Samsung and Nvidia shared big news about a new AI chip, Groq 3, at the GTC 2026. This chip is created by Groq but will be manufactured by Samsung using its 4nm technology.

This announcement is important because Nvidia had stopped working with Samsung some years ago and chose TSMC to produce its chips instead. Now, Nvidia is returning to Samsung, which is a positive sign for Samsung’s chip business.

Groq 3 is not like most AI chips. Many AI chips today depend on HBM to achieve fast performance. However, Groq 3 uses a different method. It has a large amount of built-in memory, SRAM. Because of this, it can move data very quickly, reaching speeds up to 150TB per second. This makes it much faster than many chips that rely on HBM.

Source – Nvidia Blog

This chip is specially made for AI tasks like chatbots and real-time responses. It works very well for tasks where quick answers are needed, such as talking to AI assistants. It is also useful for agentic AI, where AI systems can perform tasks on their own step by step.

On the other hand, Nvidia’s Rubin chips are more general-purpose and better for training AI models and handling large data tasks. Both types of chips can work together in future systems.

The post Samsung to produce Groq 3 AI inference chip for Nvidia appeared first on Sammy Fans.