SemiAnalysis

SemiAnalysis provides research, industry models and consulting on semiconductors and AI infrastructure for AI companies, chipmakers, cloud providers and investors. Its AI Cloud TCO Model examines the economics of buying accelerators and selling GPU compute, while its accelerator and datacenter models track chip production and forecast IT power capacity. Networking and wafer fabrication models extend that coverage to interconnects and semiconductor equipment, connecting technical requirements with infrastructure economics.

Founder Dylan Patel remains CEO and chief analyst, having developed SemiAnalysis from a solo venture in 2020. Its public engineering work includes InferenceX, which benchmarks inference across accelerators and serving software with reproducible recipes, logs and results. Its AgentX workload replays the structure of long coding sessions, including subagents and repeated cache reuse, without exposing original user content. This lets engineers examine serving behavior beyond isolated prompts, including how routing and memory management affect performance.

The business combines subscription research with institutional models, retained advisory work and bespoke projects. In August 2026, SemiAnalysis reported operating its benchmark matrix across more than 1,000 chips and approximately 2 MW of compute, and spending over $3 million on the dataset underlying AgentX. It also entered a semiconductor research partnership with Tema ETFs in 2026, with a suite of funds planned.