AI is rapidly transforming every industry, and deep tech is no exception. In this episode, we sit down with Joseph Krause, co-founder and CEO of Radical AI, a company poised to revolutionize how we discover and develop new materials. Radical AI is building a self-driving lab where AI autonomously designs, tests, and discovers materials, accelerating R&D at an unprecedented scale. This groundbreaking approach has attracted a historic $55M seed round from major investors like NVIDIA and Raytheon. Joseph, a US Army National Guard veteran with a PhD in materials science, shares his unique journey from military service and academia to deep tech investing and finally, entrepreneurship. We dive into Radical AI’s “materials flywheel” concept, how they are increasing the speed of discovery by 370x, and why New York City is the perfect place to build a world-changing company.
→ Enjoy this conversation with Joseph Krause, on Spotify, Apple, or YouTube.
→ Subscribe to our newsletter and never miss an update.
Nataraj: You have a diverse background from materials science, working in the Army National Guard, and a little bit of venture capital at AlleyCorp, who I’m assuming is also an investor in Radical AI. How did these different experiences converge and lead to starting Radical AI?
Joseph Krause: Great question. I always love the Steve Jobs quote about how you can never connect the dots looking forward, only looking backward. For myself, I very much feel this is true today. Going back to when I was in graduate school, I was pursuing a materials science PhD at Rice University. I was working on a lot of what we call functional materials, or materials that can be used in real applications, and I was frustrated with my lack of ability to push things into industry. That actually isn’t poor performance from academia; it’s just not what academia is truly focused on. Academia’s job is to drive our fundamental understanding of science. For me, I wanted to take that fundamental understanding, work on applied applications and materials, and really put them into products in the world.
At a similar time as I was going through my PhD, I was separately serving in the US Army National Guard. It’s a part-time military service where you can get called up into active duty if needed. I’ve always been a big fan of the military and wanted to serve, so it was a very important thing for me. I had this opportunity come up from the Army Research Lab, which is the US Army’s corporate development research lab, where they were doing materials research but thinking about problems for the Army and how these things can impact the future warfighter. I got really interested in doing science that I care about the impact of. I’m separately serving in the National Guard and I care about the warfighter having the best technology. So I took a journeyman fellowship at Army Research Lab, working on neuromorphic computing.
Even here, it’s still fundamental because it’s at the core, cutting edge of research. I still felt this lack of ability to execute on translation and pushing things out of research into the market. So I took a leave of absence from my PhD, joined AlleyCorp, which is a VC based here in New York City, and I started investing in material science, semiconductors, 3D printing, and informatics to understand how to commercialize materials. While at AlleyCorp, me and one of my other co-founders, Jorge Colindres, came up with this inception for Radical AI. We found our third co-founder, Gerbrand Ceder, and we rolled out and started the company from there. Looking back, it’s a very linear trajectory, but at the time, I had no idea what the next step was.
Nataraj: Looking back, what were the traditional ways of doing things that were preventing this commercialization of new materials or discoveries?
Joseph Krause: I think two things. One is focus. The primary purpose of academic research is not to commercialize applications. It can be an outcome, which is why there are tech transfer offices, but it’s not the core driver. The driver is discovering new fundamental phenomena.
Number two is that the commercialization of science typically ends up in a product. If you think about semiconductor transistor technology, the materials research behind that goes into a chip which is then put in advanced packaging so we can use it in a smartphone or a computer. If you work on transistor technology, you’re still a couple of steps removed from the product that will actually utilize those materials. That disconnect is hard to bridge. Because you don’t have this focus and you’re not building the end product, you can’t really solve commercialized problems. The ones that do, like nuclear fusion, take that scientific invention and start using it in a real product, like a reactor. Bridging that gap, in my opinion, really requires private enterprise. That’s why you’ve seen companies from fusion to space with SpaceX and Blue Origin building private companies on really advanced scientific capabilities.
Nataraj: When you have a good founding team and some funds, what’s the next step? If it’s an Uber for something, I would go and create a new app. What was the product that Radical AI created first?
