SuperRadiant, a startup founded by three Brown University researchers with backgrounds in high-energy physics and robotics, surfaced last fall with an ambitious pitch: autonomous systems that can design experiments, execute them physically, and learn from the results without human supervision. The Y Combinator-backed company joins a crowded and well-funded field trying to solve what has become an unexpected bottleneck in the age of powerful AI models capable of predicting novel molecules and materials. The physical validation of those predictions still happens at human speed.
The company's emergence reflects a broader shift. As frontier AI systems grow more sophisticated at generating scientific hypotheses, theConstraintError shifts from idea generation to proof. Laboratory automation has evolved from glorified pipetting robots into something closer to embodied scientific intelligence, and the market opportunity has grown accordingly. The laboratory automation sector was projected to reach $10 billion globally in 2026 and $18.4 billion by 2033, according to Grand View Research. The embodied AI category, which pairs reasoning with physical action, stood at $4.67 billion in 2025 and could hit $67.63 billion by 2033 if current growth trends hold, Grand View Research reported.
Investors and government agencies appear convinced the moment has arrived. In July of last year, the National Science Foundation committed between $380 million and $400 million to launch a national network of AI-programmable cloud laboratories. Carnegie Mellon University's AI Science Foundry, one of the inaugural awardees, connects more than 80 robotically controlled instruments across two facilities into a unified platform, according to the university. Boston University and Stanford-led collaborations each secured $20 million under the same program.
"With frontier models becoming increasingly capable of predicting novel molecules and generating scientific hypotheses, the bottleneck in science is shifting toward the ability to validate those predictions," Michelle Lee, founder and CEO of Medra, wrote last June when announcing a DARPA-funded partnership. Medra, which had raised more than $60 million as of June 2026, opened its ML001 autonomous laboratory in April of last year after a 77-day construction sprint. The company calls its platform an "AI Experimentalist," terminology that has become common across the sector.
An Ecosystem Takes Shape
The landscape includes both pure-play automation providers and hybrid models. Emerald Cloud Lab operates remote life-science facilities in Austin and South San Francisco. Strateos offers automation-as-a-service. Automata's LINQ platform powers what the UK's Royal Marsden NHS Foundation Trust has described as the country's first robotic genomic testing facility for cancer patients, operational since mid-2024 in a regulated clinical environment. Radical AI announced a $55 million "Series Seed+" round in mid-2025 for materials discovery, bringing in Gerbrand Ceder as chief science officer. Recursion, now a public company, runs high-throughput phenomics and reported $213 million in cash inflows from its Roche/Genentech collaboration by 2025, according to SEC filings.
McKinsey identified autonomous experimentation as central to future R&D stacks in a January 2025 analysis. The consulting firm has estimated that generative AI could add $60 billion to $110 billion annually to pharmaceutical and medical-products industries. A subsequent McKinsey report forecast capacity uplifts of 21 to 30 percent in wet labs and data analytics through agentic AI and automation, emphasizing a shift from pilots to broader transformations.
Academic results lend weight to the commercial enthusiasm. Google DeepMind's GNoME system predicted 2.2 million new crystal structures in 2023. Lawrence Berkeley National Laboratory's A-Lab followed by synthesizing 41 new inorganic materials in 17 days of continuous operation, connecting prediction to robotic execution without human intervention, according to Nature papers published in late 2023. The Abolhasani Lab at North Carolina State University has reported self-driving fluidic laboratories for perovskite nanocrystals and doped quantum dots in publications through 2024.
"The promise is that an AI agent will be much, much faster than just relying on human knowledge to explore that experimental space," Milad Abolhasani, the ALCOA Professor at NC State, told Optics & Photonics News in April 2025.
A benchmarking study published in Digital Discovery late last year proposed common "Acceleration Factors" for self-driving labs and found a median speed-up of roughly six times across the literature, with higher gains in higher-dimensional search spaces. The underlying dynamics favor automation: laboratory robotics reached $2.9 billion globally in 2026, up from $2.4 billion in 2023, and could hit $3.9 billion by 2030, Grand View Research reported. The materials informatics market, smaller but faster-growing, has been estimated variously at $134.6 million in 2023 heading toward nearly $400 million by decade's end in one forecast, or closer to $250 million in 2026 en route to $1.61 billion by 2036 in another, according to mid-2026 data from Grand View Research and Research & Markets.
