Inside a midtown Manhattan lab, a robotic arm lifts small glass bottles and carefully measures out pellets of iron and other elements. A furnace melts them into alloys. Other machines analyse composition, test hardness, and measure resistance to heat and oxygen. Nobody designed today’s experiments. The AI did — and it’s been running since 4am.
This is Radical AI’s vision of the future of scientific discovery, and it’s already producing results.
The 20-Year Problem
Developing a new material from scratch — hypothesising, synthesising, testing, iterating — typically takes human scientists two decades or more. It’s a process constrained at every step by the limits of human attention, working hours, and the sheer volume of existing literature no single researcher could ever fully absorb.
The need for new materials has never been greater. Clean energy, fusion reactors, next-generation jet engines, semiconductors, robotics — virtually every frontier technology depends on materials that either don’t yet exist or haven’t been optimised. The gap between what we need and what we have is a materials problem as much as anything else.
One Scientist, Ten Problems at Once
Radical AI’s approach flips the traditional model. Rather than a team of scientists grinding through one challenge at a time, their AI system can read 10,000 scientific papers in five seconds, cross-reference 380,000 published studies and 57 million internal data points, and simultaneously generate hypotheses, run quantum chemistry simulations, and analyse experimental results — in parallel, around the clock.
When a client brings a problem to the lab, the process starts with a specification: here are the properties we need this material to have. The AI takes it from there, considering what has been tried before — including failures, which rarely appear in published literature but are captured in Radical’s own lab database — and proposing anywhere from a dozen to several hundred candidate materials to test.
The physical lab then gets to work. Running nearly autonomously, it can execute up to 50 experiments a day — a pace the company aims to push to 100 by the end of the summer. A human materials scientist, for comparison, might run 50 experiments in an entire year. As each result comes in, the AI updates its models in real time, refining its hypotheses and designing the next round of tests without waiting for human review. If it has a new idea at 4am, the lab simply starts running.
Human scientists remain in the loop, providing qualitative notes — observing cracks in a sample, noting unexpected behaviour — that help the AI develop something closer to scientific intuition over time. The relationship is collaborative, not hands-off, but the ratio has fundamentally shifted: one scientist managing ten problems simultaneously, rather than ten scientists focused on one.
Early Results
The company is still in the process of filing patents and hasn’t disclosed specific details of its discoveries. But in one recent campaign, the system identified around 300 novel material compositions over just 16 weeks. The most promising were sent to Purdue Applied Research Institute for independent verification — and outperformed the leading existing material in their category.
Current focus areas include high-performance alloys for jet engine components, which need to withstand extreme heat while lasting longer and burning fuel more efficiently, as well as materials for defence applications and fusion energy — a field the CEO describes as fundamentally a materials problem, given the extraordinary demands placed on every component inside a reactor.
The Bigger Picture
Radical is not alone. A small but growing cluster of startups is building similar AI-driven autonomous laboratories, each targeting different material classes and applications. One is working on AI-designed materials to remove PFAS — so-called forever chemicals — from water supplies. The space attracted hundreds of millions in venture funding last year, suggesting the investment community has concluded this approach is real.
What makes Radical’s system distinctive, its CEO claims, is the active learning loop — the AI isn’t just designing and testing in sequence, but continuously updating its understanding from live experimental data, making new hypotheses in real time rather than waiting for a batch of results to be processed.
The lab is currently moving to a larger facility at Brooklyn Navy Yard, where expanded automation will push throughput further still.
The underlying ambition is straightforward, if sweeping. Materials sit beneath almost every meaningful technological advance — energy, computing, transportation, medicine. A lab that can discover them in weeks rather than decades isn’t just faster. It’s a different kind of tool entirely.