Archive - SemiAnalysis

TPU Inference Externalization Full Steam Ahead - InferenceX

InferenceX, Up to 50% Better Performance per Dollar, Rapid Externalization of TPU stack, Growing Customer Base, Ironwood, TPUv8i, Reducing CUDA Moat

Sep 7 • Alec Ibarra, Cam Quilici, Bryan Shan, Wenyao Gao, Daniel Nishball, Zane Fong, and Dylan Patel

Korea’s Trillion-Dollar Sovereign AI Investment: Nvidia Wins, Hynix Loses

Korea hosts a Squid Games, National AI Tournament, the best non-Chinese open source model gets eliminated, why Nvidia needs open source, implications…

Sep 1 • Max Kan, Ray Wang, Myron Xie, and Dylan Patel

August 2026

Most Neoclouds Suck At Security

OpenAI vs HuggingFace, Container Escapes, Kernel Bypass, Network Policies, Security Keys, Multi-tenant Grafana, and a ClusterMAX 3.0 Preview

Aug 30 • Jordan Nanos, Sam Harshe, Pratt Bhatt, Billy Cao, Jack Carson, and Dylan Patel

OpenAI Jalapeño: Better Than Nvidia Blackwell

OpenAI’s self-designed ASIC compared with Rubin, Jalapeño’s TCO, throughput per MW, and spicy deets

Aug 25 • Bryan Shan, Myron Xie, Jordan Nanos, Wega Chu, Clara Ee, and Dylan Patel

AgentX - InferenceXv3: Does CUDA Moat Hold up in Agentic Inferencing?

$3 Million USD dataset open sourced, 1 Mil+ Context Length, Multiturn, Sub Agents 95%+ KVCache HitRate, GB300 NVL72, MI355, B200

Aug 24 • Cam Quilici, Bryan Shan, Alec Ibarra, Daniel Nishball, Zane Fong, Kimbo Chen, and Dylan Patel

Are Open Models Catching Up?

Comparing open vs. closed models across the eras of frontier models, Is the gap narrowing?

Aug 21 • Evan Cloutier, Max Kan, Jordan Nanos, and Dylan Patel

Cerebras's Next Generation CS-4: Fast Just Got Faster

Double the Performance, Double the Power, Double the Fun

Aug 19 • Myron Xie, Bryan Shan, Wega Chu, Jordan Nanos, and Dylan Patel

$12B of US ratepayers' money wasted on a modeling mistake and PJM wants to do it again

American Grid design needs an overhaul, Why it is good to be full of cold air.

Aug 16 • Robert Boswall, Reyk Knuhtsen, Jeremie Eliahou Ontiveros, Oliver Kennon, Dylan Patel, and Dashboard American

Ultra-High Interactivity on NVIDIA GPUs? - TileRT InferenceX

Can TileRT software on NVIDIA GPU compete with Cerebras, Groq LPU, SambaNova? Batch Size 1, Disaggregated engine, high throughput prefill engine, high…

Aug 10 • Bryan Shan, Daniel Nishball, Cam Quilici, Kimbo Chen, Alec Ibarra, and Dylan Patel

SpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker

Inference at 100B/GW/year, SpaceX's stellar pace, Microsoft's 10GW 2026 Awakening, Azure Can Grow Triple-Digits

Aug 7 • Jeremie Eliahou Ontiveros, Reyk Knuhtsen, Jordan Nanos, Max Kan, Dylan Patel, and Muhammad Zuhair

Gemini is Cooked but GCP is Cooking

GCP YoY rev growth >100%, DeepMind's long term failure is Google Cloud's short term gain

Aug 7 • Max Kan, Joey Brookhart, Doug O'Laughlin, and Dylan Patel

Kimi K3, The Manos, The Mythos, The Legendos

Kimi K3’s architecture: compressed memory, attention across depth, latent expert routing, and serving performance

Aug 3 • Kimbo Chen, Shubham Choudhari, Bryan Shan, and Dylan Patel

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts