‍A city that already knows how to move money

Stockholm in September holds its light past ten at night, and Viggo had picked a spot for our first real conversation that made his point before he said a word: a converted print shop on Hornsgatan, three floors up, with the whole of Riddarfjarden laid out through the windows and a ferry crossing it every few minutes without a single passenger appearing to check a card reader. Every ticket, every coffee at the kiosk on the quay, every small transaction moving through this city runs on rails that Swedes stopped thinking about a decade ago. It is an easy country to build the future of payments in, because the present already looks a little like it.

Viggo talks about payments the way a sailor talks about tides, patient and almost fond, like something with its own schedule that nobody gets to argue with. I had expected a pitch. What I got, over two hours, was closer to a man explaining weather.
‍

The fourth wave

“I have watched money learn to move in a new way four times now,” he said, turning his coffee cup a quarter turn without drinking from it, the way people do when they are about to tell you something they have told themselves many times before. In 2000 he was at Spray here in Stockholm, building websites while money started moving over the internet before anyone had agreed to call it ecommerce. At Spotify he watched streaming rewire how people pay for software, from ownership to subscription. In 2022 he founded a blockchain company, another moment when money wanted a new shape. Each time, some of the old rules survived and something new had to be built around them.

The fourth cycle is the one he is betting the company on. There is an actor in the economy now that has never existed before, the AI agent. It has a goal and a mandate to act, and it hits a wall the moment money enters the picture.

“Almost everything in payments was designed around a human clicking a button. The people who saw the agent wave first were AI people, not payments people. I had watched money find a new way to move three times before. I recognised the fourth.”
- Viggo Stenseth

A dreamer, a pragmatist, and the hacker between them

The office belongs, on paper, to three people, though only Viggo was in the room that afternoon. Ingemar Svensson joined by phone from a client site in Dubai, his voice arriving thin and slightly delayed through the speaker Viggo had propped against the window. Tommy Berglind wandered in halfway through with two paper bags of cinnamon buns and an apology about a delayed train on the tunnelbana.

Viggo is the hacker in this configuration, a title he carries with a specific kind of pride. His role is defined by a deep-seated need to take systems apart and put them back together again, driven by the conviction that you cannot truly understand how a piece of technology works, or where it might fail, until you have seen its guts laid out on a workbench. It is this mindset that translates the dreamer's ambition and the pragmatist's constraints into the actual architecture of the fourth wave.

Ingemar is the one Viggo calls the dreamer, an investment banking technology veteran who first sketched what became SolvaPay on the back of a client presentation he was supposed to be finishing. “He is the one who insists we aim at the whole thing rather than a feature,” Viggo said. Down the phone line, tinny and half a beat behind, Ingemar put it more bluntly: “If you are only going to solve the easy half of the problem, do something else with your life.” Coming from someone who spent a decade inside institutions built to reward caution, the line carried a certain weight.

Tommy is the pragmatist, the one who, in Viggo's phrase, never stops advocating for the user, and also the reason the other two are in a room together at all. He and Ingemar had worked together before. He and Viggo go back years further than that, through skydiving, techno and code, three pursuits that reward the same instinct: trusting a system you did not build and cannot fully see. “We found each other before we found the idea,” Tommy said, setting the cinnamon buns down like evidence. “The trust came first. The company came later.”

Between the three of them they have built payment infrastructure across Spotify, Handelsbanken, Bank ofAmerica,Merrill Lynch, and Barclays. Interestingly, their deep expertise in existing infrastructure produced a deep-seated conviction that existing rails need to be overcome and rebuilt from the ground up.

‍“A dreamer, a pragmatist and me in between means an idea gets pushed to its most ambitious version and pulled back to something we can actually ship, often in the same conversation.”
- Viggo Stenseth

The friction, Viggo admitted, comes from the same place as the strength. In a team this small, working a problem with no natural edges, the three of them are constantly switching between discovery and delivery, sometimes several times inside one afternoon. Knowing when to keep dreaming and when to close the laptop and ship, he said, is the hardest rhythm they have found.
‍

The night of the tweet

He traces the real conviction back to a headline, scrolled past on an ordinary evening at home in late 2024. Marc Andreessen had sent bitcoin to an AI agent on X so that it could self improve and spread out into the wild on its own. Viggo stayed up most of that night and wrote a full page in his journal about why the world needed what he called an agentic bank, because the financial system as it stood had nowhere to put what he had just watched happen.

“Agents were still fairly unsophisticated back then, to be fair,” he said. “And soon after, the big card networks put out headline announcements about agentic SDKs that turned out to be mostly empty under the hood.” That combination, a real glimpse of the future sitting beside incumbents clearly caught off guard by it, stuck with him. He called Ingemar within the month. The three of them started building together in August 2025.
‍

The checkout page is already dead

Viggo's counterintuitive bet, the one most people in payments still consider premature, is that you build for in chat and in agent payments now, before any of the standards have settled. “Most sensible people say it is too early,” he said. “Agent payments are not mainstream yet, the protocols are still moving, regulation is unclear. We bet the opposite way. In payments, the winners get decided before the category is obvious, never after.”

In practice that means a zero code wedge that captures revenue where users already are, inside Claude, inside Cursor, inside ChatGPT, while the larger agentic payment layer gets built underneath it.

