[Mechanism Capital Partner: AI Physical Data Scale to Expand 100x by 2026] Mechanism Capital partner Andrew Kang posted on the X platform, stating that by 2025, the robotics field will have resolved long-standing challenges in model architecture and training. Significant advancements will be made in data collection technology, data quality understanding, and data formulation, giving AI companies the confidence to finally invest in large-scale data collection. Companies like Figure, Dyna, and PI are leveraging innovative reinforcement learning (RL) technologies to achieve over 99% success rates across various real-world application scenarios. Additionally, progress in memory technology has broken the "memory wall." NVIDIA's ReMEmber utilizes memory-based navigation, while Titans and MIRAS have implemented memory during testing. Improved virtual localization models (VLM) mean virtual localization arrays (VLA) will possess better spatial understanding capabilities, along with significantly enhanced throughput in data annotation and processing workflows. By 2025, the market will begin to experience zero-shot capability mapping, visual sensitivity, and general physical reasoning enabled by data scale. By 2026, the scale of physical AI data is expected to expand 100 times.