Where AI Meets the Future of Science
The worlds of artificial intelligence and experimental science are colliding in an unprecedented way — and leading that charge is Periodic Labs, a startup co-founded by two of the brightest minds in machine learning: Liam Fedus, a respected researcher from OpenAI, and Ekin Dogus Cubuk, a materials science expert from Google Brain.
Emerging from stealth mode with a staggering $300 million seed round, Periodic Labs has already sent shockwaves through Silicon Valley’s venture ecosystem. The startup’s mission? To transform the way humanity conducts scientific research by integrating AI reasoning, robotics, and material simulation into a single, automated discovery engine.
In short, Periodic Labs represents the next frontier of AI scientific discovery — a vision that could accelerate breakthroughs in physics, chemistry, and energy faster than ever imagined.
The Genesis: From Late-Night Conversations to a $300 Million Dream
The idea for Periodic Labs was born in a casual conversation between Fedus and Cubuk about seven months ago. Both researchers shared a sense that the pieces of a long-standing puzzle were finally falling into place: AI models, experimental automation, and material science simulations had each matured enough to be woven into a single, unified system.
“There are a few things that happened in the LLM field, in experimental science and in simulations that kind of made this the right time,” Cubuk explained in an interview.
Indeed, a perfect storm had arrived:
- LLMs (large language models) gained reasoning capabilities that allowed them to make complex, structured decisions.
- Robotic arms became precise enough to handle delicate material synthesis tasks.
- AI-driven simulations could finally model molecular and physical systems with astonishing accuracy.
With these ingredients, the founders envisioned a lab where AI doesn’t just analyze data — it actively conducts science.
The Vision: Automating Scientific Discovery Through AI
The core of Periodic Labs’ mission lies in building what Fedus calls “a closed-loop system of scientific reasoning.” Imagine this workflow:
- An AI model generates a hypothesis or proposes a new material compound.
- A simulation engine tests that hypothesis in silico (digitally).
- A robotic system in a physical lab executes the experiment.
- The AI observes the outcome, analyzes the data, and refines its future predictions.
This continuous feedback loop — powered by large language models and advanced automation — could compress decades of experimentation into months, revolutionizing how science is done.
“Making contact with reality, bringing experiments into the [AI] loop — we feel like this is the next frontier,” said Fedus.
The approach could lead to breakthroughs in fields like superconductivity, renewable energy, pharmaceuticals, and materials engineering — industries where traditional trial-and-error methods are slow and costly.
Scientific Roots: From Google Brain to AI-Powered Labs
Periodic Labs isn’t building from scratch. Cubuk was among the researchers behind a 2023 Google Brain project that successfully used AI models to suggest new material recipes. Working alongside robotic systems, the experiment produced 41 previously unknown chemical compounds — an early glimpse into the promise of AI-driven lab automation.
The key insight? Even failed experiments generated invaluable data for model training. Unlike traditional science, where negative results often go unpublished, AI thrives on every outcome, good or bad. This data-centric philosophy could reshape the incentive structure of modern research, prioritizing exploration over perfection.
The Funding Frenzy: How $300 Million Came Together Overnight
When news of Fedus’s departure from OpenAI broke, it triggered what insiders called a “VC feeding frenzy.” Dozens of top-tier investors competed for a chance to fund the next big leap in AI.
Within days, Felicis Ventures, led by former OpenAI executive Peter Deng, secured the deal. Deng recalls the moment vividly — a brisk walk through San Francisco’s Noe Valley where Fedus laid out his vision.
“He told me that everyone talks about doing science, but in order to do science, you actually have to do science,” Deng recalled.
That simple but profound statement sealed the partnership. Felicis quickly led the $300 million seed round, joined by a lineup of heavyweight angels and venture firms. Ironically, OpenAI itself did not invest, but it didn’t need to — Periodic Labs had all the momentum it needed.
The Team: A Fusion of AI, Physics, and Vision
Armed with their massive funding, the founders have assembled what can only be described as a dream team of scientific and AI talent. Among the early hires:
- Alexandre Passos — co-creator of OpenAI’s o1 and o3 models.
- Eric Toberer — a leading materials scientist behind major superconductor discoveries.
- Matt Horton — developer of Microsoft’s GenAI materials science tools.
The company’s structure is intentionally multidisciplinary. Every week, team members host graduate-level lectures to cross-train one another — ensuring that AI researchers understand materials physics, and scientists grasp the computational principles driving AI models.
This culture of deep interdisciplinary collaboration is what Fedus calls the “tight coupling” that Periodic Labs needs to achieve its mission.
Inside the Lab: Robots, Simulations, and the Hunt for New Materials
Periodic Labs has already established its first experimental facility, where AI models and robotic systems work hand-in-hand. The team is currently focused on one of the holy grails of modern science: discovering new superconducting materials — compounds that can conduct electricity without resistance at higher temperatures.
Breakthroughs in this area could enable ultra-efficient computing, energy transmission, and quantum devices, marking a monumental leap for global technology.
While the robotic systems are still in training, the lab is already generating real experimental data that feeds back into its AI models. Every test, whether successful or not, contributes to a self-improving feedback cycle that sharpens the AI’s reasoning capabilities.
Why Periodic Labs Matters: AI’s Next Great Leap
While most AI companies focus on software, Periodic Labs represents a shift toward AI systems that interact with the physical world — bridging the gap between digital intelligence and experimental science.
This approach embodies the next major evolution of artificial intelligence: AI not as a tool for prediction, but as an agent of creation.
“We take a bunch of data, and it can just regurgitate what it knows,” said Deng. “Discovering something new means letting AI test hypotheses in reality.”
By merging computation, experimentation, and automation, Periodic Labs could redefine how humanity approaches discovery — much like how the telescope revolutionized astronomy or the microscope transformed biology.
Challenges Ahead: The Reality of Scientific Discovery
Despite the excitement, Fedus and Cubuk are the first to admit that AI-driven science isn’t a guaranteed success story. Scientific discovery is inherently unpredictable, often more art than algorithm.
Hardware challenges, model limitations, and experimental noise all pose significant hurdles. Moreover, replicating the serendipity of human intuition — the flashes of insight that drive great discoveries — may still be beyond AI’s grasp.
Yet, the founders argue that even failure will produce value: every experiment enriches the data pool that future models can learn from. In this way, Periodic Labs is not just a company — it’s a self-evolving ecosystem of knowledge.
The Dawn of AI-Driven Discovery
With Periodic Labs, the line between human science and artificial intelligence is beginning to blur. Fedus and Cubuk’s vision for AI scientific discovery represents not just a new chapter in research, but a new era — one where algorithms, robots, and researchers work side by side to accelerate innovation.
Whether the startup discovers new superconductors or simply pioneers a model for hybrid human-AI experimentation, its impact will be profound.
In a world where computation increasingly shapes creativity, Periodic Labs stands as proof that the future of discovery belongs to both humans and machines.
The next great scientific revolution may not happen in a university lab — it may be unfolding right now in a room filled with robots, neural networks, and the boundless curiosity of AI.