Periodic Labs has entered the AI science race with unusually large early funding and a direct ambition: build systems that can help automate scientific discovery. The startup came out of stealth on Tuesday with a $300 million seed round and backing from Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos.
The company’s pitch is centered on AI scientists and autonomous laboratories. Instead of only training models on existing internet data, Periodic Labs wants robots to conduct physical experiments, collect results, iterate on them, and keep improving through repeated work in the lab.
A Founding Team Built Around AI and Materials Science
Periodic Labs was founded by Ekin Dogus Cubuk and Liam Fedus. Their backgrounds help explain why the startup is aiming at the intersection of advanced AI, chemistry, materials science, and robotics.
Cubuk led the materials and chemistry team at Google Brain and DeepMind. One of his projects there was GNoME, an AI tool that discovered over 2 million new crystals in 2023. Researchers say materials like those could one day help power new generations of technology.
Fedus is a former VP of Research at OpenAI and was one of the researchers who helped create ChatGPT. He also led the team that created the first trillion-parameter neural network.
The company’s small team also includes researchers tied to other major AI and materials science efforts. According to the source article, that work ranges from building OpenAI’s agent Operator to contributing to Microsoft’s MatterGen, an LLM materials science discovery AI.
What Periodic Labs Wants Its AI Scientists To Do
The central idea is not simply to use AI as a research assistant. Periodic Labs says it is building AI scientists and the autonomous laboratories where those systems can operate.
In practical terms, that means connecting AI models to lab environments where robots can run physical experiments. The company describes a loop in which robots conduct experiments, collect data, make changes, and try again.
That matters because the company sees scientific AI as needing more than internet-scale text and image data. Periodic Labs argues that the next stage requires fresh data from physical experimentation.
“Until now, scientific AI advances have come from models trained on the internet” and LLMs have “exhausted” the internet as a source that can be consumed, the company says in an introductory blog post. “[A]t Periodic, we are building AI scientists and the autonomous laboratories for them to operate.”
The company’s argument is straightforward: if models need new kinds of data, autonomous labs could produce it directly. Experiments in the physical world would not just search for useful materials; they would also create datasets that future AI systems could use to keep improving.
Why Superconductors Are First
Periodic Labs’ first target is superconductors. The company wants to invent new superconductors that it hopes perform better and possibly require less energy than existing superconducting materials.
The source article does not say that the company has already produced such a material. The goal, as described, is to use automated experimentation to search for materials that could outperform what exists today.
That makes superconductors a fitting first test for the broader Periodic Labs model. The work requires discovery, testing, measurement, and iteration. Those are exactly the tasks the company says it wants its AI scientists and robotic labs to handle.
The startup is not limiting itself to superconductors, however. It also hopes to find other new materials. The first program gives the company a concrete starting point, while the larger platform is meant to support wider materials discovery over time.
The Data Strategy Behind The Lab
One of the most important parts of the Periodic Labs plan is data collection. The company wants to collect the physical-world data its AI scientists produce as they mix, heat, and otherwise manipulate raw materials in the search for something new.
This creates two connected goals. The first is to discover next-generation materials. The second is to generate new experimental data that can be fed back into AI models.
That feedback loop is the core of the company’s strategy. A model proposes or guides experiments, robots perform them, the lab records what happens, and the resulting data becomes part of the system’s future learning process.
If the approach works as intended, the lab is not only a place where discoveries happen. It is also a machine for producing the kind of scientific data that internet-trained models do not already have.
A Crowded But High-Profile Field
Periodic Labs is not alone in pursuing AI scientists or automated chemistry discovery. The source article notes that AI as a tool to automate chemistry discoveries has been a topic of academic research since at least 2023.
Other groups are also working in this area. The field includes tiny startups like Tetsuwan Scientific, nonprofits like Future House, and the University of Toronto’s Acceleration Consortium.
What makes Periodic Labs stand out from the information in the source is the combination of its founding team, its small team’s experience, and the scale of its seed round. With $300 million in early funding and backers from across the technology industry, the company is entering the field with significant resources for an ambitious scientific automation effort.
The stakes are clear. Periodic Labs is betting that AI progress in science will depend not only on better models, but also on labs that can generate the next wave of physical-world evidence. Its first proving ground is superconductors, but its larger claim is that autonomous experimentation can become a new engine for materials discovery.