AMD acquired the young startup Mext on June 15, 2026, two months after it launched software that uses artificial intelligence models to anticipate servers’ memory needs. Its technology moves data between DRAM and NVMe flash storage before the operating system or applications request it. According to AMD, Epyc-based servers could replace more than half of their DRAM with flash while retaining 95% to 99% of their performance. An internal deployment spanning 86 to 145 hosts added about 58 terabytes of capacity for $601,000, compared with $3.3 million for equivalent DRAM capacity. AMD targets cumulative savings exceeding $60 million by the end of 2027 through the deployment of the technology across thousands of servers.
Source: Read the original article
