G

Groq LPU (GroqChip)

Hardware

by Groq

The Language Processing Unit (LPU) — Groq's own inference accelerator chip (GroqChip), deployed 256 per rack across Groq's own data centers. Not sold or licensed as standalone hardware today: access comes bundled into GroqCloud or into the dedicated regional deployments Groq itself builds and operates for governments and telecoms.

Find alternatives to Groq LPU (GroqChip)

No price is published for the chip or a rack on its own. groq.com/products/ names exactly one rack-scale SKU, the NVIDIA Groq 3 LPX (256 next-generation LPU accelerators plus NVIDIA's Vera Rubin), and marks it “When released” — not available today. Every other LPU deployment described on groq.com (Bell Canada's sovereign AI network, Aramco Digital's Saudi Arabia data center, the US Department of Energy partnership) is Groq building and operating the infrastructure itself and selling inference capacity, not selling or leasing the hardware to the customer. The only route to LPU compute with a published rate is GroqCloud (see the groqcloud record).

What it does

• Accelerator silicon
• Model inference

Sources

What it does

Official documentation · 12 Sep 2026

“Groq AI inference infrastructure, specifically GroqCloud™, is powered by the Language Processing Unit (LPU), a new category of processor... LPUs run Large Language Models (LLMs) and other leading models at substantially faster speeds.” Same two capabilities GroqCloud's own record already carries (accelerator-silicon, model-inference) — not widened, since the fix here is the missing record, not a score lift.

Sold within

Official documentation · 12 Sep 2026

“When released, each NVIDIA Groq 3 LPX rack connects 256 next-generation LPU accelerators to NVIDIA’s Vera Rubin to deliver low-latency, large-context inference for production AI.” That is the only named rack SKU on Groq's own current products page and it is explicitly future (“When released”), so it cannot source a sold_within.access: standalone claim today. Corroborating that Groq operates rather than sells its hardware: groq.com/newsroom/groq-becomes-exclusive-inference-provider-for-bell-canadas-sovereign-ai-network (“Groq Becomes Exclusive Inference Provider for Bell AI Network”, will “power Bell AI Fabric”) and groq.com/newsroom/aramco-digital-and-groq-announce-progress-in-building-the-worlds-largest-inferencing-data-center-in-saudi-arabia (facility offered “through our digital marketplace, nawat, in a flexible ‘as-a-Service’ model”). included-with-parent under groqcloud is the honest reading: a GroqCloud customer already obtains LPU-powered inference at no separate charge, which is the only route to this hardware with a published rate.

Deployment

Official documentation · 12 Sep 2026

Inferred from sold_within: since access is included-with-parent under groqcloud, this record has no deployment path separate from GroqCloud's own, which is saas.

Also from Groq

• GroqCloud Model API

  Hosted inference service that serves open-weight models on Groq's own LPU hardware, billed per token.

Open Groq in the directory

Something wrong here? Send a correction — quote this product id: groq-lpu.