Layer L2
Silicon & Memory — Leaderboard
Ranked companies in the Silicon & Memory layer of the AI infrastructure stack. Share figures carry an explicit provenance label (official, filing-derived, third-party, or site-estimate). Methodology →
| # | Company | Product | Share | Source | Customers |
|---|---|---|---|---|---|
| 1 | Amkor Technology | HBM3E / HBM4 stacked memory | – | site-estimate | – |
| 2 | ASE Technology | HBM3E / HBM4 stacked memory | – | site-estimate | – |
| 3 | NVIDIA | HBM3E / HBM4 stacked memory | – | site-estimate | Microsoft, Microsoft (Azure / OpenAI deployment) |
| 4 | SK Hynix | HBM3E / HBM4 stacked memory | – | site-estimate | NVIDIA, AMD |
| 5 | AMD | Chiplet die-to-die interconnect (UCIe / 2.5D-3D chiplet partitioning) | – | site-estimate | – |
| 6 | Micron Technology | HBM3E / HBM4 stacked memory | – | site-estimate | – |
| 7 | Samsung Electronics | HBM3E / HBM4 stacked memory | – | site-estimate | NVIDIA |
| 8 | TSMC | CoWoS 2.5D advanced packaging (HBM-on-logic) | – | site-estimate | NVIDIA |
| 9 | Rebellions | Chiplet die-to-die interconnect (UCIe / 2.5D-3D chiplet partitioning) | – | site-estimate | – |
| 10 | Groq | Groq LPU (Language Processing Unit) | – | site-estimate | Saudi Arabia (Aramco Digital / HUMAIN) |
| 11 | Cerebras Systems | Cerebras Wafer-Scale Engine | – | site-estimate | G42 |
| 12 | SambaNova Systems | SambaNova RDU (Reconfigurable Dataflow Unit) | – | site-estimate | – |
| 13 | Tenstorrent | Tenstorrent Blackhole-class AI accelerator | – | site-estimate | – |