What are the key takeaways from this Latent Space episode?

Key insights include: **Materials as the foundation layer:** Every technology stack ultimately depends on physical materials — GPUs require novel etching materials, batteries and solar panels (currently ~22% efficiency, theoretically 50% with perovskite layers) are pure materials problems. AI engineers seeking real-world impact should reframe their work toward this foundational layer rather than software-only applications.; **Physics Processing Units (PPU):** Welling frames physical experiments not as validation endpoints but as computational units running in parallel with digital simulations. CuspAI's platform routes cheap computational screening first, eliminates poor candidates, then escalates to expensive experiments — treating nature itself as a fast, programmable co-processor alongside data center compute.; **Platform architecture — generative + multi-fidelity digital twin:** CuspAI's platform pairs a generative candidate model with a multi-scale, multi-fidelity digital twin that filters candidates through progressively expensive evaluation steps. LLM-based agents now autonomously orchestrate literature search and workflow execution, built incrementally by first running workflows manually, then automating piece by piece.

What did Max Welling discuss on Latent Space?

Prof. Max Welling, co-founder of CuspAI, explains how his company uses AI-driven platforms to search the entire space of possible materials — not just known ones — to accelerate solutions for climate change, carbon capture, and the energy transition, combining generative models, digital twins, and high-throughput experimentation. Key topics include: **Materials as the foundation layer:** Every technology stack ultimately depends on physical materials — GPUs require novel etching materials, batteries and solar panels (currently ~22% efficiency, theoretically 50% with perovskite layers) are pure materials problems. AI engineers seeking real-world impact should reframe their work toward this foundational layer rather than software-only applications.; **Physics Processing Units (PPU):** Welling frames physical experiments not as validation endpoints but as computational units running in parallel with digital simulations. CuspAI's platform routes cheap computational screening first, eliminates poor candidates, then escalates to expensive experiments — treating nature itself as a fast, programmable co-processor alongside data center compute..

How long is this episode of Latent Space?

This episode is 33 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

Latent Space

🔬Searching the Space of All Possible Materials — Prof. Max Welling, CuspAI

February 25, 2026

33 min episode · 2 min read

Max Welling

Episode

33 min

Read time

2 min

Topics

Productivity, Relationships, Investing

AI-Generated Summary

Published Feb 25, 2026

Key Takeaways

✓Materials as the foundation layer: Every technology stack ultimately depends on physical materials — GPUs require novel etching materials, batteries and solar panels (currently ~22% efficiency, theoretically 50% with perovskite layers) are pure materials problems. AI engineers seeking real-world impact should reframe their work toward this foundational layer rather than software-only applications.
✓Physics Processing Units (PPU): Welling frames physical experiments not as validation endpoints but as computational units running in parallel with digital simulations. CuspAI's platform routes cheap computational screening first, eliminates poor candidates, then escalates to expensive experiments — treating nature itself as a fast, programmable co-processor alongside data center compute.
✓Platform architecture — generative + multi-fidelity digital twin: CuspAI's platform pairs a generative candidate model with a multi-scale, multi-fidelity digital twin that filters candidates through progressively expensive evaluation steps. LLM-based agents now autonomously orchestrate literature search and workflow execution, built incrementally by first running workflows manually, then automating piece by piece.
✓Equivariance vs. data augmentation trade-off: Hard-coding rotational or permutation symmetry into neural networks reduces training data requirements but can complicate the optimization landscape. When datasets are large, data augmentation sometimes outperforms hard-coded equivariance. The practical rule: use equivariance constraints when data is scarce and the symmetry is exact; prefer augmentation at scale.
✓Industrial partnership as prerequisite for materials breakthroughs: CuspAI only invests in a new material vertical after securing a domain-expert industrial partner, such as their PFAS water-filtration work with Chimera. Each new material class requires retraining models and redesigning experimental setups, so deep partnerships — not one-off contracts — provide the feedback loops needed to reach breakthrough results.

What It Covers

Prof. Max Welling, co-founder of CuspAI, explains how his company uses AI-driven platforms to search the entire space of possible materials — not just known ones — to accelerate solutions for climate change, carbon capture, and the energy transition, combining generative models, digital twins, and high-throughput experimentation.