“The old checkout assumes the user leaves the workflow. They are in a conversation, they hit value, and then they are sent somewhere else. Every step along the way is a place to lose the sale. The next storefront is the conversation itself.”
- Viggo Stenseth

Betting on the bridge

The decision Viggo is proudest of, in hindsight, was going early on MCP, back when it was still contested and far from an obvious standard. “We saw it as a versatility bridge,” he said, “the connector that lets a business plug into the agentic economy. The one thing it was missing was a payment layer. That is exactly what we added.” A business becomes payable now, with a human in the loop today, and ready for full agent to agent transactions once that arrives.

The honest answer about money

I asked him directly about the funding gap, SolvaPay's two point four million euro pre seed against competitors who have raised considerably more. He answered without much hesitation.

“At this stage the currency that matters is whether you have earned the right to raise again by proving traction, more than how much you have already raised. Capital buys speed, but it can also lock you into a bet before the ground settles, and in this market that is a real risk.”
‍- Viggo Stenseth

On Skyfire and Nevermined specifically, he was more generous than most founders manage to be about their competition. “I am impressed by them for going extremely early, effectively the first to test agent payments in the wild. Their crypto bet looked like a natural fit for agents, and I am convinced crypto will end up somewhere in this infrastructure eventually. The front of this market, though, still runs through the regulatory and financial layer, and that moves slowly. You can already see both of them drifting back toward more conventional approaches. We are betting on meeting money where it actually moves today.”
‍

The ferry back

We left as the light turned long and copper over the water, the kind of Stockholm evening that makes the whole city look freshly rinsed. Tommy walked us down to the quay, still talking, the way people do when they have said something true out loud for the first time in a while and cannot quite stop. Ingemar's voice was still faintly audible from Viggo's phone, propped in his jacket pocket, arguing some final point about the roadmap that nobody was going to resolve that evening.

Standing at the quay, watching the same contactless ferry that had made the point before anyone opened their mouth, I kept coming back to a plain fact. This team had, between them, spent years inside the institutions that built the last version of how money moves, and had chosen to walk out and build the next one with almost nothing borrowed from the old one. Viggo said something on the way down that stayed with me longer than any of the numbers he had given me that afternoon. Money is about to be able to carry a goal, to act in someone's interest, to work quietly in the background while nobody is watching it. It sounds almost too simple, until you sit with what it actually means.

Redstone invested in SolvaPay because the founding team has lived through three prior cycles of money changing how it moves and is betting, in our view correctly, that AI agents are the fourth. The company's early wedge into zero code, in agent payments earns it real transaction volume now, while its early bet on MCP positions it as the payment layer underneath the wider agentic economy as it matures.

Redstone is one of the most active European early stage VCs and holds a top decile track record across its sector funds.

‍

‍Error compounds

Alexander joined the call and shared his screen before I had asked a single question. Not a slide with the company name on it. A drawing: a small robot at the top of a family tree, branching down into six smaller robots, "error compounds" underlined twice in red, next to a steering wheel wired into a grid of a thousand tiny mechanical parts, each one checked off or flagged. Jesper had posted a version of it on LinkedIn a few weeks earlier. Alexander pulled it straight back up.

"That's the whole pitch, if you want it in one drawing," he said. "Everyone else is building the robot family tree. We built the grid."
- Alexander Oelling

Alexander is INXM's co-founder and CEO. He spent years before this as Chief Digital Officer at Isar Aerospace and Volocopter, leading digital transformation in regulated, safety-critical environments, after founding and scaling Sensorberg into one of Europe's largest proximity infrastructure providers. I had come to hear how a serial founder, a systems architect, a product designer and a manufacturing veteran had ended up building the same company, and why they think the entire industry is arguing about the wrong bottleneck.

Four founders, one failure

None of the four arrived at this from a typical software background, which turns out to matter more than it sounds. Matthias Kainer, the CTO, has spent more than twenty-five years in software development and systems architecture, including stretches at Microsoft, AutoScout24, ThoughtWorks, Volocopter and Isar Aerospace. Jesper Bylund, the CPO, led design at Volocopter and product design at n8n before that. Kamil Klüber, the CSO, spent nearly seven years at Siemens driving digital transformation across aerospace, defence and new space, and years before that at HELLA, working the entire product development value chain in automotive. Between the four of them: rockets, air taxis, product design, factory floors. Four different industries, and, as it turned out, the same failure sitting underneath all of them.

"We compared notes across our backgrounds, and we kept landing on the same pattern. Not a lack of intelligence or data. The intelligence was there. What was missing was execution, reliable, auditable, repeatable execution at the process level. That isn't sector-specific, but structural. In that moment we stopped saying this is interesting and started saying this is the problem."
- Alexander Oelling

It is a specific kind of tolerance, and Alexander argues most software teams never develop it, because they have never operated anywhere failure has physical consequences. Rockets require determinism, you cannot patch a launch failure in production. Air taxis require certification, you cannot ship a workaround. Factory floors require reliability at scale, where a system that works ninety-seven times in a hundred is a system that fails on the other three, and that failure has a cost attached to it.

A system that sometimes works and sometimes doesn't isn't a system. It's a liability. That's exactly the lens we apply to enterprise AI. An AI that occasionally hallucinates its way through a process doesn't reduce operational risk. Instead, it introduces a new kind of risk, one that's even harder to audit than the problem it was meant to solve.
- Alexander Oelling‍

It became real at the notary

We went back to the beginning, or rather to the specific afternoon it stopped being an idea.

