AI NEWS PAGE DOSSIER

Dylan Patel

Dylan Patel's file: SemiAnalysis, chips, supply chains, GPUs, datacenters, and the independent researcher counting the physical machinery underneath the AI story.

Patel started SemiAnalysis as a solo semiconductor-analysis project, then followed the AI boom down into the physical stack: advanced packaging, HBM, accelerator shortages, datacenter power and individual facilities. The newsletter evolved into an independent research shop selling models of the machinery beneath AI.

HOW THE HELL DID DYLAN PATEL GET HERE?

  1. 2020

    SEMICONDUCTOR NERD STARTS A SOLO NEWSLETTER

    SemiAnalysis says Patel founded the project in 2020 and built it from a solo venture into a research firm covering the semiconductor supply chain, cloud infrastructure and machine learning.

    RECEIPTS: SemiAnalysisfirst-party

  2. Jul 5, 2023

    THE AI BOOM BECOMES A PACKAGING AND MEMORY PROBLEM

    SemiAnalysis published detailed estimates of H100 supply constraints and the CoWoS and high-bandwidth-memory bottlenecks limiting AI accelerator production.

    RECEIPTS: SemiAnalysisfirst-party

  3. Mar 13, 2024

    HE FOLLOWS THE CHIPS ALL THE WAY TO THE POWER GRID

    Patel and colleagues argued that AI growth was increasingly constrained by datacenter capacity, power generation and the logistics of deploying huge GPU clusters.

    RECEIPTS: SemiAnalysisfirst-party

  4. Jun 30, 2025

    THE NEWSLETTER STARTS MAPPING THE AI CLOUD WAR SITE BY SITE

    SemiAnalysis used its datacenter research to analyze Oracle’s role in Stargate, OpenAI infrastructure and large AI campuses, tying corporate strategy directly to physical capacity.

    RECEIPTS: SemiAnalysisfirst-party

  5. 2026

    FIVE THOUSAND DATACENTERS GO INTO THE MODEL

    SemiAnalysis’s Datacenter Industry Model tracks more than 5,000 facilities using permits, property records, power data, FOIA material and satellite imagery to estimate where AI capacity actually exists.

    RECEIPTS: SemiAnalysisfirst-party

  6. Sep 15, 2026

    THEN THEY MAP 300-PLUS DATACENTER MORATORIUMS

    Patel and colleagues mapped more than 300 local datacenter restrictions and estimated how much planned capacity was actually delayed, turning a political argument into a facility-level reporting project.

    RECEIPTS: SemiAnalysisfirst-party

MORE DOSSIERS