N環球 
Nvidia's Vera Rubin Enters Full Production: AI Server Node Assembled in Under 5 Minutes
Jensen Huang announced at GTC Berlin that the Vera Rubin AI platform is in full production; Google Cloud's A5X instances go live in November as Samsung, SK Hynix and Micron race for HBM4 orders.
Why it matters
Vera Rubin is Nvidia's next-generation AI computing platform, succeeding Grace Blackwell. Full production marks the formal start of the next AI data-center hardware cycle — and sets off a battle among Samsung, SK Hynix and Micron for high-bandwidth memory (HBM) orders.
Over 20x faster assembly
CEO Jensen Huang announced on October 21 in his GTC Berlin keynote that Vera Rubin is in full production. Assembling a server node now takes under five minutes, down from two hours for Grace Blackwell — a reduction of more than 95%. The new design eliminates cables and wiring and uses 100% liquid cooling.
Cloud giants move first
Google Cloud announced its Vera Rubin-based A5X instances will reach general availability in November, with plans reported for up to 960,000 Rubin GPUs. Nvidia shares rose about 0.7% intraday on the news.
HBM4 supply race heats up
Production ramp-up also boosts demand for advanced memory. Samsung and SK Hynix are competing for next-generation HBM4 supply slots, with Micron also qualified. Samsung emphasizes speed while SK Hynix pushes larger capacity, and work on the next generation, HBM4E, is already underway.
💬 Comments
No comments yet. Be the first!