Conversation
Highlights

How Radical AI is Rewriting the Economics of Materials Discovery

The aerospace industry uses materials developed in the 1940s. Defense contractors rely on alloys from the 1970s. Semiconductor manufacturers work with substrates created decades ago. This isn’t nostalgia—it’s economic paralysis.

In a recent episode of BUILDERS, Joseph Krause, Co-Founder and CEO of Radical AI, explained the structural problem: “There are two core problems inside that field today. There is timelines typically see 10, 15 plus years to do a novel discovery and then cost north of 100 million for a single material system.”

When discovery takes 15 years and costs $100 million per material, companies stop discovering. They optimize what exists. And entire industries stagnate—not from lack of interest in innovation, but from broken economics.

Joseph’s company is building scientific superintelligence to solve this, not by accelerating the existing process, but by eliminating the structural bottlenecks that make traditional materials discovery economically irrational.

The Serial Workflow Constraint

When Joseph worked as a materials scientist, the process was brutally linear: “I would make a hypothesis, I’d go read a bunch of publications about this area in my hypothesis, I would refine the hypothesis and I would go into a lab, I would make it, I would characterize it, what did I make? And then I would test it for the application and then come all the way back to the start of that process and go through it again.”

Each iteration required completing every prior step. No parallelization. No concurrent experimentation across different parameters.

Radical AI’s closed-loop autonomous system eliminates this constraint: “When you use a closed loop autonomous system, you don’t need to work in a serial fashion. You can work in a parallel fashion. And so now you can do all those tasks I just described simultaneously.”

This architectural shift compresses months of sequential work into days of parallel execution. Hypothesis generation, synthesis, characterization, and application testing happen concurrently across multiple material candidates.

Capturing Systematically Lost Experimental Data

The deeper inefficiency isn’t just serial workflows—it’s institutional amnesia at industrial scale.

Joseph estimates 90% of experiments fail. That’s normal for discovery. “That is scientific discovery. That is how you get to the 10% of things that do work. We don’t ever capture that data. It Lives in lab notebooks in your brain. You don’t talk about it in publications because it’s all the mistakes.”

The consequences are profound: “If you and I worked on the exact same material problem, I’d have no idea what you did in the lab. And you have no idea what we’ve done here.”

Researchers unknowingly repeat failed experiments. Companies can’t build on prior institutional learning. Even within the same organization, knowledge fragments across individual scientists’ notebooks and memories.

Radical AI’s system captures every experiment—successful or failed—including temperature ranges, pressure variations, precursor combinations, and processing conditions. This data doesn’t just inform future experiments; it trains their AI agents on the full experimental landscape, not just the 10% that “worked.”

Engineering Manufacturability Into Discovery

Even successful lab discoveries typically die in the valley of death between bench-scale synthesis and production-scale manufacturing.

“Fundamental research is different from commercialization in the processing. When you make a material at the lab, you still have to understand how to make that material at scale because the process of that will change the properties that you need for an application,” Joseph explains.

Traditional approaches restart the entire discovery process at manufacturing scale, exploring new questions: What temperature ranges work for 600-ton batches instead of 100-gram samples? How does humidity affect precursor reactions at scale? Do powder precursors behave differently than pellets in industrial reactors?

Radical AI captures this data during discovery: “We’re capturing that data, as I mentioned, and then we have the ranges and the form factors that we can bring to the manufacturing setting. When it comes to temperature, we know the range. When it comes to pressure, we know the range, concentration, the range, even something like humidity in that form factor, how does that impact science?”

Discovery experiments explicitly test processing parameters relevant to manufacturing conditions. Lab-scale data maps directly to production specifications, eliminating the traditional restart cycle.

Selecting Markets Where ML Has Exponential Advantage

Radical AI applied two filters for beachhead selection.

First, technical suitability—problems where their approach has structural advantage: “We want to solve problems that are very challenging to solve for scientists today.”

High entropy alloys fit perfectly. “There are probably 10 to the 40 different potential combinations. If you use the periodic table that would take humans 5, 6 million years to move through all them.”

The search space is intractable for traditional experimentation. ML-driven autonomous systems can screen vast compositional spaces that humans cannot.

