Samsung foundry nears profitability via Nvidia Groq chip production

Samsung Electronics is poised to potentially return its foundry unit to profitability this year following the mass production of Nvidia’s Groq 3 LPX inference accelerator. Samsung is serving as the exclusive manufacturer for the processors, utilizing its 4-nanometer fabrication line.

Announced at the Hot Chips 2026 conference, the Groq 3 LPX is a rack-scale system containing 256 language processing units (LPUs) specifically designed for artificial intelligence inference. The hardware is intended to work alongside Nvidia’s Vera Rubin platform, where GPUs manage training workloads while the Groq silicon handles inference tasks for AI agents.

Nvidia CEO Jensen Huang noted that inference serves as the “growth engine of AI,” stating that the LPX enables ultrafast token generation and improved throughput and efficiency. Market analysts suggest this production ramp could lead to a meaningful financial inflection point for Samsung’s contract chipmaking business as early as the third quarter.