Morgan Stanley report analyzes NVIDIA vs. AMD in the intelligent CPU standards battle

Written by: Rita

Tide Guide

NVIDIA and AMD are competing to define the standards in a $170 billion agent CPU market.

Under the AI agent wave, server CPU architectures are diverging. NVIDIA advocates for faster cores, arguing that single-core performance determines system limits. AMD, on the other hand, focuses on more cores, emphasizing that concurrent throughput is what truly matters.

NVIDIA last week unveiled its Vera CPU architecture, featuring 88 custom ARM cores and 1.2 TB/s memory bandwidth, built on a monolithic compute die design. The rationale is that agent-based AI requires frequent interaction between CPU and GPU, with each cycle dependent on the completion of the previous step, making single-core performance critical to overall response speed. AMD is set to respond this Thursday at its AI Day, with market expectations pointing toward a strategy focused on "more cores." Currently, under a 100kW deployment scenario, the EPYC 9965 achieves 2.4 times the rack-level throughput of Vera.

Bank of America’s assessment of this strategic rivalry is clear: whoever defines the next generation of industry standards will win.

NVIDIA's Vera CPU differs from traditional server processors that focus on "core stacking." While 88 cores are not particularly impressive in the server CPU market, NVIDIA emphasizes "maximum single-thread performance." Vera offers a memory bandwidth of 1.2 TB/s and an on-chip interconnect bandwidth of 3.4 TB/s, with all design elements aimed at enhancing single-core efficiency.

NVIDIA's logic is that agent-based AI differs from one-time, massive parallel computing, instead involving repeated interaction cycles between CPU and GPU. Steps such as tool invocation, code execution, retrieval, and orchestration each depend on the completion of the previous step. Insufficient single-core performance will slow down the entire agent's response time, causing the GPU to wait and thereby reducing the overall utilization of the AI factory.

AMD’s design philosophy is fundamentally different. AMD believes that production-grade AI more closely resembles a distributed software platform, where components such as databases, APIs, vector stores, orchestration engines, caches, and middleware run in parallel. In this scenario, system bottlenecks are determined by the number of concurrent workflows that can be supported within a fixed power budget. The EPYC 9965 delivers 2.4 times the rack-level throughput of Vera in a 100kW deployment, and the next-generation EPYC 6 is expected to reach 3.3 times.

Beyond the core number debate of "faster and more," there is another, more subtle yet equally important front: the ecosystem battle between x86 and ARM instruction sets.

NVIDIA Vera is based on the ARM architecture. NVIDIA believes that as long as the microarchitecture is excellent, instruction set differences are not significant. Can ARM handle AI workloads? The answer is yes. Can ARM run enterprise software? If its performance fully surpasses x86, the software ecosystem will follow.

AMD and Intel clearly hold different views. Their argument is straightforward: agent AI is expanding from model inference to enterprise workflows—scenarios such as databases, middleware, security platforms, and enterprise applications, which for decades have been optimized for the x86 architecture. The feasibility of ARM replacing x86 cannot be proven solely by a few AI test results.

AMD's AI 2026 Day on Thursday will be the first public response in this roadmap battle.

Bank of America expects AMD not to compete solely on benchmark scores, as that would fall into NVIDIA's framework of evaluating "single-core performance." AMD needs to redefine the competitive landscape by shifting from "single-core performance" to "agent throughput in real-world production environments."

Bank of America believes this is the core of the debate. The deciding factor is not which technology is superior, but which industry-accepted standard ultimately prevails.

The insight in Bank of America's report lies in framing the CPU technology roadmap battle as a struggle for control over industry standards.

The computational requirements of agent AI differ from those of traditional AI training. Traditional training is primarily based on large-scale parallelism, whereas agents combine serial loops with concurrent scheduling. Which architecture is superior depends on how the evaluation criteria are defined. If the industry prioritizes "response latency per agent," NVIDIA's approach is more reasonable. If the industry prioritizes "number of agents supported per rack," AMD holds an advantage.

AMD’s AI Day on Thursday will be a key milestone in this debate. However, Bank of America’s implied view is that the outcome of this technological rivalry won’t be decided in the short term. The ultimate winner may not be the one with higher benchmark scores, but the company that succeeds in driving industry adoption of its own defined metrics.

For investors, the investment value of this debate lies not in determining which path is right or wrong, but in understanding each company’s strategic bet: NVIDIA is betting that agent-based AI is extremely sensitive to latency, while AMD is betting that the core demand in production environments is concurrency density. Both paths could prove valid, depending ultimately on how AI applications actually evolve.

Bank of America has given both companies a "Buy" rating, with a target price of $350 for NVIDIA and $620 for AMD. This indicates that Bank of America views this competition not as a zero-sum game, but rather as a scenario where both companies can achieve growth through differentiated strategies, albeit along different paths.

Disclaimer

This article is a summary and interpretation by Chaoxiang Research of a third-party brokerage research report (Bank of America Securities, July 22, 2026). The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution, not the views of Chaoxiang Research, nor do they constitute any investment advice.

The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.