Joseph Krause: The most important thing to think about is technology, and then from that technology, what is the product you develop? We create new materials, and we want to commercialize those materials by actually selling them at scale. The area that we are working in is very much in aerospace, defense, and energy, where we use these structural metals to build components. Our product, the output of our technology, is novel materials that enable new applications.
The technology piece is how we do that. I love using SpaceX as an example. They focus on bringing things to orbit and eventually getting to Mars. The way they do that is with advanced rocket capabilities that are reusable and more cost-effective. That technology is the means of how they get things into space. If our end goal is to develop incredible materials that enable new applications, the way we do that is our materials flywheel. This is where our AI modeling, engine, and fully robotic self-driving lab come in. They allow us to do materials discovery 370 times faster than a human scientist, bring in different data sets like patents or scientific literature to make new hypotheses, and build a high-throughput capability where we test tens of thousands of materials per year, not just a few. When you bring all those together, you have this materials flywheel that allows you to very quickly discover novel materials and then go into applications that will directly use them for new advancements.
Nataraj: Who are the typical customers looking for new materials? And secondly, you mentioned a 370x speed advantage. Where in the flywheel does that come from?
Joseph Krause: Great questions. First, who buys materials? My favorite question to answer, because it’s every single company on earth. It doesn’t matter if you are in aerospace, automotive, manufacturing, defense, climate, energy, semiconductors, or even athletic apparel. Every single industry in the world is a direct result of novel material advancement. But all of them feel this problem of very long timelines—typically 10 to 20 years—and incredible cost, north of a hundred million for a single material system. So you arguably have one of the largest markets in the world, we just cannot execute on it fast enough.
On the second piece, digital research is challenging in materials, and we cannot only do digital-based research. So much of the know-how in making a material at scale is not just how to make it, but how to make it in large quantities with the properties you need—what we call the processing of a material. Most of the IP of the biggest material companies is trade secrets around processing. So if you never make it in a lab, see its properties for an application, and understand the conditions to scale it—temperature, pressure, oxidation, concentration—then you truly can’t capture the value. We do a lot on the digital side with machine learning and generative technology for inverse design, but at the end of the day, we still need to make that material in a lab. This gap is the hardest thing to understand and where all the value sits in the materials industry today.
Nataraj: How do you pick which direction to pursue? Are you only going in directions where you have pre-commitments from customers, or do you have your own ideas of what new material to find?
Joseph Krause: Great question, because focus is very important for an early-stage startup. It’s actually both. The classes of materials we work in are directly driven by customers. We do not want to randomly work on materials with no identified customer problem. That’s the first thing. The second reason it’s both is that it’s up to us to come up with the new material to solve those problems. A perfect example is our work in the hypersonic space. Hypersonic missiles have a problem where the alloys they use today cannot keep up with the composites inside. The problem is how to drive higher heat and better mechanical performance at high temperatures. That problem is from the end customer, but the solution we’ve come up with is a high entropy alloy. We are designing novel high entropy alloys for this application. We take the problem from an end customer, and then we think about what material system can solve that problem in a way that creates a huge market opportunity.
There’s a third option, which is something like a room-temperature superconductor. The industries that would benefit are endless. There’s not really a market for that today, with customers saying, “I want to solve this in the next six months.” But if you had one, every company would be interested. We call those enabling technologies. However, focus is key. Radical AI needs to show the world that we have the best scientific discovery flywheel and can drive real commercial value. Once we do that, then we can focus on more enabling things like a room-temperature superconductor.
Nataraj: A couple of years back, there was a viral paper about reproducing a superconductor at room temperature. Are you talking about a similar thing?
Joseph Krause: Yes, LK-99 specifically. And that’s a perfect example of where experimental validation of a material is very important. This paper came out claiming room-temperature superconducting ability. People ran simulations to try to confirm it, and some thought they had. But when it came to reproducibility and experimentation, we could not replicate that material. It’s a perfect example of where experimental work is equally as important as simulation work.
Nataraj: What type of people are you hiring? And what type of AI are you using? Are you building your own models?