"It became real at the notary. Until then, it's all ideas. You meet a future customer, an investor, you talk a bit here and there, and it stays a pleasant, hypothetical conversation, because you can't raise money before there's a company to raise it into. Then the notary turns to you and says, sign here, you're founding the company now. That's the moment it stops being virtual."
- Alexander Oelling

He is careful not to dress this up as a leap of faith. "You don't know yet whether it will fly. You have a feel for the market, real experience in it, enough conversations behind you, and a conviction. At some point you simply find it, when the team is together and you understand the market, then you can see where it's heading. So you say, now we build it." The conviction, he told me, came from a gap he had been staring at for years without a name for it. Nobody was actually serving the enterprise customer, and given the sheer number of industries and niches involved, no single company ever would on its own.

One bet

This is where the actual bet gets interesting: people initially thought the team was being too conservative, too technical, or simply late to the party. While most of the market raced toward autonomous agents, INXM bet on something narrower, and in their telling, more useful. AI that plans, and an orchestration layer that executes.

"Not because we didn't want AI to do the work. We absolutely did. But autonomous and reliable weren't going to arrive at the same time, and in enterprise you can't ship the gap."
- Alexander Oelling

The product is called Orchestrator, and the underlying approach is what the team calls Compiled AI, a term with an origin story Alexander is oddly insistent on getting right. "The honest answer is we didn't coin it alone. It comes from a research paper, on deterministic code generation for LLM-based workflow automation. When Matthias read it, his reaction was simple, these are the words I've been looking for. We'd already built the thing. We just didn't have the word for it." The mechanism, once you see it, is straightforward. An LLM is used in a compilation phase to generate executable, auditable code for a given process. That code gets validated. Only then does it run, deterministically, with zero tokens spent at runtime.

A workflow tool executes the steps someone defined last year, built by professional process engineers in complex tools. The Orchestrator executes the steps that are right today, in the customer's own systems and language, without needing an engineer to rebuild the whole process every time something changes. "An operator on a factory floor doesn't want to interact with AI," Alexander said. "They want their task done. And the plant manager doesn't want to trust a black box. They want to see what ran, when, why, and what they can change."

An eighty times cost difference

The figure Alexander keeps returning to is the cost curve. ‍

"A conventional LLM approach runs roughly a hundred thousand or more tokens per SAP transaction. Our compiled execution runs the same transaction in one thousand to twelve hundred and fifty tokens, because the intelligence was spent in the compilation phase, not the execution phase. That's an eighty times cost difference, on a process that runs thousands of times a day."
- Alexander Oelling

He is careful to frame this as an architecture decision rather than a bet against the models improving. "Even if execution reliability eventually becomes a native property of the models themselves, the cost structure is already decided. It's not a bet that today's models stay flawed. It's a bet that the economics compound regardless."

I asked the obvious sceptic's question. A CIO has just spent ten million euros on SAP. Why does he now need another layer on top of it? Alexander didn't flinch. "Because the ten million didn't buy you a process layer. It only bought you a system of record. Your SAP knows what happened. Your orchestration layer decides what happens next, coordinates it across every system involved, and executes it without a developer every time a process changes. Two different problems. We don't compete with SAP. Instead, we make the SAP investment actually executable in an AI world."

What it took

The part of the conversation that stayed with me longest had nothing to do with tokens or orchestration layers. I asked what he draws on when things get hard, and he didn't reach for a business anecdote.

I've been doing this a long time, over twenty years as a serial founder, and more resilient with every one of them. The sharpest turning point was getting seriously ill. I had cancer, and it forced me to look at everything from a completely different angle. You learn a certain humility toward the universe, and at the same time a hard realism. That's what it takes to make it as a founder. It's a constant struggle with yourself, and I won't pretend I've fully mastered it.
- Alexander Oelling

He credits staying restless, building teams that do not need him standing over them, and, deliberately, keeping the rest of his life stable. "None of it works without a stable environment and the backing of your family. You can't afford too many construction sites in your life at once. With a small child at home, you don't need a second source of chaos on top of the startup."

That same instinct for where not to spend energy shows up in how he talks about the hardest problem the company has actually had to solve, which, tellingly, was not the natural-language pitch that sits at the centre of INXM's own product. "Honestly, that isn't the hard part. The hard part is understanding the customer's use cases and putting the customer first. For that you need process and industry experts, not the software developers the same project used to require." The genuinely difficult part, in his account, is product-market fit in the plainest sense, getting honest about whether technology that impresses everyone in a demo can be pressed into a form that creates measurable value for a specific buyer.
‍

The walk out

Before I left, I asked what has to be true in three years for any of this to have worked. Alexander didn't reach for a growth number. "Customers, customers, customers. Projects that are simple to connect. AI that's not only safe, practical and transparent, but genuinely easy to adopt." Zoom out to the ten-year version and the ambition gets larger without getting vaguer, a company that behaves the way AWS does inside a business today, not a tool anyone evaluates on a shortlist, but the layer the business quietly runs on. "Companies don't think about AWS as a vendor but as the layer their business runs on. That's the position we're building toward."

