Groq Requests More AI Chip Capacity From Samsung

According to Korean media reports, AI chip startup Groq has recently asked Samsung Electronics' foundry division to increase production capacity, aiming to raise its 2025 AI inference chip output from about 9,000 wafers to roughly 15,000 wafers.

Samsung is currently manufacturing AI chips for both Groq and the Korean startup HyperAccel using its 4nm process technology. HyperAccel's processor is fully produced by Samsung's foundry business. Industry sources noted that Samsung has applied several process improvements during the production of Groq's chips to help boost overall performance. With strong demand and relatively high pricing for advanced 4–5nm nodes, securing such orders is considered strategically important for Samsung as it competes with TSMC in the advanced semiconductor manufacturing market.

Analysts say that most of the chips produced in 2025 will likely be sample units used to verify real-world AI inference performance. Large-scale production is expected to begin in 2026, paving the way for broader commercial deployment. Although Groq's current order volume is still relatively small, Samsung's foundry division is actively pursuing similar projects to strengthen its position in the growing AI inference chip segment.

At the same time, NVIDIA is expected to unveil an inference-focused chip based on Groq's architecture at the GTC 2026 conference. Reports suggest the design could use static random-access memory (SRAM) instead of high-bandwidth memory (HBM). This approach may improve data transfer speed and power efficiency while also lowering chip costs, potentially accelerating the adoption of AI inference processors.

With continued investment from companies like NVIDIA and Groq, the AI inference chip market is widely expected to grow rapidly in the coming years, intensifying the competition between Samsung and TSMC in advanced semiconductor manufacturing.