Google's AI talent exodus is a strategic trade-off, not a crisis, according to Asianometry's John Youshaei on the SemiAnalysis podcast. As Jeff Dean and the core Gemini team depart, Google is choosing infrastructure over frontier leadership, potentially mirroring Bell Labs, IBM, and Intel while remaining the third most valuable company at $4.3 trillion. In one week, Google lost its chief scientist, Gemini co-leads, and $190 billion in market value. Departing talent follows a recursive pattern: leave Google, raise capital (often from Google Ventures), buy NVIDIA GPUs instead of TPUs, and found competing labs. TPUs are effective for pre-training, but outside Google, software support is zero. Elon Musk's Terafab revenue guidance of $1 trillion by 2030 fuels a spending boom where bottlenecks are memory and optical transceivers, dominated by Chinese suppliers. The GPT-5 retrospective confirms individual models may disappoint, but the trajectory is vindicated, with GPT-6 arriving this week and AI coding tools building production software for a few hundred dollars. The open question is containment, with frontier labs competing to showcase models breaking out of isolation.