AMD's hidden AI weapon may finally be exposed

AMD 's artificial intelligence story has mostly been judged against Nvidia 's dominance in graphics processing units.

BofA Global Research lifted its price target for Advanced Micro Devices (AMD) to $620 from $550, saying the next phase of AI infrastructure may not rely solely on GPUs. As agentic AI workloads generate new bottlenecks within data centers, central processing units are becoming increasingly crucial, the firm argues.

The target represents almost 11% upside from AMD's price of $557.89 at the time of the report. That is hardly a moonshot call.

It's a more important kind of call: a bet investors may be underestimating how much AI demand may increase AMD's server CPU business while its GPU roadmap ramps.

That includes AMD's next-generation Venice CPU, MI455X GPU, and Helios rack-scale solution, which BofA said it expects to benefit from strong EPYC server CPU demand and increased supply visibility.

That's why AMD's Advancing AI event on July 23 is so important.

Investors are now asking more than whether AMD can catch Nvidia in GPUs. The question is whether AMD can utilize its CPUs, GPUs, and rack-scale systems to be a more complete AI infrastructure supplier.

BofA said AMD is well positioned because of its CPU and GPU demand, supply visibility, and upcoming AI product launches.

One issue dominates the AI chip market: who can beat Nvidia?

For AMD, the question has traditionally been about GPUs. Nvidia ( NVDA ) is still the king of AI accelerators, and AMD has been striving to establish that its Instinct portfolio can grab major hyperscale demand.

In agentic AI, CPUs are responsible for such things as tool calls, code execution, orchestration, memory management, and system coordination. If those operations lag, pricey GPUs may just lie idle.

The company already has a strong franchise in server CPUs with EPYC. If artificial intelligence systems need more and more CPU content with each generation, then AMD's server business could be a bigger portion of the AI trade than investors who just think about GPUs understand.

BofA currently anticipates the server CPU market growing to almost $170 billion by 2030 from around $35 billion in 2025. Over time, the company anticipates AI-related CPUs to make up most of that market.

And the AI story doesn't have to be just about AMD gaining share in GPUs. It might also help if the AI data center design itself becomes more CPU-heavy.

BofA anticipates AMD's sixth-gen EPYC Venice CPUs to prolong the company's high-core-count advantage, which may be a big deal for agentic artificial intelligence systems requiring more parallel computing and higher rack-level throughput.

The CPU discussion is more critical than ever as AI workloads evolve.

BofA frames the battle as a debate between high-frequency CPUs and high-core-count CPUs.