Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) is drawing attention to the evolving competition in the AI server CPU market after providing its most detailed look yet at its Vera CPU architecture, with Bank of America highlighting a growing debate over how AI infrastructure performance should be measured.
In a note to clients, the bank wrote that the emerging competition centers on two different approaches to agentic AI. Nvidia is focused on reducing the "time-to-complete-an-agent," while Advanced Micro Devices (NASDAQ: AMD) is emphasizing the "number-of-agents-per-rack." Bank of America estimates the server CPU total addressable market could expand to approximately $170 billion by 2030.
According to the analysts, Nvidia's Vera architecture combines 88 custom ARM-based Olympus cores with 1.2TB/s of memory bandwidth and 3.4TB/s of on-die fabric bandwidth. The company argues that as agentic AI workloads become more common, CPU latency, memory responsiveness and GPU utilization will become increasingly important, making faster per-core performance a key advantage.
Bank of America noted that Nvidia's design also reflects a broader co-design strategy across its AI platform, including Rubin GPUs, networking and storage technologies.
The report comes ahead of AMD's AI Day on Thursday, where Bank of America expects the company to present its own view of AI infrastructure performance. Rather than focusing on faster individual agents, AMD is expected to emphasize rack-level throughput and the ability to support more concurrent AI workflows.
The analysts noted that AMD estimates its current EPYC 9965 (Turin) processor delivers roughly 2.4 times the rack-level throughput of Nvidia's Vera baseline in a modeled 100-kilowatt deployment, while its upcoming EPYC 6 (Venice) platform is projected to deliver up to 3.3 times the throughput.
Beyond performance metrics, Bank of America said the rivalry also reflects a broader architectural debate between ARM and x86 processors. Nvidia's approach suggests processor architecture is less important if higher single-thread performance improves AI agent execution, while AMD and Intel are expected to argue that x86 retains an advantage through its long-established software ecosystem and compatibility with enterprise applications.
Bank of America wrote that the central question for investors is whether future AI deployments will be constrained more by the speed at which an individual AI agent completes a task or by the number of AI agents that can operate simultaneously within a fixed power budget. The firm expects AMD's upcoming event to focus on shaping that discussion rather than benchmark comparisons alone.
Nvidia shares added 3% at $214, to mark an almost 15% gain so far this year.
