Samsung Foundry Begins Mass Production of NVIDIA Inference Accelerator 'Groq3'... Is Profitability Turning Point Near?

Groq3 Secures First Customer and Begins Full-Scale Mass Production
Samsung Foundry to Handle All Manufacturing
Advanced Process Competitiveness Demonstrated, Profitability Expected

NVIDIA's inference-only accelerator, 'Groq3 LPX', which is produced by Samsung Electronics, has secured customers and officially entered mass production.

According to NVIDIA on August 26, the company announced on August 24 (local time) that mass production of Groq3 LPX had begun. Groq3 LPX is a system that combines 256 inference-only Language Processing Units (LPUs) into a rack the size of a cabinet, and is featured in NVIDIA's next-generation artificial intelligence (AI) platform, 'Vera Rubin'.

NVIDIA explained that Groq3 LPX "enables agentic tasks such as coding to be performed in minutes rather than hours." NVIDIA has designed its AI systems so that traditional Graphics Processing Units (GPUs) handle general-purpose tasks such as model training, while the fast-responding Groq3 LPX is responsible for the inference phase, where a trained AI generates answers.

Jensen Huang, CEO of NVIDIA, Hwang Sangjun, Vice President in charge of memory development at Samsung Electronics (left), and Han Jinman, Head of the Foundry Division (right), are posing for a commemorative photo at the Samsung Electronics exhibition booth during 'GTC 2026' held on March 16 (local time) in San Jose, California, USA. Samsung Electronics.

Coinciding with mass production, NVIDIA has also secured a client. The company noted that the first customer for Groq3 LPX is Navius, an AI cloud operator. Jensen Huang, CEO of NVIDIA, emphasized, "Inference is the growth engine of AI," and added, "At a time when global demand for AI computation is accelerating, Groq3 LPX will once again bring a massive leap forward in AI throughput, efficiency, and responsiveness."

All Groq3 LPX units will be manufactured by the Foundry Division of Samsung Electronics. Groq3 LPX is produced at the Pyeongtaek Campus using Samsung Electronics’ 4nm (1nm = one billionth of a meter) process. With Groq3 LPX now entering full-scale mass production, it is expected to prove Samsung Foundry’s advanced process competitiveness. If NVIDIA’s customers utilizing Groq3 LPX for AI inference continue to increase, Samsung Electronics will also be able to expand its production volume of AI accelerators.

The Samsung Electronics foundry is rapidly securing global clients, focusing on 4nm processes. Last month, Samsung Electronics signed a memorandum of understanding (MOU) with Broadcom to cooperate on memory and foundry businesses, with total collaboration valued at 200 billion dollars over five years through 2030. In addition, Samsung Electronics plans to begin production of Tesla's 'AI5' chip using its new 2nm process at the Taylor, Texas plant in November or December.

The full-scale mass production of Samsung Electronics Foundry’s advanced processes is fueling expectations for a turnaround into profitability. Samsung Electronics’ foundry division has posted multi-trillion won deficits in the past, but analysis suggests the company narrowed losses in the second quarter of this year through higher utilization rates and better yields in its advanced process lines. Analysts in the securities industry predict Samsung Electronics could return to profitability either by the end of this year or next year.

This content was produced with the assistance of AI translation services.