"It is a question we spend more time on inside Redstone than founders probably assume. We argue about it constantly across the team: how much of the stack above the model gets absorbed by the model itself, and how much stays a separate, defensible layer that someone has to build on purpose. Early on that felt like a question with a comfortable answer arriving eventually. The longer we sit with it, the harder it gets to say with any confidence which side of that line INXM's own bet lands on, and I have come to think that is less a gap in our own thinking than the honest state of the argument right now."
- Michael Brehm

Walking back out past the whiteboard, the robot family tree still had error compounds underlined in red. I found myself thinking about the grid next to it instead, small, checked, auditable transactions, unglamorous by design. INXM's argument, in the end, is a quiet one. The industry has spent two years asking its models to be smarter, when the thing actually breaking in production was never intelligence at all.

Redstone invested in INXM because the company solves the execution problem of enterprise AI, a structural bottleneck that persists no matter how much model intelligence improves. INXM arrives at the precise moment enterprises are moving from AI pilots to production deployments, where reliability, auditability and cost per transaction decide who actually keeps the budget.

Redstone is one of the most active European early-stage VCs and holds top-decile track record across the sector funds.

‍

‍The valley pulls you in

The woman at the rental desk in Munich asked whether I wanted the upgrade, and I said no, because you do not need much car to get to Tirol. Ben climbed into the passenger seat with two coffees and a briefing folder, and we began our journey south.

You feel the Alps before you see them. The autobahn runs flat and businesslike for the first hour, business parks and rapeseed fields and the occasional furniture warehouse the size of a village. Then, somewhere past Rosenheim, the ground stops pretending. The mountains come up out of the horizon all at once, and by the time you cross into Austria at Kufstein the road has narrowed into the Inn valley, grey rock standing close on both sides, old and completely uninterested in you.

Innsbruck sits at the bottom of that valley the way a coin sits in a cupped hand. The Nordkette rises almost vertically off the northern edge of the city, you can stand on a shopping street and look straight up at alpine pastures. We did not stop there. Götzens is a few minutes higher, on the sunny terrace along the southern shoulder, one of those villages where the road folds back on itself twice and then the whole valley drops away beneath you. It is not a place you would find by accident. You have to be going there.

Which is, in a way, the whole point. The best hardware in Europe is very often being built somewhere you have never heard of, by people who prefer it that way.

Three founders and a leaf

We met the three of them in a small bakery on the main road, the kind with four tables, a glass case of Krapfen, and a coffee machine that was paid off a decade ago. Thomas Rettenwander, Johannes Mandler and Elias Hirschbichler. No pitch, no laptop. Thomas ordered for the table, and then, before anyone had asked a single question about the company, he started talking about leaves.

Thomas is the one who sees the shapes. He grew up here, in the mountains, and he has the particular relationship with the natural world that people who grew up inside it tend to have, unsentimental and very close. He doesn’t look at nature as a backdrop or a piece of scenery. To him, it’s a system perfected over hundreds of millions of years: flawless, functional, and completely solved.

He spent his childhood summers by the mountain streams, watching hours slip away as he moved rocks in the rushing water. He noticed how the current instantly adapted to the changes, and how the pebbles took on smooth, precise shapes over centuries just to survive the friction. Stacking those stones on the bank, he realized their geometry wasn’t an accident. The water had carved away everything except the most efficient possible shape. To him, the stream wasn’t a playground. Instead, it was a place where physics spent thousands of years getting the answer exactly right.

“Pick up a leaf and hold it to the light. The veins run where the load runs, and nowhere else. A dragonfly wing is built the same way. Nature has been solving the lightweight problem for a few hundred million years, and the answer never changes. Put strength only where the force actually is."
Thomas Rettenwander

He took a paper napkin and drew a wing. Then he explained the thing I have not been able to stop thinking about since. A dragonfly wing is not a uniform sheet. The big vein along the leading edge is thick, because that is where the bending happens. The veins thin out toward the back, where there is almost nothing to carry. Near the tip sits a small dark cell, the pterostigma, a deliberate lump of extra mass whose only job is to stop the wing from fluttering itself apart at speed. Nothing is even. Nothing is decorative. Every gram is either doing a job, or it is not there.

‍

A dragonfly wing carries its material where the force is: heavy along the leading edge, sparse at the trailing edge, with the pterostigma near the tip tuned to damp flutter. The same idea, translated into a loaded bracket, is the whole of fibionic’s argument.

“We have known this for a hundred years, and we build almost everything the opposite way. Beams. Plates of even thickness. Material smeared across places where it does no work, because that was the version a machine could turn out. The clever version, the one that copies the leaf, we kept for the few places where money is no object. A satellite. The frame under a Tour rider.”
Thomas Rettenwander

I said it reminded me of Gaudí, and Thomas grinned like I had passed a small test. Antoni Gaudí used to hang chains from the ceiling of his studio, let each one settle into the curve a chain naturally takes under its own weight, then photograph the model and turn the image upside down. The arch he built in stone was simply that hanging curve, inverted. He worked this way for decades, most famously on the crypt at Colònia Güell, a testing ground he treated as a monumental model for the Sagrada Família itself. Instead of forcing a shape onto the architecture, he listened to what the physics demanded and got out of the way. You can still see the same logic standing in Park Güell today, where the viaducts run on columns that lean and branch the way tree trunks do, built from rough stone taken straight off the hillside, carrying carriage roads overhead without a straight line or a right angle anywhere in the structure. Nothing there was decided at a drafting table. It settled into place the way water finds its channel, and then someone built it in stone. That is the entire idea here. Do not decide where the material should go. Let the load decide, and follow it.

