Kandou AI and Kestra: Where the next layer of infrastructure is emerging
Source: Startup Insider – Original Episode (episode in German)
Two funding rounds that show just how deep the AI infrastructure market has become: Kandou AI from Switzerland raises $225 million in a Series A round for chip-to-chip interconnects—the physical connection between GPUs and memory that's become a bottleneck. And Kestra, an open-source orchestration platform, secures $25 million for coordinating complex AI workflows in enterprise environments. Björn Rieckhoff and Jan Thomas break down both funding rounds: why strategic investors like Synopsys and Cadence send a strong signal, what it means when a funding round doubles as a marketing tool, and how the job description is currently shifting from CTO to CDO.
Key themes in this episode
- Kandou AI – $225M Series A: Chip-to-chip interconnects as a bottleneck in GPU clusters. Copper vs. optical connections and why Synopsys and Cadence as strategic investors send a strong signal.
- Funding Rounds as a Marketing Tool: Kandou has already raised over $500 million and is still labeling this round as a Series A—what’s behind this decision and why the label is becoming less and less meaningful.
- Kestra – Open-Source Orchestration: $25 million Series A for coordinating AI workflows, data pipelines, and infrastructure automation in enterprise environments. Over 25,000 GitHub stars and 30,000 organizations.
- Picks and Shovels in the AI Gold Rush: As the AI market matures, it's giving rise to increasingly specialized infrastructure providers—on both the hardware and software sides.
- Changing Job Roles: From domain expert to workflow strategist—how AI is fundamentally shifting the demands on CTOs, CDOs, and startup teams.
- Investor Intelligence: Why the traditional VC list is obsolete for founders and how data-driven investor matching improves the fundraising process.
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. I hope you had a nice Easter.
Björn Rieckhoff: I had a wonderful Easter. Sunny, but extremely windy. I was at the Baltic Sea with my family and was almost blown away.
Jan Thomas: Yeah, these are just turbulent times right now. You could say that in abstract terms, I guess. We last saw each other at Delta Campus. You just told me in our preliminary chat that you’re there every now and then. So, if anyone wants to meet you in person, they should probably subscribe to the calendar there somehow. Julian Teicke does that really well, I have to say. Always great topics.
Björn Rieckhoff: Great topics, great people. I get the impression that a fantastic community is forming. Berlin auf die Eins, but Delta Campus itself is also really special. I really like the focus, and of course, in the context of the startup scene—potential angel investments, possible support for funding rounds—it’s always very exciting for me to meet these people there.
Jan Thomas: I didn’t even understand the acronym at first, but I actually think BADE is way cooler than Berlin auf die Eins. But as I said, it’s a great initiative. I’d say let’s dive right in. We’ll pick up a bit where we left off last time. We’d already touched on the big topics surrounding the Anthropics of this world. And today, I wouldn’t say we’re picking up right where we left off, but you can definitely tell what you’re passionate about.
Björn Rieckhoff: Yeah, I find it really exciting right now to look at all these business models and topics that go hand in hand with the growing maturity of the whole AI space. And not just looking at the model level, the agent level, and the application level, but also taking a closer look at what’s actually behind it all—what kinds of infrastructure models exist on either side of it. These are often referred to as “picks and shovels,” which basically ride the wave of the gold rush. And these are often critical infrastructure projects. And we had—you mentioned it—Nscale last time; that was a two-billion-dollar funding round in early March. We’re not quite there yet with today’s topics, but we’re still heading in that direction. First topic: $225 million—that’s not too shabby for a Series A. There’s just a lot of money involved, and naturally, there’s a lot of hype surrounding these kinds of topics.
Jan Thomas: Do you get the feeling that new infrastructure layers are somehow emerging right now? You just mentioned “picks and shovels.” I mean, that’s kind of the pitch every startup ends up making—saying, “We’re actually hoping to be the shovel sellers in the Gold Rush.” But very few actually manage to pull that off. You suddenly get the feeling here that, because the deck is constantly being reshuffled, new shovel sellers are popping up all the time.
