 |
| Nvidia Next-Generation Chips Enter Full Production, CEO Jensen Huang Announces |
Nvidia next-generation chips are now in full production, CEO Jensen Huang confirmed during his keynote at CES 2026 in Las Vegas. The new Rubin platform promises up to five times the AI performance of previous-generation chips for chatbots and generative AI applications.
The Vera Rubin system integrates 72 Nvidia GPUs and 36 central processors per server, scalable into pods with over 1,000 chips. Huang highlighted a proprietary data format that boosts token-generation efficiency by 10x despite only a 1.6x increase in transistor count.
Designed for inference workloads, the chips feature new "context memory storage" to accelerate responses in extended AI conversations. Early adopters include CoreWeave, Microsoft, Oracle, Amazon, and Google, Nvidia revealed.
The company also unveiled new networking switches with co-packaged optics to enhance large-scale AI cluster connectivity, competing directly with Broadcom and Cisco. Additionally, Nvidia open-sourced its Alpamayo autonomous driving software and its training data.
Recent acquisition of AI startup Groq’s team and technology aims to expand Nvidia’s product range without disrupting its core business, Huang said. Meanwhile, demand remains high for older H200 chips in China, pending U.S. export approvals.
With growing competition from AMD and in-house chips from Google and Meta, Nvidia is betting big on its next-generation chips to maintain its AI dominance in both training and inference markets.
