Visiting Taiwan w/ Jensen Huang | Lex Fridman
Jensen Huang reframes AI compute economics: value is tokens generated, not CPU cores rented
“AI doesn't rent cores. AI generates tokens, and they sell tokens.”— Jensen Huang
At Computex in Taipei, Jensen Huang articulated a fundamental reframe of AI compute economics: the hyperscale CPU era optimized for dollars-per-core, but AI infrastructure is now optimized for token throughput — 'the more you buy, the more you make.' He also previewed NVIDIA's latest hardware achieving 100% liquid cooling (up from 80% in Grace Blackwell) and 5-minute assembly (down from 2 hours), signaling production maturity. The shift in framing from rented compute to token economics has direct implications for how hyperscalers and enterprises will justify AI infrastructure spend.