By now we had been in the bakery forty minutes and nobody had mentioned a valuation, a market size, or a competitor. Ben caught my eye over his coffee. We both already liked them. That is not diligence, and it does not survive contact with a spreadsheet, but it matters more than the industry likes to admit. You’re going to be in the trenches with these three for the next eight or ten years. When their names pop up on your phone, you actually want to feel good about answering.

Johannes is the one who turns the idea into a customer. Where Thomas talks in centuries, Johannes talks in purchase orders. He has the slightly scarred patience of someone who has heard “very interesting, send us your deck” a hundred times and has learned to hear it, correctly, as a polite no.

He is tired of one comparison in particular. “Someone picks up one of our parts and says, ah, so you do 3D printing.” He said it with the weariness of the tenth time that week. “And I have to explain that no, we do not. 3D printing had a beautiful idea, material only where you need it, and then it ran that idea one slow part at a time, for prototypes. We took the idea and built it for real series production. It is the step that comes after additive manufacturing.”

Elias is the one who has to be honest about money, and he is the least interested of the three in making any of this sound easy. He runs the finances, which in a company like this means he is the one who knows exactly how long the runway is on any given morning.

“Deep tech does not scale because the idea is good. It scales because you survive every step, one after another. Technical proof, customer trust, financing, and teaching a market that did not know it had a problem. Each one of those can kill you, and they never arrive at the same time. It is slower than software and it eats more capital. Anyone who tells you otherwise has not built one.”
Elias Hirschbichler

Above the hall

Then they walked us over to the office, which turned out to be a single room on a mezzanine directly above the production floor. You could feel the machine through the soles of your shoes before you saw it. Two desks, a whiteboard carrying the sediment of old arguments, and a window looking down onto the hall. This is my favourite kind of company office. Nobody has spent a cent on it, because every cent has gone downstairs. You can feel the founder spirit in a room like that; it is the physical proof that the company still puts everything into the thing itself.

Standing at that window, they explained how it actually works, and the market opened up in front of us as they talked.

The process has a deliberately unglamorous name, Fastest Fiber Placement, and one number that does all the arguing. As little as a minute. That is the cycle time they hit on a shallow part, fibre laid along the load path the way the leaf lays it, which they call a world record and which nobody has yet beaten.

That number is the entire line between a science project and a business. A part you lay up by hand over an afternoon belongs in a wind tunnel. A part that comes off the line every minute belongs in a catalogue, at a price an ordinary product can carry.

‍

‍

“Carbon fibre has been extraordinary for forty years,” they said, “and the reason it never reached your car or your running shoes was never the fibre itself but that nobody could lay it down fast enough, cheaply enough, or cleanly enough to make the arithmetic work. A production problem wearing the costume of a material problem.”
Thomas Rettenwander

Fix the production, Johannes said, and the logic flips. Suddenly the leaf-shaped part is not the expensive option. It is the cheap one, because you have stopped paying for all the material you used to waste. Stand at that window and start multiplying, every beam, every plate, every bracket built the dumb way, and the size of the thing becomes slightly hard to breathe around.

They had gone hunting for real demand in sport, on purpose. The performance requirements are real, the barrier to entry is lower than it is in aerospace, and the customer can feel the difference in their hands: a stiffer bike part, a surf fin tuned between flex and rigidity, a sole that springs back. With Selle Italia they put reinforcement structures into series at thousands of parts a month, and kept them there.

“Deep tech does not scale because the idea is good. It scales because you survive every step, one after another. Technical proof, customer trust, financing, and teaching a market that did not know it had a problem. Each one of those can kill you, and they never arrive at the same time. It is slower than software and it eats more capital. Anyone who tells you otherwise has not built one.”
Elias Hirschbichler

"That is the moment it stops being research." Johannes said. "When a real manufacturer rebuilds part of their product around what you can do, and then reorders. You are not running a pilot any more. Instead, you are a supplier."

Investors had been wary, Elias admitted, the way the whole market is wary of anything with a machine in it right now. Hardware reads as risk. What eventually moved the conversation was not a sharper pitch. It was a reorder.

The other edge of that patience is quieter, and it is the part that made me lean in at the window. Once the process runs and the patents hold, what protects the company is not a line of code a rival can rewrite over a weekend. It is years of physical reality that nobody else has lived through yet. A slow moat. And these three seemed genuinely comfortable with slow, which is rarer than it sounds, and hard to separate from where they sit, a village outside Innsbruck ringed by firms that have stayed relevant for decades.

Elias framed it as a European point more than a Tirolean one. “Europe has the ideas and the people,” he said. “What it too often lacks is the nerve to industrialise them at home, instead of selling the patent and watching the factory get built somewhere else. We kept all of it here. Design, machines, production. Partly stubbornness. Mostly a bet that the next decade rewards the people who can actually make things.”

The machine

Then we went downstairs, and I stopped taking notes.

I will not pretend I can describe the mechanism properly, partly because they are careful about what they show, and partly because a good deal of what makes it work is not visible at machine speed. What you see is fibre being placed, fast, along curved paths, the material fanning and turning the way it had on the finished part Thomas handed me in the bakery, dense in one corner and almost bare in another. From a step back the surface looks less manufactured than grown. It does not look like a printer patiently building up a prototype. Instead, it looks like a production line that happens to be laying down exactly the pattern a leaf would.

