Samsung Kicks Off 4nm AI ASIC Production in Texas
Samsung has reportedly begun producing next-generation AI ASIC chips on its 4nm node in Texas. The move signals a significant ramp-up of advanced, onshore semiconductor manufacturing in the U.S. to meet surging global AI demand.
The Taylor, Texas plant is a centerpiece of Samsung's expanded US investment, now totaling over $40 billion, bolstered by up to $6.4 billion in direct funding from the CHIPS and Science Act. This initiative is set to create a comprehensive semiconductor ecosystem in Central Texas, including two new logic fabs and an R&D facility. The first publicly named customer for the 4nm process at the Texas fab is AI startup Groq, founded by former Google engineers. Groq's next-generation Language Processor Units (LPUs) will be manufactured by Samsung, a significant move for the startup that has raised $367 million to date and aims to challenge established players in AI inference. While initially slated for 4nm production to begin as early as 2024, the timeline has faced delays with mass production now anticipated in 2026. Reports suggest these delays are due to a mix of factors including evolving market demand and yield adjustments, with Samsung potentially shifting the new fab's focus to its more advanced 2nm process to better compete with rivals like TSMC. This onshore production move taps into the booming market for custom silicon, where the AI chip market is projected to reach $154 billion by 2030. The trend is driven by the need for specialized processors that offer greater performance and energy efficiency for specific AI workloads compared to general-purpose GPUs. Samsung's improved 4nm yields, reportedly reaching 70-80%, are attracting other customers beyond Groq, including a reported $100 million order from US-based AI startup Tsavorite Scalable Intelligence for its all-in-one OPU. This demonstrates a growing market for fabless AI companies seeking to bring custom designs to life. The new fab places Samsung directly in the center of the strategic "build vs. buy" decision facing hyperscalers and large enterprises. While companies like Google with its TPUs and AWS with Trainium are developing their own custom silicon, the immense cost and complexity of manufacturing create a massive opportunity for advanced foundries like Samsung to serve the rest of the market.
