SemiAnalysis digests

Summarized and annotated for the non-specialist.

• How GLM5.3 Sparse Attention Affects HBM Memory Usage (1107 words · 2026-09-28)
• Intel Panther Lake Teardown (1114 words · 2026-09-26)
• The Chinese AI Infrastructure Boom: Introducing the SemiAnalysis China Datacenter Model (1185 words · 2026-09-25)
• ClusterMAX 3.0: The Industry Standard GPU Cloud Rating System Returns (1288 words · 2026-09-23)
• Computation and Data Movement for Inference (1305 words · 2026-09-21)
• Engrams Embedding Entendre: Codesign for Efficient DRAM/SSD Offloading (1063 words · 2026-09-18)
• Everyone Says Datacenter Moratoriums Are Killing the US Buildout. We disagree (1073 words · 2026-09-15)
• A Brain Too Big to Carry — On-Device vs Datacenter Inference (1187 words · 2026-09-14)
• Vera Rubin NVL72 Agentic Inference: 67x better Performance per Dollar (1177 words · 2026-09-14)
• Long Live the Short King: Why 4-hi HBM Wins (1165 words · 2026-09-13)
• Nvidia’s Backstop Universe – Heads I Win, Tails Who Loses? (1217 words · 2026-09-11)
• What is So Hard About Behind-The-Meter Power For Datacenters? Part 1 (1383 words · 2026-09-10)
• Where Does a Robot Think – On-Device vs Datacenter Inference (1223 words · 2026-09-09)
• TPU Inference Externalization Full Steam Ahead (1147 words · 2026-09-07)
• Korea’s Trillion-Dollar Sovereign AI Investment (1129 words · 2026-09-01)
• Most Neoclouds Suck At Security (1205 words · 2026-08-30)
• OpenAI Jalapeño: Better Than Nvidia Blackwell (1210 words · 2026-08-25)
• Cerebras's Next Generation CS-4: Fast Just Got Faster (1130 words · 2026-08-19)
• $12B of US ratepayers' money wasted on a modeling mistake and PJM wants to do it again (1336 words · 2026-08-16)
• Ultra-High Interactivity on NVIDIA GPUs? - TileRT InferenceX (2752 words · 2026-08-10)
• SpaceX 10GW in 2027 – Why It’s Real, Will Drive $500B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtak… (1760 words · 2026-08-07)
• Kimi K3, The Manos, The Mythos, The Legendos (3057 words · 2026-08-03)
• The Wild Wild West Of LEGO Datacenters (3375 words · 2026-07-29)
• Can AMD break the CUDA Moat? AMD Advancing AI 2026 (4280 words · 2026-07-25)