You develop, in this job, a physical reaction to the moment a demo stops being a demo. Something in the room changes when the thing on the bench is obviously not a trick. Ben and I did not say anything to each other. We did not need to.

The road back

We drove out of the valley in the late afternoon, the light going long and gold across the rock, the Inn grey and fast on our right. Ben talked for the first ten minutes and I drove and mostly listened.

You spend a great deal of this job being sold to. People arrive rehearsed, having decided in advance which version of themselves you want to meet. What we had just spent a morning with was the opposite. Nobody had inflated anything. Elias had spent as much energy telling us why it was hard as why it would work. Thomas spoke about the company as something that simply needed to exist. It’s the kind of project you pursue because it is too important to leave unbuilt, no matter how steep the odds or how distant the payout. Johannes had a customer who reorders, which is the only sentence in a pitch that cannot be faked.

The honest tell of a good founder is what they admit they have not cracked, and here it was not technical. It was the shape of the company. Fibionic does not want to stay a part maker. The longer game is to sell the whole logic, the machines, the software, the optimised material, and let other manufacturers run the process inside their own lines. That is a far harder company than the one quietly turning out bike parts in Götzens. It asks an entire industry to rebuild its production around your idea, and that kind of trust is not won with a demo.

Crossing back over the German border, we weren't debating the viability anymore. We were just arguing over the scale of it, and how fast we could move.

“We are not building for a story,” Thomas had said, back in the bakery, turning the part over in his hand. “We are building because something needs to exist. A leaf does not waste material. We just want to stop doing it everywhere else, at scale, at a price people will actually pay.” Plenty of companies can make a beautiful optimised part. Almost none can make it at the price of the dumb one. That last clause is the whole company, and it is the reason we drove home knowing we had stumbled onto something that could rewrite the rules of manufacturing.

Alpine SICAF (Euregio+ & Redstone Advised) backed fibionic because the company attacks the right bottleneck: not the material, but the cost and speed of making it. The economics that kept optimised composites locked inside aerospace and elite sport are exactly what fibionic dismantles, at the moment industry can no longer treat performance, sustainability and price as a choice of two. Redstone is one of Europe’s most active early-stage investors, with a top-decile track record across its sector funds.

Redstone is one of the most active European early-stage VCs and holds top decile track record across the sector funds.

‍

The most dangerous job you never think about

Sitting down with the team at Tethys, the first thing that hit me was the raw reality of their mission. They are throwing everything they have into fighting the terrifying, claustrophobic conditions of commercial diving.

Commercial diving is one of the most lethal professions on earth. Imagine being dropped into a freezing river with zero visibility. You can’t see your own hand in front of your face. The flow is ripping, constantly trying to slam you into concrete pillars or drag you downstream. You have exactly 30 minutes of air. In that time, you have to blindly feel your way around massive underwater structures, trying to find a hairline crack in a bridge foundation or a leak in a pipeline. Your adrenaline is spiking, your body is freezing, and if your tether gets snagged on a piece of debris, you don’t come back up.

"Seeing a diver go through that changes how you look at the entire industry. We were, back then, sitting in an unpretentious room in Zurich, surrounded by prototype hulls and stripped-down sensor rigs.”
— Jonas Wüst

It’s barbaric that in 2026, we still send human beings into high-risk meat grinders just to take a look at a concrete wall. It is an environment built to break the human body. Robots need to take that hit first. They need to go into the dark, map out the danger, and do the heavy lifting before a human ever touches the water.

That is the core of Tethys. It started as an elite robotics project, but it quickly became an urgent mission to fix a massive, life-threatening blind spot in our global infrastructure.

The chocolate milk problem

We tend to think our world runs on clean software and radio signals. But the physical truth is much messier. The real backbone of our civilization: the pipelines that supply our cities, the fiber optic cables that keep the internet alive, the foundations of offshore wind farms, and the massive ports that drive global trade, lies entirely underwater. And right now, we are virtually blind to what is happening down there.

The market has chronically misunderstood this problem. For years, legacy players thought the solution was just adding a crisper camera to a traditional Remotely Operated Vehicle (ROV). But a better camera is useless when the water looks like chocolate milk and the current is spinning the drone like a top. The real bottleneck isn’t the camera. It’s the brutal, unyielding workflow.

Tethys took a fiercely counterintuitive bet. While the rest of the industry assumed you needed multi-ton naval ships, massive cranes, and a small army of specialized technicians to run a serious underwater mission, Tethys believed they could shrink that entire footprint into a compact, ultra-rugged system. By combining edge-computing autonomy with highly proprietary sensor fusion, they built a drone that handles raging currents and navigates completely without GPS. They called it the Tethys ONE. It’s a system that a two-person team can deploy from a simple inflatable boat or a shoreline in minutes, turning what used to be a massive maritime operation into a routine, repeatable task.

Underwater Drone, Tethys Robotics

You can't hotfix a sinking drone

Rather than an award or a tech breakthrough, it was the intense stress of real-world accountability that marked Tethys's evolution into a high-stakes business.

