Jensen Huang Teases Feynman Architecture: 2028 Inference-First Chip Designed Around Agent Long-Term Memory

Financial Content / MarketMinute·medium signal

Alongside the Vera Rubin launch, Huang teased the Feynman architecture (2028 target) as an inference-first chip designed specifically for agentic AI workloads — featuring enhanced KV Cache for long-term agent memory, multi-node memory sharing, and TSMC's 1.6nm process. The framing signals NVIDIA's thesis that the next hardware generation must be architected around autonomous agent requirements, not training throughput.