Artificial intelligence has moved from the digital realm into the physical laboratory, autonomously conducting chemistry experiments and designing new materials with minimal human oversight. Researchers at Princeton University have demonstrated an AI-powered robotic system called Qumus that can create graphene and fabricate quantum devices entirely on its own, while startups and academic teams are racing to build the next generation of "AI scientists" capable of accelerating materials discovery.

How Are AI Systems Actually Running Lab Experiments?

The breakthrough lies in combining large language models (LLMs), which are AI systems trained on vast amounts of text to understand and generate human language, with robotics, computer vision, and automated laboratory equipment. Qumus operates like a small research team, with a lead AI agent orchestrating the work while specialized sub-agents handle tasks such as project planning, device design, and physical processing.

The physical setup includes robotic arms, microscope systems, temperature-controlled stages, and machine-vision systems capable of identifying microscopic material flakes. When a human user asked the system "Can you give me a graphene flake?" Qumus interpreted the request, checked its database, and autonomously carried out exfoliation and flake-search procedures until it produced a graphene sample. The only human involvement required was supplying raw materials and electricity.

What makes this particularly striking is the system's ability to recover from unexpected errors. During one experiment, a researcher intentionally removed a chip that Qumus was actively processing. The system detected the problem using computer vision, confirmed the chip was missing, and generated a new plan to restart the experiment. In another instance, when an AI model incorrectly labeled a material as graphene instead of hexagonal boron nitride (hBN), Qumus identified the inconsistency and restarted the process until it successfully produced the requested material.

Why Does This Matter for Climate and Energy Technologies?

The traditional path to discovering new materials is painfully slow. Metal-organic frameworks, materials that won the 2025 Nobel Prize in Chemistry for their potential to deliver clean energy, clean water, and clean air, required 20 years of research and over 120,000 synthesized variations to produce just one material that works for carbon capture, and only under specific conditions.

AI-driven materials discovery could compress this timeline dramatically. Mohamad Moosavi, an assistant professor in chemical engineering at the University of Toronto and Vector Institute Faculty Member, explained the fundamental shift: "With deep learning, we can make molecules differentiable. This is pretty exciting because it's for the first time in the history of human beings that we can deal with chemicals, molecules, and materials as a continuous variable".

"With deep learning, we can make molecules differentiable. This is pretty exciting because it's for the first time in the history of human beings that we can deal with chemicals, molecules, and materials as a continuous variable," said Mohamad Moosavi, Assistant Professor of Chemical Engineering at the University of Toronto.

Mohamad Moosavi, Assistant Professor of Chemical Engineering, University of Toronto

Traditionally, materials exist as categorical variables, discrete entities that engineers struggle to optimize. Deep learning changes this by creating mathematical representations of molecular structures, allowing researchers to treat materials as continuous variables for the first time. This means they can apply powerful optimization techniques from engineering to design materials, chemical processes, and devices simultaneously rather than sequentially.

What Are the Key Capabilities of Current AI Laboratory Systems?

• Autonomous Experimentation: Qumus can receive natural-language requests, design experimental workflows, operate lab hardware, analyze results, and generate reports with little or no human intervention.
• Complex Device Fabrication: The system successfully fabricated a graphene transistor by designing a multilayer device architecture, searching material inventory, generating device layouts, and performing dry-transfer stacking in approximately 90 minutes involving roughly 30 procedural steps and 18 decision-making calls.
• Iterative Optimization: When tasked with creating a graphene flake larger than 200 square micrometers with no prior experimental history, Qumus independently explored fabrication parameters including substrate temperature, heating time, massage cycles, and tape peel-off speed, eventually succeeding after several iterative runs spanning more than four hours.
• Multi-Model Flexibility: The system was tested with large language models from OpenAI, Google, Anthropic, xAI, Alibaba, and DeepSeek, with each model exhibiting different behavioral tendencies in terms of caution, efficiency, and willingness to act quickly.

How Much Investment Is Flowing Into AI-Driven Materials Discovery?

The commercial interest in AI-powered scientific discovery is intensifying. Periodic Labs, a startup founded by Liam Fedus, the former vice president of research at OpenAI, and Ekin Dogus Cubuk, a former research scientist at Google DeepMind, is in advanced talks to raise at least $500 million in a new funding round that could value the company at $7.5 billion.

This represents a dramatic rise for the San Francisco-based startup, which was founded less than a year ago and focuses on using artificial intelligence and automated laboratories to accelerate scientific discovery in physics and chemistry. If completed at the reported valuation, the company's value would rise nearly six times from the $1.3 billion valuation it received during its $300 million seed round announced in September 2025.

Investor demand has been extremely strong, with the funding round reportedly oversubscribed and discussions already taking place around an additional funding round at an even higher valuation. The funding round is expected to be led by AMP, an investment vehicle created by former Andreessen Horowitz general partner Anjney Midha.

Periodic Labs is developing what it describes as an AI scientist capable of conducting scientific experiments through autonomous robotic laboratories. The company's technology combines artificial intelligence models with automated lab systems designed to run thousands of experiments in areas such as chemistry and physics. One of the company's current research areas involves searching for new superconductors capable of operating at higher temperatures, materials that could eventually improve energy systems, electronics, and industrial technologies.

What Challenges Remain Before AI Labs Can Scale?

Despite these advances, researchers acknowledge that current systems face significant constraints. The Qumus platform remains constrained by hardware speed rather than AI reasoning, meaning much of the system's total runtime is consumed by physical processes such as mechanical exfoliation and microscope inspection rather than computational thinking.

Additionally, the field is still in early stages. Periodic Labs has already attracted high-profile talent from major AI companies, with reports indicating the company has hired more than 20 researchers from firms including Meta, OpenAI, and DeepMind, with some employees reportedly leaving large compensation packages to join the venture. This talent concentration suggests the field is moving quickly, but also that expertise remains concentrated among a small number of organizations.

The convergence of strong scientific talent, commitment to sustainability, and advanced AI capabilities is positioning certain regions as destinations for this innovation. Moosavi noted that Canada has a unique opportunity: "We have the opportunity to generate new technologies in Canada because we have the best scientists, we have the best engineers, and we have a society that is keen and interested in sustainability. That's pretty rare and also a huge opportunity for us, as researchers at Vector Institute, to catalyze this innovation".

Moosavi

"We have the opportunity to generate new technologies in Canada because we have the best scientists, we have the best engineers, and we have a society that is keen and interested in sustainability," noted Mohamad Moosavi.

Mohamad Moosavi, Assistant Professor of Chemical Engineering, University of Toronto

The shift toward AI-driven materials discovery reflects a broader recognition that artificial intelligence's most transformative applications may lie not in consumer applications but in accelerating scientific research itself. As these systems mature and hardware constraints are overcome, the timeline for developing breakthrough materials for climate solutions, quantum computing, and advanced electronics could compress from decades to years or even months.