Answer extracted from the Invest Like The Best podcast — listen to the full episode below.
What makes Tony Zhao and Cheng Chi's approach to robotics generalization and data collection particularly innovative?
Tony Zhao and Cheng Chi at Sunday Robotics reframe data scarcity as a solvable technical problem rather than a fundamental blocker. Instead of chasing internet-scale datasets, they focus on collecting data cheaply and cleverly while maintaining the real-world distribution required for effective robot generalization.
Their core innovation lies in inverting the typical robotics challenge. Most teams assume that building robots that generalize across tasks and environments requires massive, expensive datasets collected at scale. Sunday Robotics takes a different path: they solve for how to extract maximum learning value from minimal, intelligently gathered data.
This approach matters because it unlocks robotics research for teams that don't have access to billions in compute budget or thousands of robots. As Sarah explains in the episode, their belief that data constraints can be engineered away stands in sharp contrast to the fatalism many roboticists express about needing massive volumes.
Sarah — Early-stage investor at Greylock focused on AI, biology, defense, and robotics. She leads a venture fund with partners including Mike, Pranav, and Bella, and is drawn to high-agency founders who solve frontier problems with clarity and deep technical intuition.
The practical impact becomes clear when you look at Sunday Robotics' timeline. Within just two years of existence, they are aiming to deploy general semi-humanoid robots in homes by the end of the year. That pace depends entirely on treating data collection as a design problem, not a resource problem. They ask: How can we collect data in the cheapest way possible while still maintaining the distribution that supports real environments and real tasks?
This is distinct from both the frontier labs approach (throw compute at scale) and from academic robotics (perfect data collection, slow iteration). A point discussed in detail in this podcast, their philosophy of intelligent data scarcity reflects a deeper belief: that the best innovations often come from working backward from constraints, not forward from unlimited resources.
The legitimacy of this approach rests on translating cheap data collection directly into model learning. It's not about cutting corners on data quality; it's about being surgical with data gathering strategy so that every sample learned from improves the robot's real-world performance. This conversation with Sarah captures why investors are watching this team closely — they represent a different operating model for robotics, one that could reshape which teams can compete in the space.
See also
What regulatory and infrastructure barriers are preventing rapid scaling of compute capacity for AI development?
Sarah identifies compute availability and energy supply as critical constraints over the next five to ten years. She spoke with a hyperscaler infrastructure leader who stated nothing would move the needle for compute capacity at sufficient scale before 2030.
What psychological challenges do frontier AI researchers face when working at scale in compute-intensive environments?
Sarah observes that many researchers feel disempowered by the massive scale and compute requirements of frontier labs. A large contingent believes either in recursive self-improvement leading to exponential intelligence within one to two years, or they feel sidelined by the infrastructure demands.
What concerns do investors have about research-heavy AI companies when assessing capability and impact?
Sarah expresses concern that many investors are making large-scale research bets without fundamental intuition or a grounded point of view on the business fundamentals, competitive positioning, or reproducibility of results.