Björn Rieckhoff: That’s true to some extent, but I think it simply goes hand in hand with increasing maturity. I believe there’s a growing, deeper understanding of where friction points arise throughout the entire value chain. There are increasingly specialized providers who are targeting precisely these friction points and solving them. And then you often realize that the markets they’re addressing aren’t actually that small after all. And this gives rise to very, very valid, very good, and profitable business models. I think it’s simply this accompanying maturation process that’s now beginning, where many—not just investors, but also founders—are taking a look at what actually exists to the left and right of the major models and applications to position themselves well here. And I think there’s still a lot of greenfield out there.
Jan Thomas: You just mentioned that we’re going to start by talking about the Series A round—225 million from Switzerland. I didn’t realize at first that the funding came from Switzerland; that’s quite remarkable. But the company—I’m referring again to this Series A round—was actually founded back in 2011.
Björn Rieckhoff: It’s called Kandou AI. And the recent announcement of the Series A funding round, 225 million, is also a relatively large Series A. As you say, it comes about 15 years after the company was founded, which is relatively late. I think it’s probably a matter of the window of opportunity being absolutely there right now. And right now, the positioning fits very, very well. It was founded at EPFL and essentially handles the transfer of data between chips—that is, the GPUs and memory—which has increasingly become a bottleneck in recent months. Let me try to break down chip-to-chip interconnects for the listener. In modern AI clusters, you no longer have a single computing unit; instead, there are dozens or even hundreds of GPUs. These are actually connected to each other via physical lines and linked to the memory. And the problem you have is that in recent years, GPUs have become dramatically faster and capable of processing more data, but the connections haven’t gone through the same innovation cycle at all. And the result is that you now have relatively expensive GPUs that can move data very, very quickly and are essentially waiting for more data to come through the lines so they can continue working. Kandou AI is essentially developing technology to send more data per second through the lines. And they’re doing this specifically with copper lines. In early March, there was actually a company in the U.S., Ayar Labs, which Nvidia also invested heavily in—a Series E round, I believe, with $500 million in funding. They do this with optical cables. It’s the same distinction you probably know from home: are you still on copper or do you have fiber? That same difference exists inside these clusters. And Kandou, with its approach to making copper connections significantly more efficient, is right at the forefront of this.
Jan Thomas: And I want to come back to this Series A, because I looked it up on Crunchbase and I’ve never seen anything like it before. As I said, the company was founded in 2011, but has already completed several rounds, some of which were labeled Series A. So Series A, B, C, D—it’s all there—and now Series A again. At least according to Crunchbase and the articles we’ve read. I’ve been back and forth researching this with Perplexity and ChatGPT. There was this theory that it could be a strategic realignment and sort of a fresh start, because this round is just massive. But Bessemer Ventures, for example, was already involved before—I think they’ve been on board since the seed round. Accordingly, the company has a pretty impressive reputation and has already raised over 500 million in total.
Björn Rieckhoff: Yeah, it’s really exciting, and I think you can definitely see these two dynamics at play. Of course, it could be related to what you just described—this realignment or perhaps a refinement of positioning. But I’m also seeing this a lot myself right now: I think the name of the funding round has become more of a marketing tool than ever before. Yes, and it’s often about wanting to have a sort of typical large funding round in each phase.
Jan Thomas: You’d really have to stay in the seed round, right?
Björn Rieckhoff: Is that the 50 million pre-seed?
Jan Thomas: Yeah, exactly. That’s not such a bad idea, is it?
Björn Rieckhoff: Yeah, but it does send a signal, of course—you shouldn’t take it at face value, and it makes sense to look at the background. At the same time, what ultimately matters is the outcome and what you actually achieve with that capital. Whether it’s called Series A, B, or C doesn’t really matter. Of course, it’s important to check afterward, from a corporate law perspective, whether the shares issued still have seniority over those that were previously in place. Or does this also have to do with the fact that, in a worst-case scenario, the Series D investor you mentioned—Bessemer or the like—is now actually in a worse position than SoftBank or Maverick Silicon, which has just come on board?
Jan Thomas: Totally. So, that was the other suspicion. It’s kind of a marketing Series A. That they’re basically trying to label it differently. Unfortunately, since it’s a Swiss company, you can’t really look into the commercial register. That’s why you can’t find out these very specific details about the share distribution. But let’s stick to the market. I mean, you can see from the size of the round that there’s apparently a lot of hype involved. Does that surprise you?
