How Radical AI is building a self-driving materials lab

New York City–based autonomous material-science discovery firm Radical AI, founded in 2024, aims to accelerate materials R&D with the integration of AI, engineering and materials science in its self-driving lab. Its lab can create and characterize more than 25 alloys in a single day.

The company’s AI system screens billions of material compositions to predict structures and physical properties to identify experimental candidates, which are then synthesized and characterized, generating data that is fed back into the prediction engine to close the loop.

Radical AI’s self-driving lab

Radical AI’s approach reflects a broader, industry-wide trend. According to Cypris R&D Intelligence, AI-assisted researchers are generating 44% more material discoveries than traditional methods, and the technology could compress traditional 10-20 year timelines down to 1-2 years in some cases.

“Machine learning is game-changing for materials discovery because it saves scientists from repeating the same process over and over while testing new chemicals and making new materials in the lab,” Kristin Persson, director of the materials project at Berkeley Lab, told Berkeley Lab News Center.

Radical AI’s lab aims to build a fully closed-loop system that goes from discovery to the manufacturing of new materials. The company’s AI system can read publications, make hypotheses, send materials to the lab to be synthesized, characterized and tested, capture and analyze data that the system then uses to design the next loop.

Entering the loop

Once the loop begins, Joseph Krause, CEO, explained, multiple steps happen simultaneously. “So we might be reading publications and coming up with the new experiment as we are characterizing the last thing that we just made and pulling real information out in real time,” he said.

Krause brings materials science expertise from his PhD work at Rice University and time at the Army Research Lab before joining AlleyCorp as an investor. He co-founded Radical AI with Jorge Colindres (President, also from AlleyCorp) and Gerbrand Ceder as chief science officer, who served as a principal investigator at Lawrence Berkeley National Laboratory’s autonomous laboratory.

The company raised a $55 million Seed+ round in July 2025 led by RTX Ventures, with participation from NVIDIA’s NVentures, Eni Next, and AlleyCorp, followed by a reported $60 million Series A.

Competitors like Lila Sciences, which has raised hundreds of millions in funding, and Periodic Labs, founded in 2025 by prominent AI researchers, are building similar “self-driving labs.” Academic institutions have also made strides. Argonne National Laboratory’s Polybot screened 90,000 material combinations in weeks, a task that would typically take months manually, while Berkeley Lab’s A-Lab has successfully synthesized materials predicted by Google DeepMind’s AI models.

The evolving human-machine dynamic

Radical AI’s long-term aim is an end-to-end automated loop, but the lab is not fully hands-off yet: some testing is only semi-autonomous, and scientists still guide analysis and labeling.

While all samples are autonomously synthesized, tensile testing is only semi-autonomous. Other tools, such as SEM, XRD, XRF and oxidation, are fully autonomous.

“That is a really important piece of being able to connect in full flywheel as well. If you can’t pull that data in and understand all of that and analyze that, you’re not really building knowledge. You’re just reading information,” Krause said, “If you can download an image, but you cannot analyze it like a scientist can, you don’t have scientific knowledge. You just have the ability to pull an image into your model. That is where scientists really weigh in today, to continue to train our systems.”

Bottlenecks remain

“The biggest bottleneck in materials science is always processing and manufacturing,” said Krause, “That know-how is very hard to generate and takes a very long time.” The materials science industry’s 10- to 20-year timelines are largely driven by the processing and manufacturing phase.

Radical AI is still tackling this challenge. The company has not yet moved to manufacturability, Krause explained. While the company is keeping this phase in mind, it is still in the discovery and testing phase.

AI in materials science still is something of a work in progress. “How it works today and how we envision it are still somewhat different. There’s just a lot of tool building that needs to be done,” Ceder said in MIT Technology Review.

Krause said the company plans to eventually expand beyond discovery into manufacturing, vertically integrating as a materials supplier.

“If we can vertically integrate and then bring our AI and autonomy to the manufacturing process as well, then we can connect novel discovery and testing of manufacturability in a couple weeks versus the 10 to 15 year process that is today. That’s the exciting opportunity for the technology,” he added.

Filling the data gap

Most experiments fail, but the results of failed experiments are rarely published to be seen by the wider scientific community. This means that scientists at different companies and institutions are likely repeating the same failed experiments others have already conducted. At Radical AI, all data, including that from failed experiments, is recorded, indexed and taken into account when making predictions and hypotheses.

“This ability to capture the know-how and build on top of it is where experimental results is so important,” said Krause, “We don’t do that in typical human science operations today.”

This is also important for training the AI model, Krause explained. “If you can’t stack information so that you can actually learn, then those models are only going to be as good as simulation, which is where the industry is today,” he said.

We don’t just use simulated data. We use real experimental data that we can stack on top of one another so that we continually increase knowledge. That’s why data is so important.

How the Genesis Mission could fit in

Announced by the Trump administration in November, the Genesis Mission is “a dedicated, coordinated national effort to unleash a new age of AI‑accelerated innovation and discovery that can solve the most challenging problems of this century.” The mission will run within the Department of Energy (DOE), run by the Under Secretary for Science Dario Gil.

The goal is to build an “integrated discovery platform” that will be “the world’s most complex and powerful scientific instrument ever built,” according to a DOE press release.

The Genesis Mission will focus on three initiatives, aiming to enable “American energy dominance,” advancing discovery science and ensuring national security. Ultimately, the goal is to “develop an integrated platform that connects the world’s best supercomputers, experimental facilities, AI systems and datasets across every major scientific domain to double the productivity and impact of American research and innovation within a decade,” according to the mission’s website.

On Dec. 20, Radical AI joined the Genesis Mission. “The Genesis mission, in our opinion, is something that reflects the current time in science, shows strong leadership and scientific discipline and also is an opportunity to demonstrate the output advanced science can have,” said Krause. “This is an opportunity to show the American public how strong science is in the United States and show the capability to actually build the most advanced scientific tool ever built, that includes HPC and quantum AI and self driving labs or robotic automation in this single closed loop system. That is exactly what we believe in as a company, and that is why we feel so honored to be selected to be a part of that.”

Where materials R&D could go next

“We think what we’re building is one of the most important companies in the world,” Krause said, “Because the impact can be deeply felt across everyone in the human race. It doesn’t matter what industry you care about, automotive and aerospace, manufacturing and defense, climate, energy, semiconductors, electronics, the most important industries in the world are a direct result for materials, R&D. That is the opportunity that we see, and that is what we get really excited about when we think about building a vertically integrated AI and autonomy driven materials manufacturer and supplier. That is what the industry needs.”

While Radical AI is pitching an end-to-end, AI-driven materials pipeline, the challenges of getting there will take considerable physical validation while navigating an increasingly fierce competitive landscape. The company has yet to tackle the manufacturing bottleneck that drives the industry’s longest timelines, and faces competition from well-funded rivals and established academic labs. But with $55 million in funding, a leadership team including former Berkeley Lab principal investigator Gerbrand Ceder, and selection for the Genesis Mission, Radical AI has positioned itself at the forefront of an industry transformation that could reshape how the world develops new materials.