AI supply-chain thesis — mapping bottlenecks, focus companies, and supply-chain exposure for investors.
**Bottleneck theme:** Custom Silicon
**Focus:** $NVDA — NVIDIA CORP
NVIDIA is the anchor of every AI portfolio — and consensus is already paying. The franchise is structurally protected by the CUDA software moat, a generational silicon roadmap (Hopper → Blackwell → Blackwell Ultra → Rubin → Rubin Ultra → Feynman) that competitors can match in hardware but not in ecosystem, and a near-monopoly grip on TSMC CoWoS-L allocation through Rubin generation. Hyperscaler 2026 capex tracking >$700B, with NVIDIA capturing the dominant share of accelerator spend, makes the multi-year revenue picture the cleanest in semiconductor history.
The investment case is "hold, don't add aggressively" rather than fresh accumulation: expectations are extremely high, position sizing is large in most portfolios, and the highest-conviction marginal buyers have already bought. The bear case isn't competitive — it's *valuation* combined with the magnitude of any negative surprise (a hyperscaler capex pause, a CoWoS supply hiccup, a ROCm parity event at AMD, or the China revenue cliff under further export controls). Treat as the core AI exposure but lean into the secondaries — the supply chain that is forced to ramp around NVIDIA — for incremental alpha.
