TUM.ai and incubating startups differently

The U6 runs north out of Munich until the city thins into research buildings, car parks, and a great deal of sky. Garching is where the Technical University of Munich keeps its engineers and physicists, far enough from the center that the students talk about going into town as though it were a separate country. I had the whole walk from the station to rehearse my questions, and I would end up using almost none of them.

I had come to spend a day with TUM.ai, a student initiative that three separate people had described to me, in almost the same words, as basically running an incubator. I was skeptical. I have seen enough operations like this to expect a slide deck and a foosball table. What I found was closer to a very organized club that happens to grow companies as a side effect of taking its own members seriously.

The initiative is about five years old, which in student time is ancient. It started in 2020, when a handful of them decided that a university this good at machine learning ought to have somewhere for undergraduates to actually do the work rather than read about it. It has since grown to somewhere around ninety active members, sorted into departments with the seriousness of a small firm. One runs the hackathon, one handles partners and sponsors, one writes software, one keeps the legal and finance side honest, and one, the reason for my visit, runs the incubator.

They call it the E-Lab. A member walked me through it as the place where the initiative's most promising founders spend roughly fourteen weeks turning an idea into something with a shape. Equity-free, she said, and I wrote that down twice, because it is the detail that everything else hangs on.

Equity-free, she said, and I wrote that down twice, because it is the detail that everything else hangs on.

To see why that matters you have to picture the thing it refuses to be. The standard accelerator is a machine with a known output. A cohort goes in, takes money in exchange for a slice of the company, spends a season being told to grow faster, and comes out the far end at a demo day, pitching to investors in a darkened room. It works, sometimes. The incentives all point one way, toward the exit, and everyone in the building knows it. The students at TUM.ai take no equity, run no fund, and have nothing to sell when the fourteen weeks are up. That changes what the room is for.

The room, when I finally found it, was a booked seminar space with the heating set slightly too high and a whiteboard someone had failed to erase. Six or seven people were arguing about a pricing model. A supermarket sandwich sat unwrapped and forgotten next to the projector. Nobody was performing for me, which I appreciated, given that I had spent the train ride rehearsing intelligent questions I would now get to ask none of.

One of the founders talked me through what he was building. I will not describe it in detail, because it was early and half of it will have changed by the time you read this. What struck me was how the others responded. They poked at it, and they were not gentle. A woman across the table kept coming back to one assumption she thought was wrong, and would not let him wave it away with a bigger number. There was no investor in the room to impress, so the only thing worth doing was to be right.

※

Much of this feeds off the initiative's louder event, the Makeathon, a three-day hackathon they run once a year. Something like five hundred students and young professionals turn up to build against real problems that real companies bring in. I did not see it. I was told, more than once and with a certain fondness, that people barely sleep. It works as a wide mouth at the top of everything. People come for a weekend of building, some of them stay, drift toward the venture side, and end up in the seminar room with the terrible coffee two years later.

I kept asking who was in charge, and kept getting slightly different answers. This is because everyone graduates. The whole apparatus is run by people who are, by design, on their way out. A leader this year is an alumnus the next, and the thing has to be rebuilt from memory and handover documents every year or two. I found this alarming. They seemed to find it obvious. One of them pointed out that a company built to survive that sort of turnover is probably built rather well.

They are not hostile to money. The venture team is glad to introduce founders to the actual investors around Munich, and TUM as a whole has a serious startup apparatus of its own. But the students are not the ones holding the cheque, and that gap, small on paper, seems to be the whole point. They can afford to tell a founder the idea is not working, because their own outcome does not depend on the founder pretending otherwise.

I left before they were finished. The last U6 into the city is not truly late, but late enough that I did not want to walk that unlit road to find out. From the platform I could still see the seminar room, one lit window in a long dark building, the pricing argument grinding on without me. I never asked the questions I had rehearsed. It did not seem to matter much. A few dozen students were up there working on companies they will never own a piece of, and the train came, and I got on it.