Chinese labs sitting on efficient model architectures while US players burn capital on compute-heavy scaling. Anthropic's cost structure already spooked markets - a DeepSeek-style release showing 10x better performance per dollar would accelerate the commoditization narrative and crater valuations across $MSFT $GOOGL $META AI infrastructure bets. Timing matters: maximum psychological impact while Western labs are locked into expensive training runs they can't abort. Efficiency arbitrage is the real moat killer here.