In a conversation on the a16z podcast, Fei-Fei Li, co-founder of World Labs, and Yunzhu Li, co-founder of Scenix, argued that building useful robots is a data problem, not a hardware problem, and that the path to reliable machines runs through simulation. Their companies have joined forces to build an infrastructure platform that is model-agnostic and embodiment-agnostic, selling digital twins and training environments rather than robots themselves. They stated that simulation is essential for counterfactual reasoning, systematic state-space coverage, and fast iteration that real-world data alone cannot provide. They poured cold water on near-term humanoid hype, pointing out that human-level power efficiency remains decades away and that the most promising near-term deployments will be specialized robots in semi-structured environments like warehouses and assembly lines. The article notes that when Waymo trains its self-driving systems, it relies more heavily on billions of hours of simulation than on real-world miles driven, and Fei-Fei Li described self-driving cars as "the simplest kind of robots." For investors watching the $88 billion robotics market, the message is that companies building digital world creation platforms may capture more durable value than those bolting legs onto AI models.