Back to today’s briefing

Story perspectives

Nvidia Says Groq 3 LPX Accelerator Hits Full Production

8/25/2026

27 3 Full Breakdown

Summary

Story summary

Nvidia announced the Groq 3 LPX accelerator entered full production, funded by a $20 billion Groq licensing deal signed Dec. 24 2025.
It delivers 3,400 tokens per second on Gemma 4 31B with a 100,000-token context, four times faster than rivals.
Nebius became first AI cloud to adopt accelerator via its Token Factory service, with Nvidia shipping racks later in 2026.

History

Historical Context

In 2015 Sundar Pichai became Google CEO, guiding the Alphabet subsidiary toward AI hardware after the 2014 $400 million DeepMind acquisition. Building on its own server farms, Google launched the first Tensor Processing Unit in 2018, delivering 15 petaflops and sparking a data-center chip race that now includes rivals such as Nvidia.

1 of 2

Story summary

Nvidia announced the Groq 3 LPX accelerator entered full production, funded by a $20 billion Groq licensing deal signed Dec. 24 2025.
It delivers 3,400 tokens per second on Gemma 4 31B with a 100,000-token context, four times faster than rivals.
Nebius became first AI cloud to adopt accelerator via its Token Factory service, with Nvidia shipping racks later in 2026.

1 / 2