Overview
- Nvidia says its Rubin AI platform is now in production with shipments expected in the second half of 2026, and claims up to roughly 90% lower inference cost per token versus Blackwell.
- Management disclosed a 2026 backlog exceeding $500 billion across Blackwell, Rubin, and related networking products, and reported more than 2 million H200 orders worth about $54 billion.
- Chinese authorities have asked some companies to pause H200 purchases as they weigh domestic procurement, creating near‑term uncertainty even as Nvidia prepares to resume sales in China this year.
- Micron reports its high‑bandwidth memory supply is fully allocated for calendar 2026 and projects strong multi‑year HBM growth, highlighting memory as a key constraint for AI buildouts.
- Wall Street remains broadly positive on Nvidia with expectations near 50% revenue growth in 2026, while noting integration demands such as 800‑volt power systems and ongoing competitive, supply, and policy risks.