Latest
Huawei and SMIC at 7 nm: Capability Assessment
One Year After the October 7, 2022 Controls The October 7, 2022 U.S. export controls aimed to limit access to advanced computing, AI, and supercomputing for military-relevant use cases. One...
Huawei and SMIC at 7 nm: Capability Assessment
One Year After the October 7, 2022 Controls The October 7, 2022 U.S. export controls aimed to limit access to advanced computing, AI, and supercomputing for military-relevant use cases. One...
Bottlenecks And Lead Times Across CoWoS And HBM...
Overview Generative AI demand is constrained by advanced packaging capacity, particularly for chip-on-wafer-on-substrate (CoWoS) assemblies that integrate leading-edge logic (e.g., 5 nm) and high-bandwidth memory (HBM). Capacity shortages are expected...
Bottlenecks And Lead Times Across CoWoS And HBM...
Overview Generative AI demand is constrained by advanced packaging capacity, particularly for chip-on-wafer-on-substrate (CoWoS) assemblies that integrate leading-edge logic (e.g., 5 nm) and high-bandwidth memory (HBM). Capacity shortages are expected...
Lam Research Faces Tokyo Electron In Next-Gen N...
Overview NAND Flash remains the lone segment delivering consistent density/cost scaling as logic and DRAM slow. Since the first commercial 3D NAND (2013), density has improved ~30% per year by...
Lam Research Faces Tokyo Electron In Next-Gen N...
Overview NAND Flash remains the lone segment delivering consistent density/cost scaling as logic and DRAM slow. Since the first commercial 3D NAND (2013), density has improved ~30% per year by...
Cost Analysis: Training And Inference Economics...
Executive Summary A widely held view in industry is that GPT-4’s architecture remains undisclosed less for safety reasons than because it is replicable. Comparable or superior systems are expected from...
Cost Analysis: Training And Inference Economics...
Executive Summary A widely held view in industry is that GPT-4’s architecture remains undisclosed less for safety reasons than because it is replicable. Comparable or superior systems are expected from...
Networking Requirements For MI300 Clusters In A...
Context: Short Supply, High Margins, Urgent Need for Options Industry demand for AI compute far exceeds supply. With GPU shortages and datacenter accelerators reportedly priced at multi-fold markups over manufacturing...
Networking Requirements For MI300 Clusters In A...
Context: Short Supply, High Margins, Urgent Need for Options Industry demand for AI compute far exceeds supply. With GPU shortages and datacenter accelerators reportedly priced at multi-fold markups over manufacturing...
Memory Pricing Pressure In AI Servers
Market Snapshot and Misalignment The rush to deploy AI training and inference infrastructure has lifted many component stocks, sometimes beyond their direct exposure. For example, Credo gained about 27% in...
Memory Pricing Pressure In AI Servers
Market Snapshot and Misalignment The rush to deploy AI training and inference infrastructure has lifted many component stocks, sometimes beyond their direct exposure. For example, Credo gained about 27% in...