Overview
- J.P. Morgan’s Harlan Sur now models far higher revenue per gigawatt, estimating Nvidia at about $35–$40 billion, AMD’s new Helios platform near $20 billion, and Broadcom around $27 billion, with further upward revisions expected as new capacity agreements surface.
- TSMC and Nvidia produced the first Blackwell wafer on U.S. soil at TSMC’s Phoenix facility, signaling the start of U.S. volume production, as TSMC maintains elevated annual capital spending of roughly $40–$42 billion to meet AI demand.
- Nvidia’s letter of intent with OpenAI ties up to $100 billion of investment to roughly 10 gigawatts of systems, while AMD’s pact covers about 6 gigawatts and includes warrants for OpenAI to buy up to around 10% of AMD’s shares; AMD stock has climbed roughly 42% since Oct. 6.
- Strategists and investors are increasingly debating circular ‘vendor financing’ and the durability of the AI capex cycle, with Ned Davis Research’s Ed Clissold warning such cycles often end poorly even as others see ample funding and long runways for spending.
- Yahoo Finance reports xAI is signing a roughly $20 billion Nvidia GPU deal for its Memphis Colossus 2 data center, underscoring continued multibillion‑dollar compute commitments that suppliers can plan against, though the companies have not confirmed the agreement.