Founded by former OpenAI and DeepMind researchers, Periodic Labs aims to revolutionize materials science through autonomous labs and AI-powered experimentation.
Out of Stealth, Into the Spotlight
Periodic Labs emerged from stealth mode this week with a staggering $300 million seed round, making it one of the best-funded scientific AI startups ever. Backers include Andreessen Horowitz, Nvidia, DST, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos—a lineup that underscores investor confidence in the company’s ambitious mission: automate science.
- The funding gives Periodic Labs a rare advantage—deep pockets and deep expertise—from day one.
- The startup joins a growing wave of AI-first labs aiming to transform scientific R&D through automation.
Meet the Founders: Proven AI Pioneers
Founded by Ekin Dogus Cubuk and Liam Fedus, the startup brings together top minds from the intersection of AI and materials science.
- Cubuk previously led materials and chemistry research at Google Brain and DeepMind, including work on GNoME, the AI tool that discovered over 2 million new crystals in 2023.
- Fedus, a former VP of Research at OpenAI, co-created ChatGPT and led efforts behind the first trillion-parameter neural network.
Their team also includes talent from Microsoft’s MatterGen, OpenAI’s Operator, and other advanced AI-for-science efforts.
Vision: AI That Experiments and Learns in the Real World
Periodic Labs doesn’t just want to simulate science—it wants to perform it. The company’s core thesis is to build “AI scientists”—systems that conduct real-world experiments, gather data, and iterate autonomously.
- These labs will be equipped with robotics and physical instrumentation to conduct experiments on materials like superconductors.
- As experiments unfold, the AI will collect real-world data, adjust variables, and try new configurations—essentially learning by doing.
This marks a shift from traditional AI models, which are trained almost entirely on internet data. Periodic aims to generate new scientific knowledge, not just remix existing information.
Focus on Superconductors—For Now
The company’s first scientific target is the discovery of next-generation superconductors—materials that conduct electricity with zero resistance.
- Current superconductors often require extremely low temperatures, making them impractical for wide use.
- Periodic Labs hopes to discover higher-performance materials that are cheaper, more stable, and energy-efficient.
However, superconductors are just the beginning. The ultimate goal is to uncover new classes of materials that could underpin everything from clean energy to quantum computing.
A New Frontier in AI: Physical World Data
Periodic Labs’ approach centers on producing fresh, proprietary data through experimentation. Unlike traditional LLMs, which are limited by the finite information on the internet, Periodic’s AI will generate its own training data through observation and trial.
- “The internet has been exhausted,” the team wrote in their launch blog. “We’re building the AI scientists to create what comes next.”
- The vision is to create a self-evolving scientific engine, where data creation and AI training are part of the same loop.
This could not only lead to novel discoveries, but also advance the capabilities of AI itself by grounding it in physical science.
The Bigger Picture: A Race to Automate Research
Periodic Labs isn’t alone in this vision. Other startups like Tetsuwan Scientific, nonprofits like Future House, and academic hubs such as the University of Toronto’s Acceleration Consortium are also exploring the intersection of AI, automation, and materials discovery.
- However, with $300M in funding and a top-tier team, Periodic Labs may now be the best-positioned player in the space.
- If successful, it could drastically accelerate the pace of innovation, making scientific discovery not just faster—but smarter.