Björn Rieckhoff: That doesn’t surprise me. When I think about it, there are two indicators: You just mentioned the cap table, and what we see there is that Synopsys and Cadence Design Systems are already invested. As I said, I’m not really from the industry myself, but both are essentially operational companies in exactly that market. Ultimately, they aren’t traditional VC investors, but rather strategic partners—and, as a result, direct competitors in a way. Which is already a very strong signal that this technology is of central importance going forward. You could almost say: if two major competitors both need to be at the table, it’s indispensable for the market. And whoever doesn’t have a stake in it risks losing influence in all the next-generation cluster designs. It’s a real quality signal. The team likely played them off against each other very well to get both on board — and to secure their own position without becoming dependent on either one. I find that very interesting because in later stages, the deliberate choice is often to not bring any strategics on board at all. But if you bring two, you actually gain more leverage for the next phase.
Jan Thomas: But when it comes to exit channels, would you say they’re also shooting themselves in the foot, because you could argue that in a round of that size, a pre-exit strategy is already sort of implied? If there are two competing companies on the cap table, maybe they’re blocking each other, right?
Björn Rieckhoff: Yeah, or as you say, maybe an exit to the relevant companies isn’t even planned at all—instead, the goal is more about independence, perhaps continuing in some kind of joint venture, maybe with other monetization methods, such as not just a traditional sale of the product, but perhaps licensing the technology, and so on. So there are several possibilities here, regardless of whether an exit actually happens. But I do agree with you—that’s obviously not the most straightforward path.
Jan Thomas: And I asked you if that surprised you, because I somehow didn’t see it coming—after all the GPUs, then we had the LLMs, then energy was the big topic, and now the next dimension is basically being added, and somehow this whole universe is suddenly becoming so complex.
Björn Rieckhoff: It will. We already touched on this last time at Nscale—everything surrounding the data centers themselves, the orchestration of the entire infrastructure, is becoming very, very relevant. And then you basically keep digging deeper; you end up looking at every single data flow, every environmental factor necessary for the technology—or rather, every necessary environmental factor that you can optimize or improve to achieve better performance. Higher productivity at the end of the day. So it’s incredibly layered. I think things will sort themselves out a bit when it comes down to who actually gets how much of the value creation. But we also touched on this in the last episode—these absurd volumes that large tech companies are currently feeding into data centers. Of course, there’s a lot of room there to somehow optimize key friction points in the data flow to make the whole thing more efficient.
Jan Thomas: Let’s move on to the next topic. That’s a bit easier for me because, for example, Kandou didn’t have any client logos on its website. So it’s a bit hard to figure out what kind of company it actually is and who its client groups are. Now, with the next company, Kestra, they at least have some big, very diverse names: Xiaomi, Toyota, Fila, Apple, JP Morgan, Bloomberg, and T-Systems. I think that’s an interesting range.
Björn Rieckhoff: That’s an interesting range. And it’s also interesting, of course, to consider what “customer logos” mean in this context. Because that’s also exciting—Kestra is an open-source platform. So Kestra, to give a quick background, has just raised a $25 million Series A funding round, led by RTP Global, but investors like Alven were also involved and brought the total funding to $36 million, I believe. And Kestra solves—it’s also infrastructure in a way, though not on the hardware side, but rather on the software side. They solve a coordination problem. So running AI in real enterprise environments doesn’t just mean calling the large language model; you have hundreds of dependencies to manage and orchestrate. There’s no standard for that yet. And Kestra has essentially built an open-source orchestration platform—a kind of control center for AI workflows, data pipelines, infrastructure automation, and so on. To provide individual companies—and as you say, the logos—which are all essentially enterprise clients, with a uniform standard. And the fact that it’s all open source is often a very, very strong indication that it emerged specifically from a practical problem and is being adapted by the developers at the individual companies themselves, without every company you see on the website necessarily paying for it yet. It’s more about active usage. I believe the solution has over 25,000 GitHub stars and is already being used quite extensively. They say they’ve already integrated 30,000 organizations. That’s impressive.