"The The shift happened when the conversations changed. Prospective clients stopped patting us on the back and saying ‘cool technology’; they started looking at us with desperation and asking, ‘Can you deliver two systems by next month? We urgently need to inspect an international subsea cable, and every day of delay is costing hundreds of thousands of dollars while increasing the risk of a public safety crisis.”
— Jonas Wüst

In hardware, you can’t fake readiness. If software bugs out, you push a hotfix. If an underwater drone loses navigation in a harbor, it sinks into the mud or gets crushed by a freighter. Out here, customer safety hangs on every line of code, every hull seal, and every link in the supply chain. You know you’re building a real enterprise when you stop wasting time convincing people the problem exists, and start sweating over how to scale fast enough to match their operational panic.

‍Out of the lab, into the freezing water

At Redstone, I see hundreds of deep-tech pitches per year. Most of them are looking for a problem they can solve with their shiny new technology. Tethys was different. They were staring directly into a multi-billion-dollar infrastructure crisis that everyone else was trying to ignore because it was too damn difficult and too dangerous.

Yes, Tethys has a world-class engineering team. But we invested because the problem they’re tackling is an absolute emergency. Inspecting underwater infrastructure the old way is slow, prohibitively expensive, and puts lives at risk. Tethys completely changes the math by replacing human exposure with silicon and software.

Ultimately, it came down to how they build. Instead of staying insulated in academia, Tethys took their hardware straight into freezing lakes and zero-visibility rivers to prove their autonomy worked where it actually matters. They chose the mud over perfect theory. When you combine that kind of grit with the massive commercial pull we're seeing from the defense, energy, and maritime sectors, the opportunity becomes undeniable. They are fundamentally changing how industry operates underwater.

Leave the danger to the machine

Today, Tethys is a rapidly growing team of 20 builders in Zurich. The focus for the next three years is clear: industrializing the platform, securing repeatable supply chains, and building out an automated data layer that takes raw subsea sensor feeds and turns them into instant structural reports.

But when I ask them about the ten-year horizon, the conversation always circles back to the human element.

"Success means that ten years from now, sending a human diver down for a blind, exploratory inspection will be viewed as an ancient, unacceptable risk.”
 — Jonas Wüst

The goal isn’t to replace the veteran inspectors who know these underwater structures inside and out. The goal is to give them a shield. In a decade, that same highly experienced diver won’t be risking a fatal pressure accident in the North Sea. That diver will be sitting safely in a control room, driving a fleet of autonomous Tethys units and making critical decisions based on flawless data.

The expertise stays with the human. The lethal risk stays with the machine. That is the future Tethys is building, and that is exactly why Redstone stands behind them.

Alpine SICAF (Euregio+ & Redstone Adviced) invested in Tethys because the company systematically solves the “first-mile problem” of maritime data acquisition: a structural, multi-decade bottleneck that has been continuously ignored even as global reliance on subsea infrastructure intensifies. Tethys arrives at the precise historical inflection point where physical AI and ruggedized autonomous systems transition from expensive experimental luxuries to absolute civilian necessities.

Redstone is one of the most active European early-stage VCs and holds top decile track record across the sector funds.

I first met Franz Tschimben three years ago in Berlin, at one of those sparse, under-heated startup events where everyone looks slightly uncomfortable. Our conversation developed differently than expected. Franz stood out asking the right questions, rather then just doing his normal 2min elevator pitch. About what we were seeing in the market, what was actually dumb versus real, why everyone seemed to be talking past each other. The conversation stuck with me.

So I went back to South Tyrol. To meet the whole team, including Harald and Burkhard. And I figured that they build what might be the most unglamorous, most necessary infrastructure company in AI right now. We sat in a room in Brixen on a gray afternoon. Coffee, whiteboard, visions and some good laughs. Here is what I figured:

The 3D Digital Twin Factory

Franz returned to South Tyrol after many years in Silicon Valley; Burkhard Güssefeld (CTO) has followed a teenage obsession with graphics cards and light physics all the way into computer vision and 3D reconstruction; Harald Oberrauch (President) brings decades of industrial company-building from the region. ALLSIDES was founded out of Covision Lab, an AI research centre in Bolzano. Their timing couldn't be better. The market for AI training data is projected to reach nearly $10 billion by 2029, and physically correct 3D has become the critical bottleneck inside it.

The Wrong Market, Twice

Burkhard was convinced early on: by gaming. “In the late 90s I was obsessed with graphics cards, GPU technology, shaders. The gaming industry had massive demand for realistic 3D assets. I thought that's the market.” Except gaming doesn't work that way. They build most assets themselves. They weren't paying for photogrammetry services. The real urgency was e-commerce: Amazon rendering millions of products, adidas and Zara needing every shoe in every angle. But even that wasn't the full story.

Two years in, the labs started coming: Meta, OpenAI adjacents, robotics companies. Asking for data that can teach machines how the physical world actually works. “That's when we realized,” Harald says, “we weren't building a 3D content company. We were building infrastructure for the next generation of AI.” All three shake their heads when I ask if they saw it coming.

Nearly all classical computer vision algorithms rest on brightness constancy: the idea that an object looks the same from every angle. It is also physically wrong. Velour reflects light differently than polished metal. Pearlescent surfaces shift colour with viewing angle. The data is broken before it is ever used. ALLSIDES inverts the logic. Those variations are treated as signal, not noise. A single ALLSIDES scanner now produces more than 30,000 physically accurate 3D models per year.

