Company File
The king: GPUs, the CUDA moat, Mellanox networking, and the expanding software stack.
Plate I — Pillar-Purity Score
The reference AI-infrastructure stock. All roads lead here.
Methodology — How we score
Plate II — The Thesis
Why it’s on the list.
Nvidia sells the GPUs every hyperscaler is buying, the CUDA platform that locks them in, and increasingly the full rack-scale system via Mellanox-derived networking and DGX. The thesis is that Nvidia keeps its 90%+ training share while incrementally taking networking and software dollars.
Plate III — Supply Chain
Upstream & downstream.
See the full chain plate for all 37 names.
Upstream — what feeds NVDA
Plate IV — Thesis Breakers
What would invalidate this.
- Risk 01Custom-silicon (TPU, MTIA, Trainium) displaces material GPU share
- Risk 02Open-source CUDA alternative matures
- Risk 03HBM / CoWoS supply loosens too fast
Plate V — Also on the Silicon plate
Siblings by pillar.
- TSMScore 10Fabs every leading-edge AI chip on earth — the real bottleneck of the entire stack.
- ASMLScore 10EUV lithography monopoly; no EUV means no sub-5nm chips.
- ALABScore 10PCIe retimers and CXL connectivity — the glue chips inside every AI rack.
- CRDOScore 10Active electrical cables and SerDes for AI-rack interconnect; pure-play AI networking IC.
- AVGOScore 9Custom silicon partner to Google (TPU) and Meta (MTIA); dominant AI networking ASICs.
Close
Next earnings: 2026-05-27.
The Sunday brief decodes earnings that matter. Five minutes. Free.
Research only. Not investment advice. Do your own diligence.