Perspectives.
Separate from data-driven research — a track for haelangdal's subjective analysis, theses, metaphors, and sentiment-cycle reads. New releases are locked to upgraded tiers.
Reports
Deep dives & market readsThe CPU in the Inference Era — What Is Measured Fact and What Is Forecast
Decomposing the claim that 'inference needs this much CPU' into measured fact, stakeholder forecast and offsetting factors — three tiers of evidence and the Korean memory derivative
The Accounting of GW - Paper Capacity vs Physical Capacity in AI Data Centers
A site-by-site reconciliation of announced pipelines against energized capacity - delays do not destroy demand, they shift it to the right
How AI Gets Connected — Reading the Connectivity Value Chain as Layers
Dissecting AI infrastructure's third bottleneck, 'connectivity,' from the basics (scale-up and scale-out) to the 8-layer value chain — where copper's time and light's time split
MLCC and FC-BGA — The Benefit Structure of Two Battlefields, Finished Goods and Materials
Dissecting the AI-server component supercycle as a vertical value chain — where finished goods (Samsung Electro-Mechanics et al.) and materials each benefit
The Place Memory Took — Vera Rubin Rewrites the Data Center BOM
Within a VR200 NVL72 rack at $7.8M, memory takes 25.6% — the "HBM 15.5%" proposition evolves into "memory 25.6%"
AI Data Center BOM Reach Rate
AI Data Center Deep Dive — Part 3. NVIDIA gets a share, HBM, optical, and power must follow at theirs. Reverse-engineering revenue ceilings by industry through BOM accounting identity.