Substack

SemiAnalysis

Everything we have picked from this writer, newest first.

Intel’s Panther Lake puts 18A PowerVia backside power and RibbonFET GAA transistors into commercial silicon, with logic density similar to…

SemiAnalysis Sep 26

China’s datacenter fleet exceeds 24GW across 1,000 facilities, while Alibaba, Tencent and Baidu doubled capex to $20 billion in 2Q26 amid…

SemiAnalysis Sep 25

Nvidia’s Vera Rubin NVL72 delivers up to 67x better agentic inference performance per dollar than Blackwell, according to SemiAnalysis…

SemiAnalysis Sep 14

Nvidia’s Rubin Ultra cuts HBM capacity to 192GB per GPU as 8-hi and future 4-hi stacks improve bandwidth economics and ease wafer…

SemiAnalysis Sep 13

Nvidia disclosed $530 billion of off-balance-sheet commitments, including AI cloud floors and data-center guarantees, to finance AI…

SemiAnalysis Sep 11

South Korea’s sovereign AI push channels GPU spending to Nvidia while domestic-model ambitions risk conflicting with Samsung and SK Hynix…

SemiAnalysis Sep 1

OpenAI’s Jalapeño inference ASIC beat Nvidia Blackwell on claimed performance per watt in InferenceX tests, using HBM4 and a…

SemiAnalysis Aug 25

AgentX 1.0 open-sources a 1 million-context agentic coding benchmark, testing Nvidia and AMD inference hardware on multi-turn production…

SemiAnalysis Aug 23