Jan Thomas: I find it interesting because, when I was trying to make sense of this for myself, I wondered if this might eventually become something that people actually build themselves. I mean, the more mature AI becomes, the easier something like this might be. At the same time, though, I also thought that companies like Anthropic or OpenAI are probably eager to buy something like this. It has a bit of an agency feel to it, I think. I don’t know how you see it.
Björn Rieckhoff: Yeah, I know what you mean. I think the use cases—when you’re looking at companies like JP Morgan Chase, Bloomberg, or Crédit Agricole—often become too complex for you to just build them yourself, so to speak. I think it really comes down to hardcore data transformation and infrastructure. That’s why I picked out these two topics—because Kandou is more on the hardware side of infrastructure, and Kestra could almost be described as infrastructure on the software side. I think it’s primarily intended for engineers who run truly complex, data-intensive workflows. And that’s no small feat. It’s really about very, very large-scale operations. Honestly, I’ve been comparing it in my head to something like n8n—an automation tool that’s essentially designed with a “no-code, low-code first” approach, allowing you to connect APIs and automate workflows almost visually, in a Zapier-like way. And with Kestra, you’re in a different realm—it’s really about building hundreds to thousands of workflow executions daily. That gets complex accordingly. You actually have an environment there that appeals more to software developers than to business users, and the focus is more on enterprise requirements. It’s no longer about me cobbling together my own tool, as I used to do in the past.
Jan Thomas: But don’t you think that—end-to-end, I think the comparison is spot on—but don’t you think that, for example, one of Kestra’s biggest customers or users could be Accenture? That they’d just say, “We understand this tool; we use it to enable enterprise customers to set up all their workflows properly.” Maybe you could even spot a LangChain in the corner somewhere. My feeling is that you just have a lot of companies that have now understood what’s possible with AI, but are still lagging a bit behind in adoption because, in companies, people first need to be educated—they need to be enabled first. And then, kind of like what Palantir does, people are actually deployed on-site. That’s actually a typical consulting budget, really.
Björn Rieckhoff: I think it’s also very much about having a deep understanding of the processes that need to be managed and coordinated in the first place. And I believe that, as you just pointed out, it often makes a lot of sense to have support from an external third party—a very technical consulting firm. Organizations like Accenture, for example, are, I think, ideally suited for this. And I agree with you there—you probably understand the complexity, or you might not necessarily be in a position to assess the complexity yourself, to then choose the best possible tool for the job. So I do think you need help there. I think I’d probably place Kestra somewhat in the enterprise solutions category alongside something like Microsoft Azure, a Data Factory, or something similar. You’re currently active in that space. And these are often topics that technical consulting firms tend to address. I completely agree with you on that.
Jan Thomas: Yeah, I’m just smiling because I think the job description has changed drastically, or rather, the skill set required. I mean, what you need to bring to the table to make a company future-proof—compare it to, say, five years ago, before ChatGPT 3, and today. It’s totally crazy. It’s a whole new setup and orchestration of tools you need to know, plus data flows, automation, and so on.
Björn Rieckhoff: Yeah, totally, and above all, it’s super fast-paced, as you mentioned. So what was relevant five years ago might already be outdated today, and you just have to keep up with the times very, very aggressively and, in a way, build in redundancy so that you can improve or replace the systems relatively quickly. Because at its core, you’re still solving the same business problem. But the way you solve it is becoming significantly different. I recently read a LinkedIn post — I think it was Peter from Cleandum — that's a nice analogy here: the skill set of startup employees has been shifting dramatically. Away from domain experts with ten-plus years in a specific field, toward younger people who think natively in agents and workflows and ask themselves: how do I make this as productive as possible? And I think this can likely be viewed in a similar light in these highly complex enterprise environments, where the role of the CDO is changing significantly.
Jan Thomas: And what I find exciting about it is, well, for one thing, the speed at which it’s changing is, of course, incredible. And then there’s that great quote from Wayne Gretzky, who once said, “You don’t have to be where the puck is; you have to be where the puck is going.” And anticipating that today — understanding what a perfect company you’re currently transforming will actually look like in three to five years? Super exciting. I can also imagine a lot of people sitting in the office at night, crying into their beer. Because they just don’t know where to go from here — that’s pretty intense.
