Micron Explores Near-GPU NAND Flash to Run Bigger LLMs - TechPowerUp

Micron is exploring near-GPU NAND flash technology to support larger-scale large language models (LLMs), marking a significant development in AI computing infrastructure. As LLM parameter sizes contin... Read original →

Micron’s exploration of near-GPU NAND flash targets a critical bottleneck in AI computing, reshaping data center architecture by integrating storage closer to compute units. This move accelerates the convergence of memory and processing, enhancing efficiency for large language model training. Upstream, semiconductor fabrication and materials must evolve to meet higher performance demands; downstream, AI chipmakers like NVIDIA and AMD may integrate storage compute into their designs. Geopolitical tensions, especially U.S. export controls on advanced tech, heighten supply chain risks, pushing companies to restructure operations. Competitors such as Samsung and Kioxia are likely to accelerate investments in similar near-memory computing technologies to maintain market leadership. Within 12–24 months, near-GPU storage will become a decisive factor in AI infrastructure, driving industry-wide shifts toward low-latency, high-bandwidth solutions and establishing new competitive barriers.