If Mark Zuckerberg offers you a multi-million dollar package to join Meta’s elite Superintelligence Labs, you sign on the dotted line. That’s how you declare your arrival in Silicon Valley.
But AI researcher and IIT Bombay alumnus Rishabh Agarwal refused Zuckerberg’s offer and instead chose not to follow the standard playbook.
An IIT Bombay graduate with an All India Rank (AIR) 33 in JEE, Rishabh has worked at the world’s top AI labs, including Google Brain, DeepMind, and Meta, before co-founding Periodic Labs.
And what does Periodic Labs do? Backed by NVIDIA, Jeff Bezos, and other luminaries, Agarwal’s startup aims to create AI scientists that accelerate real-world scientific discovery.
An IITian who went to the frontiers of AI research
Agarwal earned his B.Tech in Computer Science and Engineering from IIT Bombay, following which he pursued a PhD at Mila under advisors Aaron Courville and Marc Bellemare. His early career included time at Geoffrey Hinton’s team in Google Brain, Toronto.
He later became a staff research scientist at Google DeepMind. His contributions span reinforcement learning (RL), large language models (LLMs), and scaling techniques. Notable achievements include an Outstanding Paper Award at NeurIPS, pioneering work on on-policy distillation for LLMs, popularising offline RL, and core contributions to models like Gemma and Gemini.
He also served as an Adjunct Professor at McGill University and mentored numerous students who advanced to roles at OpenAI, xAI, and Google DeepMind.
Agarwal’s Meta chapter that made him famous
In early 2025, Agarwal joined Meta’s Superintelligence Labs, led by Alexandr Wang. After just five months, though, he faced a life-altering decision. While most accounts describe a roughly $1 million (Rs 9.5 crore per annum) offer, Agarwal clarified that the actual compensation package was “an order of magnitude higher” than $1 million for a seat in one of the industry’s most sought-after AI labs.
Despite the lucrative job, Agarwal chose to let it go, citing the need for a different kind of risk and impact after years at the biggest labs. He then went on to establish Periodic Labs as his next venture.
Founding Periodic Labs
Agarwal is a founding member of Periodic Labs, which raised a massive $300 million seed round. The company is building AI scientists paired with autonomous labs to drive breakthroughs in materials science, physics, and more, with applications ranging from superconductors to semiconductor improvements. It is an early-stage AI startup backed by heavyweights like Andreessen Horowitz, NVentures (NVIDIA), and Jeff Bezos.
Founded by AI pioneers Liam Fedus (co-creator of ChatGPT and former OpenAI VP of Research) and Ekin Dogus Cubuk (former DeepMind lead behind the revolutionary materials-discovery tool GNoME), Periodic Labs is attempting something radically different from traditional software startups.
Instead of training AI strictly on digital text and images, they are building an AI Scientist, which is paired directly with autonomous physical laboratories to unlock breakthroughs in the physical world.
Periodic Labs uses advanced reinforcement learning (RL) models to orchestrate real-world robotics. The AI reads scientific literature, generates a hypothesis, and commands a physical robotic lab to mix powders, run chemical synthesis, and test materials end-to-end.
By utilising what they call a “physical reward function,” nature itself becomes the AI’s training environment. The system doesn’t just record success – it learns massively from failed, unpublished experiments (“negative data”) to rapidly iterate faster than humanly possible.