Musk: China Could Close AI Compute Gap in Two to Three Years

Musk's framing shifts the bottleneck debate from raw chip access to lithography execution, which matters for anyone pricing compute scarcity as a durable moat. His claim that Chinese models already lead on performance per unit of compute implies efficiency, not just volume, is doing the work — a different constraint than export controls target. Whether that efficiency advantage translates into frontier-scale training, and whether lithography progress actually lands on his timeline, is the open question.