The SuperRadiant Team
SuperRadiant's three founders bring credentials from disparate corners of computational science. Owen Tower, the CEO, holds a PhD in physics from Brown and has worked on nanofabrication and microfluidics with machine learning methods, contributing to research at NASA's Jet Propulsion Laboratory and CERN's Large Hadron Collider, according to the company's Y Combinator profile. Cooper Niu, the CTO and a Brown physics PhD candidate, created ALBERT, an autonomous AI particle physicist that reportedly rediscovered the Standard Model and recovered the top quark mass, predicting 178.9 ± 5.0 GeV from pre-1990 data, work detailed on arXiv in March of last year. Benedict Quartey earned a PhD in computer science from Brown, focusing on generally capable robots and spatial world models, and previously worked at the Boston Dynamics Robotics & AI Institute.
Tower announced the startup's launch and Y Combinator acceptance in a LinkedIn post in mid-September, describing the mission as building "Embodied Scientific Intelligence." The company's website states that SuperRadiant's systems "observe, hypothesize and experiment" to iterate toward discovery, positioning the technology as general-purpose and adaptable rather than fixed high-throughput equipment. The company has not disclosed funding details, customer contracts, or headcount, and declined to provide specifics on current deployment timelines or pilot partnerships.
Precedent and Evolution
The concept of autonomous scientific agents has roots stretching back years. Robot Scientist Adam ran closed-loop hypothesis–experiment–analysis cycles in yeast functional genomics from 2004 to 2009, with results published in Nature and Science. Eve followed from 2012 to 2015 with drug-screening automation, according to papers in Nature and Royal Society Open Science. IBM's RoboRXN, active from 2019 through at least 2024, provided cloud-based autonomous synthesis and language-to-procedure translations, demonstrating early AI experimentalist components, according to IBM Research documentation.
More recent demonstrations have shown greater sophistication. Automata and CellVoyant partnered in January of last year to demonstrate adaptive, AI-driven, closed-loop cell culture that operates autonomously around the clock with real-time protocol adjustments. Follow-up communications cited an "8× improvement" in differentiation efficiency, according to Automata case studies and CellVoyant LinkedIn posts. Medra's ML001 laboratory, operational since April of last year, supports end-to-end experimental design with a "Physical AI lab in the loop," Lee said in the June announcement. The facility integrates with DARPA programs.
Open Questions and Infrastructure Challenges
Standards organizations and federal programs signal infrastructure maturation, though challenges remain. The NSF's Programmable Cloud Laboratories program solicitation, published in August 2025, outlined four-year node operations and plans to recruit new users across disciplines. SiLA 2, an HTTP/2-based interoperability standard for lab devices and software, has advanced through the past few years alongside OPC UA's Laboratory & Analytical Device Standard and integration with Allotrope ontologies, according to standards documentation.
Reviews published in Materials Horizons in March of last year argue next-generation self-driving labs must become generalizable across modalities and reconfigurable with minimal downtime, shifting from point solutions to whole-lab orchestration. Papers in Nature Communications and Nature Reviews Chemistry emphasize safety frameworks, risk matrices, and the need for human-AI collaboration guardrails. The NIST AI Risk Management Framework, released in January 2023 with a generative AI profile in mid-2024, provides voluntary guidance applicable to agents orchestrating experiments.
"Self-driving labs are the new AI asset countries are pursuing in hopes of gaining an economic and security edge," Abolhasani wrote in a LinkedIn post last year. The UK government issued preliminary market engagement documents for "Sovereign AI — Autonomous Labs" programs in November 2025, according to government procurement records.
The laboratory automation and AI-for-science sectors face open questions around validation standards, reproducibility across platforms, and integration with legacy laboratory information management systems. A PRX Energy paper in March 2024 cautioned on unsupervised discovery claims and validation issues in high-throughput inorganic materials prediction, a reminder that the technology remains unproven at scale.
Near-term milestones for the field include results from the first NSF PCL nodes, expected to showcase usage metrics, safety reporting, protocol libraries, and inter-node orchestration. Scaling generalist AI experimentalist stacks from single domains to cross-domain laboratories that span biology, chemistry, and materials science represents the next technical challenge, along with enterprise bridges to existing electronic lab notebooks and data systems via standardized APIs.
"Self-driving labs will serve as collaborators for human researchers, significantly reducing the time and cost required to reach scientific solutions," Abolhasani told Phys.org in an April 2025 interview. Whether venture-backed startups like SuperRadiant, established cloud-lab operators, or academic consortia capture that value remains an open competitive question. The sector is building toward a projected market that could exceed $18 billion by the early 2030s, but the path from prototype to widespread adoption has historically proven longer and more complicated than early enthusiasm suggests.
