0:00
/
0:00
Transcript

AI in Materials Science with Henrik Hagemann

Lessons from AlphaFold, what AI means for materials science and why more AI use means massively more forever chemicals in the environment.

Thank you to everyone who tuned into my live video with

, founder of Puraffinity, a company that removes forever chemicals from water.

Here are a few points that stand out for me:

  1. Every cutting-edge GPU fab uses a huge amount of virgin PFAS; the chip boom is literally built on forever chemicals.

  2. Graduate-track openings “have fallen off a cliff – the first question is: can an agent do it instead?

  3. It’s easier to take a materials-science PhD and up-skill them in AI than the reverse.

  4. Today Henrik would build the same company with four to eight people instead of 200. (Donkeycorns, my favourite topic!)

  5. Gen-AI is “a new continent of PhD-level labour that costs 25 ¢ per hour and works 24/7.”

  6. Henrik’s moonshot vision? A $400 M full-stack AlphaFold-for-Materials to wipe out pollution – in-house synthesis, testing, scale-up & regulatory.

  7. Real bottleneck in materials science isn’t AI, it’s testing. Real-world validation is now the gating factor: four weeks to three months to see if a material actually works.

My reflections

Two things stand out for me here. One is the trend I see everywhere: progress requires fewer people and more compute. Great for progress, less great for people who never get hired. Also, while everyone seem to have a preference for hiring experienced people if they are hiring at all, where will the experienced people come from later if juniors aren’t getting jobs?

Another is that AI in materials science holds an incredible promise and at the same time creates new problems. Again: great new materials, but to make them we’ll have to use up water and make even more forever chemicals to make AI chips.

Someone said, I forgot who, that the defining characteristic of our capitalism is that we’re trying to make money to insulate ourselves from problems we created by making money. It feels this way with AI: we badly need to keep making tech and scientific progress to solve our current problems and yet it’s the pursuit of this progress is what largely got us into this situation in the first place.

When this pattern shows up on an individual level, the right but painful decision is often to cut the losses, accept the pain and start the journey back to normality, but it doesn’t seem remotely possible for a civilisation. And I’m not sure it’d be a good idea: personally, I’ll take the world with PFAS, AI and capitalism over the times when there was no pain relief for surgeries and life expectancy was 30-something years.

P.S. Today I learned that to invite the guest from the audience to join us on stage, they need to be on mobile, not laptop, so if you’d like to join a future conversation, make sure you join a future Substack Live on your smartphone.

Timestamps

00:00 Welcome

01:18 Henrik’s path: Denmark → China → Imperial College → Puraffinity

03:06 State-of-the-art in AI × materials science: sparse data, non-linear physics, the Alchemy consortium

05:53 Why scale-up kills battery & materials start-ups: manufacturing reality vs. in-silico dreams

08:49 Case study – removing PFAS from water: 4-week to 3-month test cycles as the new bottleneck

11:33 The next 3-5 yrs: better hybrid models, not just more compute

13:30 Hidden cost of AI boom: every prompt guzzles water & chips need “tons of PFAS”

16:14 “Vibe-coding” materials & building full-stack labs with <10 people

18:46 How Henrik would found “Puraffinity 2.0” today – 1/10th the head-count, 1/3 the capital

22:00 The perception gap: why half the world shrugs at GPT-4-level tools

24:00 $400 M moonshot: an “AlphaFold for pollution”

26:22 Junior hiring is broken; why sector experts → tech beats techies → sector

29:33 Fractional careers, representation & the risk of widening gaps

32:30 Funding squeeze & the new bar for product revenue

Get more from Evgeny Shadchnev in the Substack app
Available for iOS and Android