Context Drought: 1M Token Windows for 5–10 Years.

SK Hynix partners with NVIDIA / Shared entity: HBM / Shared topic / What happened next

Dylan Patel: Memory Is Now 30% of AI CapEx — HBM Demand "Completely Ignited"

Linked by a graph relationship (SK Hynix partners with NVIDIA); both cover HBM; overlapping topics (demand, memory).

Samsung competes with SK Hynix / Shared entities / Shared topic / What happened next

200+ Samsung engineers left for SK Hynix in four months, and 81.5% of the foundry division wants out.

Linked by a graph relationship (Samsung competes with SK Hynix); both cover HBM, SK Hynix; overlapping topics (hynix, memory).

Samsung competes with SK Hynix / Shared entities / What happened next / Tension

Samsung, SK Hynix, Micron sell out 2027 memory supply

Linked by a graph relationship (Samsung competes with SK Hynix); both cover DRAM, HBM; picks up the DRAM thread on 2026-08-07.

Samsung competes with SK Hynix / Shared entity: HBM / What happened next

Nvidia is testing Rubin Ultra at 192GB and stepping back from HBM4E to HBM4.

Linked by a graph relationship (Samsung competes with SK Hynix); both cover HBM; picks up the HBM thread on 2026-08-11.

SK Hynix partners with NVIDIA

NVIDIA open-weighted Magpie TTS Multilingual: 12 languages, 364M params, 32ms time-to-first-audio on B200.

Linked by a graph relationship (SK Hynix partners with NVIDIA).

Google open-sourced DiffusionGemma, a text-diffusion model that generates 4x faster, and you can run it locally today

Linked by a graph relationship (SK Hynix partners with NVIDIA).

NVIDIA Nemotron 3 Super: 120B/12B Active MoE Open Model for Agentic AI at 5x Prior Throughput

Linked by a graph relationship (SK Hynix partners with NVIDIA).

H100 uses DRAM / Shared entity: HBM / Earlier coverage

NVIDIA Secret Inference Chip with Groq LPU Technology

Linked by a graph relationship (H100 uses DRAM); both cover HBM; earlier HBM coverage from 2026-03-04.