’m sure you’re familiar with them—when you look at a timeline: which companies were the most valuable 10, 20, 30 years ago. And then you see how that changes every year and how much they’re worth. And I’m pretty sure that the most valuable companies today won’t be the most valuable companies in ten years. Instead, I think they’ll come from among those mentioned, perhaps even from the European players here. Olaf Jacobi, with Inevitable Systems from London, was convinced: This could become a $10 trillion company. Trillions, that is. A company that will somehow end up at the very top, maybe becoming the most valuable company in the world. Of course, that’s a lot of crystal ball-gazing, too. But what I’m trying to say is, I do believe that right now—look at Anthropic, how old are they? They’re four years old or something. That’s totally crazy. And now SpaceX suddenly comes into the picture, where you don’t know—it’s become this kind of hodgepodge, where you also don’t know where they’ll be in five years. Still—and maybe this ties into your point—I can also imagine that the scene, this whole Microsoft and Apple thing, might be too obvious. Maybe in the end it’ll be rare earth metals or satellite coverage or topics like that that make the difference—things we mere mortals don’t even have on our radar. I don’t know, government ties because of data centers as big as 700 Walmarts or something. Just those kinds of things.

Björn Rieckhoff: How was that recently? Anthropic bought about 200k NVIDIA chips from Elon Musk—I don’t remember the exact magnitude—because they simply can’t keep up with their own growth anymore. These kinds of dependencies are just a given.

Jan Thomas: Yeah, exactly. And those are just the areas people are focusing on right now, or at least in the short term. But still, I believe in the long run—that’s why I mentioned Microsoft and Apple—I think these are companies that are just going to keep coming. That’s probably why the comparison with the automotive industry isn’t so far off. They had a system that worked for ten, twenty, thirty years. And I think that’s being called into question now. And just the same, Anthropic and OpenAI could be called into question again in five years because someone else will come along. Yann LeCun is always running around saying, “They’re all beginners.” Others say it’s about energy, right?

Björn Rieckhoff: At some point, we’ll probably just have to ask ourselves whether we want the problem of prices keeping us afloat.

Jan Thomas: We can almost take that as our closing remark, right?

Björn Rieckhoff: Yes.

Jan Thomas: Yeah, thanks. Super exciting discussion as always, right?

Björn Rieckhoff: That was fun!

Jan Thomas: Exactly, until next time. Ciao!

---

About Björn Rieckhoff

Björn Rieckhoff is an independent advisor and business angel with nearly ten years of experience in early-stage venture capital. He helped build Cavalry Ventures as its first employee and later became a partner of the fund. Today he supports founders more directly with fundraising — sharpening their story, stress-testing business models, and setting up lean financing processes. With over 80 transactions and board seats from seed to Series B, he brings this perspective as a sparring partner for entrepreneurs.

About Startup Insider

Startup Insider is the industry portal for the startup scene in the DACH region. It covers news from all regions and industries, along with an overview of key players and events in the German-speaking startup world.

Links

• Stainless
• Anthropic
• DeepMind
• Kandou & Kestra – AI-Infrastruktur reift (vorherige Episode)