Nvidia’s Next‑Gen AI Chips Are Now in Full Production
CEO Jensen Huang announced at CES 2025 that Nvidia’s upcoming chip generation—the Rubin platform—is fully in production and already being tested by AI firms.
These chips can deliver 5× the AI computing performance of the previous generation when running chatbots and similar applications.
What the Rubin Platform Includes
Rubin consists of six different Nvidia chips.
The flagship configuration includes:
72 GPU units
36 new CPU units
Systems can be scaled into “pods” containing 1,000+ Rubin chips.
How Nvidia Achieved the Performance Jump
Huang said the gains come from a new proprietary data format Nvidia hopes the industry will adopt.
Despite only a 1.6× increase in transistor count, performance jumps dramatically due to this new data approach.
Competition Is Intensifying
Nvidia still dominates AI model training, but faces growing competition from:
AMD
Google (a customer and a rival)
Meta and others collaborating on alternative AI chips.
New Technologies Introduced
Context Memory Storage: Helps chatbots respond faster during long conversations at massive scale.
Co‑packaged optics networking switches: Compete with Broadcom and Cisco for linking thousands of machines efficiently.
Self‑Driving Car Software
Nvidia is releasing Alpamayo, a decision‑making system for autonomous vehicles.
Both the models and the training data will be open‑sourced to improve transparency and trust.
Strategic Moves
Nvidia recently acquired talent and chip tech from Groq, strengthening its AI hardware capabilities.
The company is also positioning Rubin to outperform older chips like the H200, which the U.S. government now allows to be sold to China.
