Groq has reportedly requested Samsung Electronics' foundry division to increase wafer production for its inference AI chips, reflecting the booming market demand. Industry sources expect Samsung's foundry business to boost profitability by deepening collaboration with Groq and scaling up output.
According to Chosun Biz, Groq plans to raise its 2025 wafer orders from about 9,000 to approximately 15,000 units at Samsung Foundry.
Analysts note that 2025 volumes mainly cover sample chips used to validate real-world AI inference applications, while large-scale commercial mass production is set to begin in 2026.
Although Groq's order volume remains modest, Samsung Foundry is aggressively pursuing these contracts to establish a foothold in the inference AI chip segment. Besides Groq, South Korean startup HyperAccel also relies entirely on Samsung Foundry for processor manufacturing. Currently, Samsung uses its 4nm process node to produce AI chips for both companies.
Insiders reveal that Samsung's 4nm process for Groq AI chips incorporates significant enhancements to improve performance. Given the higher cost of this node and strong industry demand for 4-5nm processes, securing such orders is strategically important for Samsung as it competes with TSMC.
Meanwhile, anticipation is rising around Nvidia's planned unveiling of an inference-dedicated chip based on Groq AI designs at GTC 2026.
Market watchers predict Nvidia will adopt Groq's architecture but replace high-bandwidth memory (HBM) with static random-access memory (SRAM). This shift could reportedly accelerate data transfer speeds, enhance power efficiency, and reduce chip costs.
The combined momentum from Nvidia's AI chip initiatives and Groq's expanded production signals robust growth ahead for the inference AI chip market.
Article translated by Charlene Chen and edited by Jack Wu
