20VC: Anj Midha on Investing $300M into Anthropic | The Early Days of Anthropic & How 21 of 22 VCs Turned it Down | The Four Bottlenecks to Compute | What the China Has Smashed and Why We Should Be Worried

The episode identifies four key bottlenecks in advancing AI: context feedback, compute, capital, and culture, arguing that culture drives algorithmic innovation. It presents a real-world example of a physical lab at Periodic Labs using LLMs to predict materials, robots to synthesize them, and machines to validate results, creating a closed-loop system. The speaker emphasizes that unique data from physical domains—like material science—is a critical, underutilized bottleneck and source of competitive advantage.

You get a concrete, operational framework for identifying where real AI progress is bottlenecked and how to build systems that close the loop between prediction, action, and physical validation.

1. 01The four bottlenecks to AI progress are context feedback, compute, capital, and culture—with culture being the most critical enabler.
2. 02Real-world data from physical systems (e.g., robotics, material synthesis) is essential for advancing AI beyond synthetic benchmarks.
3. 03Superhuman AI capabilities are already emerging in narrow domains like coding and material science through recursive automation and closed-loop experimentation.