Key Questions Answered

•Materials as the foundation layer: Every technology stack ultimately depends on physical materials — GPUs require novel etching materials, batteries and solar panels (currently ~22% efficiency, theoretically 50% with perovskite layers) are pure materials problems. AI engineers seeking real-world impact should reframe their work toward this foundational layer rather than software-only applications.
•Physics Processing Units (PPU): Welling frames physical experiments not as validation endpoints but as computational units running in parallel with digital simulations. CuspAI's platform routes cheap computational screening first, eliminates poor candidates, then escalates to expensive experiments — treating nature itself as a fast, programmable co-processor alongside data center compute.
•Platform architecture — generative + multi-fidelity digital twin: CuspAI's platform pairs a generative candidate model with a multi-scale, multi-fidelity digital twin that filters candidates through progressively expensive evaluation steps. LLM-based agents now autonomously orchestrate literature search and workflow execution, built incrementally by first running workflows manually, then automating piece by piece.
•Equivariance vs. data augmentation trade-off: Hard-coding rotational or permutation symmetry into neural networks reduces training data requirements but can complicate the optimization landscape. When datasets are large, data augmentation sometimes outperforms hard-coded equivariance. The practical rule: use equivariance constraints when data is scarce and the symmetry is exact; prefer augmentation at scale.
•Industrial partnership as prerequisite for materials breakthroughs: CuspAI only invests in a new material vertical after securing a domain-expert industrial partner, such as their PFAS water-filtration work with Chimera. Each new material class requires retraining models and redesigning experimental setups, so deep partnerships — not one-off contracts — provide the feedback loops needed to reach breakthrough results.

Notable Moment

Welling argues that reaching two degrees of warming requires not just zeroing emissions by 2050 but then actively removing carbon dioxide for another fifty to one hundred years at roughly half the current emission rate — a problem he describes as entirely unsolved and the core motivation behind founding CuspAI.

Know someone who'd find this useful?

Episode Transcript

I want to think of it as what I would call a a sort of a physics processing unit, like a PPU. Right? Which is you have digital processing units, and then you have physics processing units. So it's basically nature doing computations for you. It's the fastest computer known, possible even. It's a bit hard to program because you have to do all these experiments. It's also quite quite bulky. It's like a very large sort of thing you have to do. But in a way, it is a computation, and that's the way I wanna see it. So I wanna you can do computations in a data center, and then you can ask nature to do some computations. Right? Your interface with nature is a bit more complicated. But then these things will have to seamlessly work together to get to a, you know, a new material that you're interested in. Yeah. It's a pleasure to have Max Voling as a guest today. Max has done so much of his career that I've been so excited about. If you're in the deep learning community, you probably know Max for, his work on variational auto coders, which has literally stood the test of time or officially stood the test of time. If you are a scientist, you probably know him for his, like, binary work on graph neural networks on equivariance. And if you're a material science, you probably know about his new, startup, Cusp AI. Max has a long history doing lots of cool problems. You started in quantum gravity, which is, I think, very different than all of these other things you worked on. The first question for AI engineers and for scientists, what is the thread in how you think about provenance? What is the thread in the type of things which excite you? And how do you, decide what is the next big thing you wanna work on? So it has actually evolved a lot. In my young days, let's breathe, I would just follow what I would find is like super interesting. I have kind of this sensor I think many people have but maybe not really, sort of use very much which is like you get this feeling about getting about very excited about some problem right like it could be you know what's inside of a black hole or what's you know at the boundary of the universe or you know what what is quantum mechanics actually all about and so I followed that basically throughout my career, but I have to say that as you get older, this changes a little bit in a sense that there's a new dimension coming to it and this is impact. Working in three-dimensional quantum gravity, you pretty much guaranteed there's gonna be no impact in what you do relative you know, maybe a few papers, but not under in this world, at this energy scale. As I get closer to retirement, which is fortunately …

Get the full transcript (6,539 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all Latent Space transcripts →