# Nvidia says LPX inference racks have entered full production

_Tuesday, August 25, 2026 at 10:30 AM EDT · Infrastructure · Latest · Tier 2 — Notable_

![Nvidia says LPX inference racks have entered full production — DCD](https://media.datacenterdynamics.com/media/images/Nvidia_Groq_LPU.2e16d0ba.fill-1200x630.png)

Nvidia's LPX racks for AI inference accelerators have entered full production, according to Data Center Dynamics. The liquid-cooled platform holds 256 Groq 3 language processing units and provides 640 Tbps of scale-up bandwidth, alongside BlueField-4 DPUs, Vera CPU racks, STX storage servers and Spectrum-6 Ethernet.

Nvidia cited benchmarking results of 3,400 output tokens per second and said LPX is complementary to its NVL72 platform for latency-sensitive inference. The platform is expected to be available later this year, with Nebius among early adopters.

## Sources

- [DCD](https://www.datacenterdynamics.com/en/news/nvidias-ultra-low-latency-ai-inference-lpx-racks-hit-full-production/)

---
Canonical: https://techandbusiness.org/newswire/1Wfk3yYS83ubkTSSZg5Rur
Retrieved: 2026-08-26T20:46:42.894Z
Publisher: Tech & Business (techandbusiness.org)