Samsung tied to competitor inference chips
Samsung is publicly linked to manufacturing competitor inference silicon — Samsung showcased an Nvidia-made inference chip at GTC that uses a 4nm process, according to Reuters, and social posts say Samsung is set to manufacture Groq LPUs too reported reported. The same week the company faces a worker strike vote that union leaders warn could disrupt chip supply — a near-term supply risk for customers and partners reported.
Groq contracted Samsung Foundry in August 2023 to build its next‑generation LPU family at Samsung’s Taylor, Texas plant [announced] (prnewswire.com). Nvidia’s GTC keynote tied the recent inference design to Groq’s tech and projected at least a $1 trillion revenue opportunity for inference hardware through 2027 [said] (srnnews.com). Samsung’s foundry roadmap for those devices uses a 4nm SF4X-class node at the Taylor site, a capability Groq and Samsung public materials have highlighted for their LPU roadmap [reported] (sammobile.com). Shares of Samsung Electronics jumped as much as 5% after Jensen Huang flagged the tie‑up at GTC, while Nvidia’s Vera Rubin platform announcement listed Groq‑derived LPUs among chips Nvidia says are going into production for agentic and inference-scale deployments [market reaction] (msn.com). About 90,000 unionized Samsung workers are eligible to vote in the strike ballot (from a roughly 125,000 South Korea workforce), and union leaders say an 18‑day action beginning May 21 could hit Pyeongtaek output and therefore near‑term memory and accelerator wafer throughput [vote numbers and timing] (money.usnews.com). Industry reporting says Groq has scaled its Samsung wafer orders—raising planned output from roughly 9,000 wafers to about 15,000 wafers in recent cycles—to support higher LPU volume, a change that would increase demand on Samsung Foundry capacity if sustained [wafer volumes] (biz.chosun.com).