Second, market readiness—applications blocked by materials constraints, not lack of demand: “We wanted to pick an area where there were products ready today, but they were prevented from without novel materials. We call this enabling technology, not optimization technology.”

This distinction matters strategically. Optimization technology improves margins on existing products—important but incremental. Enabling technology unlocks entirely new product categories that cannot exist with current materials.

High entropy alloys maintain mechanical properties—strength, corrosion resistance—in extreme environments: 2,000-4,000°F temperatures, high pressures, and oxidative atmospheres. Joseph explains one specific challenge: “When you are flying at Mach 5 speeds in the atmosphere, you actually create a very oxidative environment because you are smashing into atoms in the atmosphere, creating free radicals which attack your system.”

Current materials cannot withstand these conditions. Products like advanced hypersonic systems and next-generation jet turbines exist in design but cannot be built. This creates clear demand from aerospace, defense, automotive, and energy customers.

Running Dual-Track GTM to Derisk Frontier Technology

Radical AI simultaneously pursues government and commercial customers—not as hedging, but as strategic complementarity.

Government partnerships (Department of Defense, Air Force Research Lab, DOE Genesis mission) provide three advantages:

First, access to world-class scientific institutions: “The most advanced scientific institutions in the world are funded by the US government.”

Second, funding for long-horizon applications: “Nuclear fusion…It’s actually very hard to convince investors, talent and the market that nuclear is here. But materials are a massive bottleneck for nuclear fusion reactors today. And so government’s actually the perfect entity to step in and say we’re willing to bridge that gap.”

Third, validation of the technology approach itself.

Commercial customers (aerospace manufacturers, defense primes, automotive companies, energy infrastructure providers) provide near-term product applications, faster revenue cycles, and market feedback on which material properties matter most for deployment.

Joseph views these as converging: “We don’t really care if we’re selling to defense and the US gov or we’re selling to aerospace. We just want to focus on the high entry alloy, application, discovery and scale.”

The technology breakthrough—materials that enable new product categories—applies across both tracks. Government success validates technical capability; commercial success validates market demand. Together, they derisk the long development cycles inherent in deep materials science.

Addressing the Downstream Adoption Barrier

Customer skepticism rarely centers on discovery acceleration. The objection runs deeper.

“We get asked all the time…how do you really ensure, when you move from discovery to manufacturability like you don’t, you can scale that gap, you don’t get caught in the valley of death,” Joseph explains. “I think this is a good concern because a lot of things in AI are CAPEX intensive, for example with GPUs and in science there’s a lot of unknown.”

Prospects understand AI can screen massive search spaces faster. They’ve seen lab discoveries die at scale-up. The valley of death kills most promising materials before they reach production.

Radical AI’s response isn’t a future roadmap—it’s architected into their system from day one. Manufacturing parameters are captured during discovery. Processing data flows directly from autonomous experimentation to production specifications. The gap doesn’t exist because the system was designed to eliminate it.

This transforms customer conversations from “can you scale?” to “show me your manufacturing parameter data for this material.” The question shifts from theoretical concern to technical validation.

Focusing on Technology Breakthroughs Over Market Categories

Joseph’s long-term vision extends beyond specific customer segments: “We just want to be the company who has scaled that system and can sell it to all those components.”

He draws parallels to companies that built transformative technologies rather than serving specific markets: “Look at SpaceX, you look at Tesla, you look at some of the humanoid companies today, or the aerospace companies like Boom and supersonic flight don’t worry per se about the market, they worry about the technology. Because if that technology becomes available, the markets that unfold in front of them are endless.”

This philosophy shapes everything—hiring, partnerships, customer selection. “We believe we are building a company that fundamentally impacts the human race with the output of our technology. If you are focused on a mission and not a job, this is the place for you.”

For deep tech founders, the lesson is clear: when you’re solving problems that broke traditional economics, focus on the technology breakthrough that makes new categories possible, not on optimizing within existing categories. The markets emerge from enabling new capabilities, not from incrementally improving old ones.

Radical AI isn’t just accelerating materials discovery—they’re eliminating the structural bottlenecks that determine which innovations get built and which industries can evolve beyond decades-old materials. In markets frozen by 15-year timelines and $100 million discovery costs, that’s the difference between optimization and transformation.