Skip to content

NVIDIA

Chip and systems company led by jensen-huang. Positioned by Jensen as the substrate of modern AI — not just a GPU vendor but the maker of the integrated stack (chips, networking, systems, software, data flywheel) that functions as "one computer."

Strategic frame (from jensen-huang-cleo-abram-2026)

  • Sees current AI era as a reinvention of computing, not an application layer on top of existing computing
  • Bets the next era on physical-ai — robotics, autonomous vehicles, embodied systems
  • cuda investment is the template for holding non-consensus bets through years of skepticism
  • Stack integration is the moat: competitors can match chips but not the whole system

Architectural frame (from jensen-huang-lex-fridman-2026)

  • Install base as moat — see install-base-moat. CUDA's defensibility is the developer gravity well, not the ISA.
  • Rack as product — Grace Blackwell (NVL72) was LLM-inference-optimized; Vera Rubin (with Vera CPU, Rock storage rack, NVLink 72) is agent-workload-optimized. NVIDIA now ships buildings, not chips.
  • Supply chain co-management — works continuously with ASML, TSMC (CoWoS), SK Hynix (HBM). See ai-supply-chain.
  • Open source as research signal — studies DeepSeek, MiniMax, Nemotron-3 (hybrid Transformer+SSM) to see where architectures are going, so the hardware can be there when the workload arrives.