Archive

All Analyses

3 analyses total

2026 (3)

2026

Gemma 4 MTP on GB10 — Inference 1.83x faster on 26B, 3.52x on 31B Dense

AILLMvLLM

Decomposing the AI Value Chain in the Agentic Era

AISemiconductorHBM

How DeepSeek V4's 90% KV Cache Reduction Reshapes HBM Demand — From Capacity to Bandwidth, Packaging, and Thermal Control

AISemiconductorHBM