Bank of America: The AI agent CPU standards battle—NVIDIA “faster” vs AMD “more”

Deep Tide TechFlow message. According to Tide direction research, a Bank of America report dated July 22 said that Nvidia and AMD are waging a defining battle in the field of agent-focused CPUs. Nvidia released its Vera CPU architecture last week, with 88 custom ARM cores and 1.2TB/s of memory bandwidth. It argues that single-core performance determines agent response speed, emphasizing “faster.” AMD responded on Thursday at its AI Day. Its EPYC 9965 in a 100kW deployment already has rack-level throughput 2.4 times that of Vera, arguing that concurrent throughput determines production-grade AI efficiency, emphasizing “more.” Bank of America believes that the underlying issue in the x86 vs. ARM battle is also just as critical: x86’s decades of optimization moats in the enterprise software stack are not something that can be replaced overnight.

Bank of America expects AMD to redefine the benchmarks rather than compete on benchmark scores, and that the real winner will be the company whose framework the industry is willing to adopt. Bank of America maintains buy ratings for Nvidia and AMD, with target prices of $350 and $620, respectively. It believes this is not a zero-sum game—both companies can win, just with different paths.