“The algorithm is maybe 30% of the problem. The rest is: how do you actually capture reality in a way that's useful? That's hardware. That's calibration. That's the entire data pipeline.”
— Burkhard Güssefeld

The Failure That Changed Everything

Franz had failed before, with AKER, which was basically Amazon Go for smaller supermarkets. He spent two years building the company in San Francisco. “We had Timing. We had Team fit. We didn't have real product-market fit. We had something customers found interesting, but not something they needed to pay for. And the founding team; we were good people, but we weren't right for each other. Those misalignments matter.”

“In that moment, it felt brutal. Like a personal failure.”
— Franz Tschimben

The consequence for ALLSIDES was one hardcoded constraint: objects only. No faces, no full environments, even though those categories attracted more attention and more funding. Franz and Burkhard ended those conversations early. Within six months they had designed, manufactured, and deployed their first scanners with Adidas, Zara, and Meta. Companies that expect 24/7 operation from day one. Early investors were sceptical: hardware is read as a liability in the past AI climate. What changed their minds was what the hardware demonstrably produced. The perceived liability turned out to be the moat.

The first moment of real doubt came early. “We had a technology that worked in the lab. But 3D models already existed everywhere,” Franz says. “There were days where I thought: maybe we're solving a problem nobody actually has.” What changed it was Adidas. They looked at what ALLSIDES could do and said: ‘This is what we need.’ Not ‘this is interesting.’ Not ‘this is cool.’ “Once you have one customer at that level saying yes, actually saying yes with money and not just compliments, everything shifts,” Burkhard adds. “You know you're not crazy.”

Physical AI Needs a First Mile

The shift in how ALLSIDES understood itself happened through Meta. Meta used ALLSIDES to build a digital twin catalogue and the data was immediately picked up as training material by robotics companies and physical AI labs. This is the real danger of the sim-to-real gap: a model trained on physically incorrect data fails the moment it meets an actual object. That gap cannot be patched downstream. The data has to be right before anything else happens.

The vision is to do for 3D data what NVIDIA did for compute, by becoming the underlying layer without which the next generation of AI systems cannot function. The market for AI in robotics alone is projected to exceed $180 billion by 2033. “How to be a platform, not just a tool,” Franz says when asked what they're still figuring out. “How to build something that other people can build on top of. That's the next phase.”

“The market tells you what you're building. If you listen.”
— Franz Tschimben

How a Place Changes What You're Building

Later that day, we close the talks with a cozy dinner in a small pizzeria in Bolzano's old town, the kind where they know the regulars by name. It's late, the tables around us half-empty. I bring it up casually while we're waiting for food.

“You can built this anywhere,” I say. “Why stay?”

Franz smiles. Harald doesn't hesitate.

“I never left. But growing up here, watching companies stay relevant for decades instead of burning out. That teaches you something about patience. About building things that last, not just things that grow fast and disappear later.”

He pauses. “The Silicon Valley optimizes for growth. Here, people optimize for durability and relevance.”

Burkhard nods. “I moved here deliberately. With kids, a life outside work. That changes how you think about what you're actually building.”

Franz adds something quietly: “I left for ten years. Needed to see what ambition actually looks like at scale. But when you come back…” He trails off. “You don't unlearn things. You add to them.”

“So when we started ALLSIDES,” Burkhard continues, “people kept calling with bigger ideas. Faces, environments, generative 3D. We said no. Objects only. Perfect objects.”

I ask why.

“You're not building for a story,” he says simply. “You're building because something needs to exist.”

Harald picks this up: “Most tech companies can't think like that. The pressure is too intense. But there's something about this place that lets you stay focused. You're not competing on hype. You're competing on clarity.”

He grins slightly. “Even if you're opening an office in New York next month.”

The point lands: rooted doesn't mean limited. They're building a global company, but they're building it from a place that keeps them honest.

The wine arrives. We move on to other things: kids, the mountains, how the city has changed. But something stays with me: these three didn't come to South Tyrol to be romantic founders. They came because the place forced them to ask better questions about what they were building.
‍

14 Months Later

As I'm writing this report, I think back to that Berlin conversation. Franz asking questions instead of pitching. Skeptical. Curious in a way that felt almost uncomfortable for a founder.

What strikes me now is that same quality hasn't changed. He's still asking the hard questions. Still skeptical about the narrative. Still uncomfortable with hype.

That matters more than it sounds.

There's a type of founder who becomes convinced of their own story. Starts believing the hype. Starts optimizing for the narrative instead of the problem. I've seen it happen a dozen times. Companies that had real technology, real customers, real traction, and then just… stopped learning and listening.

Franz, Burkhard, and Harald haven't done that. They still talk about what they got wrong. They still adjust when the market tells them something new. They still act like people who are solving a problem, not people who already have all the solutions.

Harald said something lately: “We're not done. We've just started.”

Most founders say that. With these three, I believed it.

‍

Alpine SICAF (Euregio+ & Redstone Adviced) invested in ALLSIDES because the company addresses the first-mile problem of 3D data capture, a structural bottleneck understood for decades, never systematically solved and arrives precisely as physical AI and generative 3D shift from experimental to critical infrastructure.

Redstone is one of the most active European early-stage VCs and holds top decile track record across the sector funds.