Why I invest at the intersection of AI and climate

A fusion magnet is a coil bent into a shape you cannot draw freehand. It has to hold a plasma steady, not for a pulse, but for as long as a power plant runs. Helical Fusion, a Japanese company working toward a steady-state helical fusion plant, is doing that, with magnet work validated in collaboration with Japan’s National Institute for Fusion Science. Following that work closely taught me one thing very fast.

The loop is slow. You draw the coil, build the magnet, test it, learn one thing you did not know, and change the drawing.

Every hard-technology company is that loop. The difference between a company that gets somewhere and one that quietly runs out of money is almost never the idea. It is how many turns of the loop they get through before the money is gone.

That is what I buy.

What I actually do

I am an impact angel investor. I write small pre-seed cheques into founders using AI to get through that loop faster, mostly on climate, and on other problems where the world is worse off if nobody solves them.

I am one person writing cheques out of my own money, not a firm. I am early at this, and I have no record to show you yet. I am building one bet at a time, and some of them will be wrong. Anyone who tells you otherwise at this stage is selling you something.

I say it plainly because founders deserve to know what they are getting. A cheque from me is not the thing that saves your company. What I can add is a few doors in Japan, some patience, and someone who has read a lot of bad decks including my own.

Why AI, and why not “AI for climate”

“AI for climate” as a category has become a way of not saying anything. There is software that puts a chat box on a carbon dataset, and there is a materials team that ran its candidate structures in simulation for months before it touched a furnace. Both get the same label. Only one of them changed how fast the company learns.

I look for the second kind. In practice that means simulating before fabricating, so a failed design costs a week instead of a quarter, and reading the whole literature of a field rather than the forty papers a human can hold in their head. The sensor data is always a mess. The question is whether the team built something that makes sense of it, or hired three more analysts.

None of that is a product feature. It shows up in how quickly a team stops being wrong about something. You can see it when you ask a founder what they got wrong last month. The fast ones have a specific answer, because they were wrong recently and it cost them four days.

The lever can be installed

Here is where I part company with a lot of people who use the word AI-native.

Some of the best climate teams I meet do not have this lever yet. A chemist who has spent eleven years on one reaction is not behind. She is holding something you cannot buy, and the loop she runs is slow for physical reasons and also for habitual ones. If the physical part is fixed and the habitual part is not, the lever is installable. That is a yes for me, and often a better yes than the team that already speaks fluent AI and has no depth underneath it.

The hard half is not technical. It is whether the founder can be told something and actually change. I have met brilliant people who cannot, and the cheque does nothing for them. I have no clean test for it. What I do is watch what happens to a suggestion I make in a first meeting by the time of the second.

Where I look

Munich, because I live here and my kids grow up here. You learn more from a founder in one hour across a table than in six video calls. I look at Cambridge as well, where the research sits and too little of it becomes a company anyone can back. And Japan. I have been convening people in Tokyo since 2024, and the industrial buyers who decide whether a climate technology ever gets an order are Japanese as often as they are European.

Those three are where I can be useful, not a claim about where good companies are. Good companies are everywhere. My time is not.

Who I want to hear from

Founders who are not the usual shape. I carry a Korean surname. I grew up a weird kid in Kawasaki and I have spent most of my life being the wrong sort of person in the room I was in. I know what it costs to walk into a meeting where everyone has already decided what you are.

So: women building hard technology, founders from countries that do not appear on anyone’s map of deep tech, immigrant founders, founders who took eight years to get here instead of three. Please write to me. I will not pretend I have solved my own bias by naming it. I have just noticed that the founders who had to fight to get to the meeting tend to be the ones still there in year four.

And it does not have to be climate. Climate is where most of my time goes, because the physics has a deadline. But if you are using this same loop against something else that matters, health, food, water, the way people find work, I want to see it.

The unglamorous part

Most of what I do is not investing. It is reading a deck at eleven at night, telling someone their pilot customer is not a customer, or writing to a person in Tokyo to ask whether a founder is worth thirty minutes of their week.

I get things wrong at a decent rate, and the wrong calls are usually the confident ones.

If you are building one of these companies, send me a paragraph. Not a deck first. One paragraph about what you are trying to do and what turn of the loop you are on right now. I read everything, and I am slower than I want to be.