According to Bank of America report issued on July 22, Nvidia released its Vera CPU architecture featuring 88 custom ARM cores and 1.2TB/s memory bandwidth, emphasizing that single-core performance determines AI agent response speed. AMD responded on Thursday during its AI Day, stating that its EPYC 9965 delivers 2.4 times higher rack-level throughput than Vera under 100kW deployment, arguing that concurrent throughput defines production-grade AI efficiency.
Bank of America noted that the underlying competition between x86 and ARM architectures is equally significant, with x86's decades of enterprise software optimization creating a substantial barrier. The bank maintained buy ratings for both companies with price targets of $350 for Nvidia and $620 for AMD, viewing this not as a zero-sum competition but as two different paths to success.