Periodic Labs, an AI startup with the audacious goal of creating an “AI scientist,” has secured a massive $300 million seed round. The funding, led by Andreessen Horowitz with participation from Nvidia, Accel, DST, and a who’s-who of tech royalty including Jeff Bezos and Eric Schmidt, values the company at a reported $1 billion pre-money, according to Bloomberg.
The startup is tackling what it sees as a fundamental limitation of today’s AI: the internet is finite. While large language models have mastered the web’s ~10 trillion tokens of data, Periodic argues that true scientific discovery requires moving beyond regurgitating existing knowledge. Its solution is to pair AI models with autonomous, robot-powered labs that can run real-world experiments, generating massive amounts of proprietary data that exists nowhere else.
Beyond the internet
Periodic’s core strategy is to create a feedback loop where an AI conjectures a hypothesis, and a physical lab tests it. The AI then learns from the results — especially the failures, which are rarely published but are crucial for learning. “Nature is the RL environment,” the company states, positioning its AI as an agent that can actively probe reality rather than just analyze a static dataset.
The company, co-founded by former Google DeepMind materials lead Ekin Dogus Cubuk and ex-OpenAI VP of research Liam Fedus, is stacked with talent from Meta, Databricks, and Samsung. Their first major target is discovering high-temperature superconductors, a holy grail of materials science that could enable hyper-efficient power grids and data centers that don’t melt themselves.
While that’s a long-term moonshot, Periodic is already working with industry partners. In a blog post, the company mentioned it’s helping a semiconductor manufacturer use custom AI agents to solve chip heat dissipation issues by making sense of experimental data faster.
This $300 million war chest will be used to scale its labs and hire more AI researchers and experimentalists, accelerating its mission to automate the scientific process itself. It’s a bet that the next great breakthroughs won’t be found by re-reading the internet’s textbooks, but by giving AI the tools to write its own.
