NVIDIA and AMD go head-to-head—who is setting the new standard for AI agent CPUs

Author: Rita; Source: Deep Tide

Tide Guide Reading

NVIDIA and AMD are fighting over who will define the standards in the intelligent agent CPU market, which is valued at around $170 billion.

Amid the agentic AI wave, server CPU architecture roadmaps have begun to diverge. NVIDIA argues for faster cores, saying that single-core performance determines the system ceiling. AMD, meanwhile, emphasizes that it’s concurrency throughput that matters most.

Last week, NVIDIA released its Vera CPU architecture, featuring 88 custom ARM cores, 1.2TB/s of memory bandwidth, and a single-chip compute die design. Its logic is that agentic AI requires repeated interaction between the CPU and GPU—each loop depends on what the previous step completed—so single-core performance directly determines the overall response speed. AMD will present its response at the AI Day on Thursday, and the market expects it to focus on the “more cores” route. At present, in 100kW deployment scenarios, EPYC 9965’s rack-level throughput has already reached 2.4 times that of Vera.

Bank of America’s assessment is clear: whoever can define the industry’s next-generation yardstick will win.

Two technical routes, two design philosophies

NVIDIA’s Vera CPU is different from conventional server processors that simply “stack more cores.” With 88 cores, it isn’t standout in the server CPU space, but NVIDIA is betting on “maximum single-thread performance.” Vera’s memory bandwidth reaches 1.2TB/s, and its on-die interconnect bandwidth is 3.4TB/s—every design choice is aimed at improving single-core execution efficiency.

NVIDIA’s logic is that agentic AI differs from one-time large-scale parallel computing and is instead more of a repeated interaction loop between the CPU and GPU. In steps such as tool calls, code execution, retrieval, and orchestration, each step depends on the previous one being completed. If single-core performance isn’t sufficient, it will slow the agent’s overall response speed, causing the GPU to wait and thereby lowering utilization across the entire AI factory.

AMD’s design philosophy is sharply different. AMD believes production-grade AI is closer to a distributed software platform, where components such as databases, APIs, vector storage, orchestration engines, caches, and middleware run in parallel. In this scenario, the system bottleneck is the number of concurrent workflows the system can carry within a fixed power budget. In 100kW deployments, EPYC 9965’s rack-level throughput is 2.4 times that of Vera, and the next-generation EPYC 6 is expected to increase that to 3.3 times.

x86 vs ARM: the hidden battle for the software ecosystem

Besides the core-count dispute of “faster and more,” there’s another, more covert front of equal importance: the choice of instruction set—the ecosystem battle between x86 and ARM.

NVIDIA Vera is built on ARM architecture. NVIDIA believes that as long as the microarchitecture is good enough, the instruction-set differences don’t matter. Can ARM handle AI workloads? The answer is yes. Can ARM run enterprise software? If performance comprehensively surpasses x86, then the software ecosystem will migrate accordingly.

AMD and Intel clearly hold a different view. Their argument is direct: agentic AI is extending from model inference to enterprise-grade workflows. In use cases such as databases, middleware, security platforms, and enterprise applications—over the past several decades—solutions have been optimized based on x86 architecture. The feasibility of replacing x86 with ARM cannot be proven by only a few AI test results.

AI Day is approaching—AMD gets its response window

AMD’s AI 2026 Day on Thursday will be the first public response milestone in this route-versus-route contest.

Bank of America expects that AMD will not fight solely by trading off benchmark scores, which would put it into NVIDIA’s “single-core performance” evaluation framework. AMD needs to redefine the competitive dimensions—shifting from “single-core performance” to “the capacity of agentic systems in real production environments.”

Bank of America believes this is the core of the debate. The deciding factor isn’t a straightforward comparison of technical superiority, but which yardstick the industry ultimately accepts and adopts.

Tide Perspective

The insight in Bank of America’s report lies in reframing what looks like a server CPU technology-route dispute as a struggle for “standard-setting power” in the industry.

The compute requirements of agentic AI differ from those of traditional AI training. Traditional training is dominated by large-scale parallelism, while agentic AI is a hybrid of serial loops and concurrent scheduling. Which architecture is better depends on how the evaluation criteria are defined. If the industry’s core metric is “single-agent response latency,” NVIDIA’s route is more reasonable. If the industry’s core metric is “single-machine agent capacity,” then AMD has the advantage.

AMD’s AI Day on Thursday will be an important milestone in this debate. But Bank of America’s implicit judgment is that the route competition won’t be decided in the short term. The ultimate winner may not be the one with higher benchmark scores, but the vendor that can drive the industry to accept the benchmarks defined by itself.

For investors, the investment value of this debate isn’t about deciding who is right or wrong on the route—it’s about understanding each company’s strategic bet. NVIDIA is betting that agentic AI is extremely sensitive to latency, while AMD is betting that the core need in production environments is about concurrency density. Both routes have plausible grounds; the determining factor is how agentic AI applications actually evolve.

Bank of America gives both companies a Buy rating: NVIDIA’s target price is $350, and AMD’s target price is $620. This indicates Bank of America believes the competition isn’t a zero-sum game—both companies can achieve growth with differentiated routes, though their growth paths differ.

Disclaimer

This article is compiled and interpreted by Deep Tide Research of a third-party brokerage research report (Bank of America Securities, July 22, 2026). The ratings, target prices, earnings forecasts, and related judgments cited in the text are the views of the brokerage’s analysts only; they represent the position of their affiliated institution, not the views of Deep Tide Research, and do not constitute any investment advice.

There are risks in the market; decisions must be made independently. This article should not be used as a basis to buy or sell any securities.