At the FMS 2026 memory event held at the Santa Clara Convention Center in California on the 6th, NVIDIA senior researcher Vikram Mailsodi predicted that the number of AI agents will surpass the world's population, with agents rather than humans becoming the primary entities questioning AI. He noted that increased agent requests require significantly higher processing speeds for databases and applications, demanding a complete overhaul of the technology layer. Samsung Electronics DS Division Senior Vice President Ki Yang-seok stated that while memory prices were previously calculated per GB, AI now offers an opportunity to change pricing methods, with processing performance growing in importance. He mentioned engineers dividing tasks between in-house AI and expensive ChatGPT, ensuring seamless operation between the two systems. Micron Vice President Jeremy Warner echoed that focus will shift from price per Gb or GB to price per I/O, with performance and power efficiency determining architecture success. Panelists agreed that HBM alone will become insufficient, with Ki predicting a 'hierarchical memory' structure linking DRAM and NAND flash. FMS Chairman Tom Coughlin argued that current AI infrastructure has reached limits handling with DRAM alone, asserting HBF has emerged as a memory tier to resolve this issue.