Joseph Krause: The people we hire span five technology buckets: machine learning and AI, software engineering, automation/robotics, mechanical engineering, and materials science. Building an interdisciplinary team is imperative. On the AI front, we use a multitude of different AI technologies in a multimodal approach. We have an agentic system that sits at the top layer using LLM technology. We don’t build LLMs from scratch; we fine-tune models from the big AI labs. Those agents have access to different tools we’ve built, like graph neural nets and generative-based models using diffusion for inverse design. We also use older AI like computer vision in our lab to track experiments and Bayesian optimization in our design of experiments loop. It’s a mix of technology across the board, from cutting-edge LLMs to well-understood architectures.
Nataraj: DeepMind has AlphaFold for predicting protein structures. Is there a similar opportunity in materials for predicting crystal structures?
Joseph Krause: Absolutely, that’s a very real thing and it’s being deeply explored. We do that internally with our own models. Big Tech also has models: Microsoft has MatterGen, Google has GNoME, and Meta has models from their Open Catalyst Project. The field is growing rapidly. There’s been some pushback that some predictions are not novel or not valid—meaning we can’t actually make them in the lab. These constraints of novelty and validity are really important. A big problem with these models is a lack of experimental data; they are trained almost exclusively on computational data. You need the ground truth from the experimental lab. The problem is that 90% of the work a scientist does doesn’t work, and we don’t capture that negative data. It lives in a scientist’s head or lab notebook. That’s why we build self-driving labs. We capture every single data point and feed that experimental data back into our AI engine to build the most proprietary dataset in the world.
Nataraj: Why pick New York?
Joseph Krause: New York is one of the capitals of the world. We love it for two reasons. One, the talent density is very high and continues to grow. There are not many elite performers who would not want to move to New York City. Two, if you want to work on advanced scientific discovery and deep technology, there are not many places to do that in New York. We might be one of the only places where you can do AI and science. If you want to be in New York and work on really hard problems, Radical is one of the few places you can go. Those two reasons are linked, and that’s why New York is super important to us.
Nataraj: What is your best AI use case for you personally or for managing your company?
Joseph Krause: For me personally, it’s research and knowledge gain. I might talk with a customer who has a very specific materials problem in their parts, for example, in the automotive industry. I’m a material scientist, not an automotive expert. I will use AI to take what would be a three or four-month learning process and get up to speed very quickly on the history of materials used in that application, the research areas pursued, and the current state of the art. I can use AI to get knowledgeable enough in an hour. It’s for longer, extended research settings where I need quick, actionable information that I can learn today and use this afternoon.
Nataraj: And are you using ChatGPT, Deep Research, or something else?
Joseph Krause: To be honest, I use all of them because I’m still in the discovery phase of seeing who’s the best. I’ve used a lot of Grok from xAI; for science, they seem very good and pull references well. I use ChatGPT a lot for idea generation. I will also use Gemini in direct comparison. Sometimes I’ll have both open in a tab, paste the same prompt in both, and see what different information I can pull. I don’t have a single winner yet.
Nataraj: Every time I see Apple’s new demo, they talk about redesigning every material. Where does that material research happen?
Joseph Krause: Apple is an incredible materials science company. A majority of their advancements come from new materials. The battery life, the colors, the aluminum and titanium they use—it’s all a materials science problem, down to the semiconductor technology. They do research internally; they have a very good materials research team. They also work with outside parties. Corning with Gorilla Glass is the one most people know. That was a material discovered 20 years before it had an application, until Steve Jobs came knocking and needed a scratch-resistant glass for a touchscreen. Apple is very good at indexing information quickly and pushing material advancements.
Joseph Krause, co-founder and CEO of Radical AI, joins the Startup Project to discuss how his company is revolutionizing materials R&D. He shares how integrating AI, robotics, and engineering into a “materials flywheel” accelerates discovery by 370x, attracting a historic $55M seed round from investors like NVIDIA and Raytheon. Joseph details the journey from his PhD and military service to building a world-changing deep tech company in New York City.
Joseph’s vision for Radical AI highlights the immense potential of integrating AI and robotics into fundamental scientific research. By creating a closed-loop, autonomous system for materials discovery, his team is not just building a company—they are building a platform to solve some of the world’s most critical challenges across aerospace, defense, and energy.
→ If you enjoyed this conversation with Joseph Krause, listen to the full episode here on Spotify, Apple, or YouTube.
→ Subscribe to ourNewsletter and never miss an update.