Björn Rieckhoff: I think you could probably organize a great meetup or conference right now, and you’d walk away with just as many questions as you walked in with. So, you can examine the problem from all sorts of angles, but nobody really has a definitive answer. And it’s kind of like trying to look into a crystal ball. But I do agree with you—the pace of change is truly enormous. And I think what’s important right now is simply to be very agile about it and say: we’re not stuck in our ways with our standards, but are very open-minded, looking at solutions that could work tomorrow, and constantly questioning how we use our tech stack.
Jan Thomas: Well, because when you think end-to-end like that. By the way, I’m curious to see if this will be the biggest exit in Berlin this year. I’m really, really looking forward to it. But two or three years ago, nobody had that on their radar. Not even Kestra. Who knows what else is coming next year. Do you maybe know that smartphone competitor Nothing? They’ve kind of predicted that almost all apps will disappear. I find that super exciting, too. It feels like everything is being turned upside down. And you don’t know what will actually become reality and what won’t. But these are mega exciting times.
Björn Rieckhoff: What I always find interesting is to question, to some extent, the thought patterns we ourselves operate within. And I think we just happened to grow up and be socialized in a world shaped by the internet and the App Store, which came along at some point, so we simply think within those patterns. So it’s also interesting to see how this actually plays out with future generations, who no longer think in terms of an MS-DOS folder structure, but are much more flexible in their thought patterns. So in my head, it still remains a certain decision tree or folder structure, something linear. And that’s just not necessarily the case anymore. And that’s how you can really see what kind of impact that can have.
Jan Thomas: I think we can still speculate a lot right now. But I believe we can say for sure that the cards are being reshuffled. Nothing will remain the same. And it’s kind of about—though I think this is actually fun—challenging yourself a bit to question the status quo of everything and ask whether it will hold up in the future. And what if not? What might the resulting picture of the future actually look like? And I think you can look in any direction and say, “Oh, things could change there.”
Björn Rieckhoff: Things could change there, and I believe—coming back to those two topics—that it’s becoming clear this goes much deeper than just the application layer and what I can do with OpenClaw itself; it’s also about where the actual levers are on the infrastructure side, both on the software and hardware sides. And I find that particularly exciting to follow as part of this maturation process.
Jan Thomas: Then I’d say let’s take a break here. Maybe you could briefly explain where you’d be right now if those two had reached out to you. Personally, I wouldn’t have identified them as exceptionally important startups, I have to be completely honest. I lack a bit of that understanding—though of course, it’s always easy to say that in hindsight, now that they’ve closed such large rounds. But if they had come to me before a round, phew, I would have had a hard time.
Björn Rieckhoff: That’s not exactly trivial. I agree with you—especially when you think about companies like Kandou—it’s often so specific and de facto deep tech that you really need a very vertical, specialized understanding of the field. And I think that’s something generalist VC funds often don’t have access to. What would I do if startups like that reached out to me? It’s often about asking the right questions and bringing some structure. What’s actually the priority? What are we working on? That sounds trivial, but it’s often lever number one — because most early-stage teams have endless opportunities but struggle enormously to prioritize or structure them. And that’s where I help with a bird’s-eye view. And that ranges from redefining an ICP and a go-to-market strategy to, well, now that we’ve laid the foundation, what is the actual narrative for investors, and which specific investors are we targeting. And so I’ve been thinking over the last few months: that antiquated VC list that founders always get forwarded by their investors — “talk to these ten funds” — nine out of ten of those conversations lead nowhere because the focus doesn’t match or the fund is heading in a completely different direction. It’s outdated. So I sat down and tried to back the whole thing with actual intelligence — looking at current investment behavior among European VCs to give founder teams qualitatively better guidance on who they should actually be talking to right now.
Jan Thomas: But to me, that’s almost a separate topic—maybe we can dive deeper into it next time, Björn. It might be a bit abstract, but there could be a lot to learn there—looking at startups through an investor’s lens, figuring out who to talk to. I think we could almost do a whole episode on that.
Björn Rieckhoff: Sure, we can do that.
Jan Thomas: Then I look forward to next time, Björn. That was a lot of fun. Thanks to you.
Björn Rieckhoff: Me too. See you soon, then. 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.
Raising a round? Curated investor lists for European early-stage rounds, delivered within 48 hours: $ cd services/investor-intelligence