Beyond Nvidia: AMD Rides the AI Agent Boom Fueled by Meta's Muse

For years, the artificial-intelligence chip trade revolved primarily around one type of processor: the GPU.

Nvidia became the defining beneficiary.

Now the rise of autonomous AI agents is highlighting another part of the data-center stack—and investors have begun paying much more attention to companies such as AMD and Intel.

AMD shares surged about 32% over a one-month period through early October 2026. On October 6, the stock gained 2.8% to close at $649.42.

The rally coincided with growing excitement around personal agents including Meta's Muse and OpenAI's Dots.

But the story is more nuanced than simply saying Meta's AI made AMD rich.

AI Agents Change the Compute Mix

Traditional generative-AI services rely heavily on GPUs because large neural networks require enormous amounts of parallel computation.

Agents introduce a different workload.

They can remain active for long periods, operate browsers, execute code, manage files, communicate with applications and perform multi-step tasks.

Ryan Shrout, president of hardware consulting firm Signal65, argues that as agents become more widely deployed, a larger share of the surrounding workload can shift toward CPUs.

That does not mean CPUs replace GPUs.

GPUs continue to perform much of the heavy model inference.

CPUs handle many of the systems tasks surrounding the model: operating systems, virtual machines, orchestration, applications, tool execution and data movement.

The AI infrastructure market is therefore becoming broader rather than simply moving from one chip category to another.

Meta's Muse Is an Early Example

Meta launched Muse as a personal AI agent on September 8.

Unlike a traditional chatbot that mainly responds to prompts, Muse is designed to act on a user's behalf and work through longer tasks.

Third-party technical inspection found at least one Muse environment running on an AMD EPYC Turin host, with a sandbox allocated two CPU cores and 8GB of memory.

That finding generated investor interest because a personal agent that gives each user persistent compute could create substantial demand for server CPUs at scale.

But it should not be overinterpreted.

Meta has not said its entire Muse fleet is exclusively powered by AMD.

A company spokesperson said Meta's system is designed to remain largely CPU-agnostic so it can source capacity flexibly.

OpenAI's Dots Shows a Similar Pattern

OpenAI launched Dots on September 29 as an agent designed to work independently on users' behalf.

Third-party benchmark results suggested that a Dots virtual environment used an AMD EPYC processor, with one observed configuration showing nine allocated CPU cores.

Again, that does not establish AMD as OpenAI's exclusive CPU provider.

The evidence demonstrates that AMD processors appear in some agentic workloads—not that every Dots session runs on the same hardware.

AMD's Strongest Evidence Comes From Meta's Broader Infrastructure

The most important AMD-Meta connection is not a reverse-engineered Muse virtual machine.

It is their official strategic partnership.

In February 2026, AMD and Meta announced an agreement covering up to six gigawatts of AMD Instinct GPU deployments across multiple generations.

The first gigawatt is scheduled to use custom MI450-based GPUs alongside sixth-generation EPYC CPUs.

AMD also says Meta has already deployed millions of EPYC processors throughout its infrastructure and will be a lead customer for its new Venice and Verano server CPUs.

That gives AMD a much broader exposure to Meta's AI growth than Muse alone.

Intel Is Benefiting Too

The CPU revival is not exclusively an AMD story.

Intel shares were reported to have gained roughly 21% over the same one-month period in which AMD rose about 32%.

Intel has been positioning its Xeon processors around agentic AI as well.

The company argues that CPUs are increasingly important for orchestration, concurrency and data movement as agents become more complex.

At the same time, hyperscalers including Amazon, Google, Microsoft and Meta are developing or deploying more custom silicon, including Arm-based processors.

That means AMD and Intel face opportunities and competitive pressure simultaneously.

Why Persistent Agents Could Consume So Many CPUs

A useful way to understand the economics is to think of a personal agent as a cloud computer assigned to a user.

An agent may need to run a browser, execute scripts, move files, talk to APIs, wait for external events and then resume work.

These are often CPU-heavy general-purpose tasks.

If a service allocates even a small number of CPU cores to each active user, millions of concurrent agents could translate into significant server demand.

However, today's configurations may not remain constant.

AI companies will continuously optimize utilization, share infrastructure more efficiently and evaluate CPUs from multiple suppliers.

So current core counts should not be mechanically extrapolated into future chip sales.

Nvidia Is Not Being Replaced

The phrase “beyond Nvidia” should not be confused with “without Nvidia.”

GPUs remain critical for training and inference.

The emerging difference is that autonomous AI systems require far more than accelerator chips alone.

They also need CPUs, memory, storage, networking, power, cooling and orchestration software.

In many agent architectures, the GPU performs the model's computationally intensive reasoning while the CPU manages the environment in which the agent acts.

That creates a larger total infrastructure opportunity.

AMD Is Building a Broader AI Stack

AMD's strategy increasingly extends beyond CPUs.

Its portfolio includes EPYC server processors, Instinct accelerators, Pensando networking, ROCm software and Helios rack-scale infrastructure.

AMD and Meta say they are co-engineering across the stack, from silicon through systems and software.

That matters because AI customers increasingly buy integrated infrastructure rather than isolated processors.

The closer AMD can tie its CPUs, GPUs and networking together, the more of an AI data center's spending it can potentially address.

The Stock Rally Needs a Date Stamp

The headline market numbers also require context.

AMD closed at $649.42 on October 6, after gaining 2.8% during the session. Its shares had risen roughly 32% over the preceding month.

By October 12, however, AMD had pulled back to $594.24.

This is why the 32% figure should be described as a historical snapshot rather than the company's current one-month return.

It also illustrates the volatility of AI-related semiconductor stocks.

Meta's Agent Push Is Still Accelerating

Meta continues to expand Muse.

On October 12, it introduced Muse Wonder and Muse Wonder Full, new frontier models designed for agentic tasks spanning coding, finance, law and health.

That suggests agentic computing is becoming a larger part of Meta's strategy rather than a one-off product experiment.

If agents become deeply integrated into consumer and enterprise workflows, the resulting infrastructure requirements could be substantial.

But the eventual winners will depend on cost, performance, efficiency, availability and software support.

Bottom Line

The rise of AI agents is broadening the semiconductor investment story beyond GPUs.

Meta's Muse and OpenAI's Dots illustrate why CPUs can become more important when AI systems remain active for extended periods and interact with real computing environments.

AMD is well positioned because Meta already operates millions of EPYC processors and has signed a major multi-generation infrastructure agreement with the company.

But several caveats matter.

Meta does not rely on a single CPU supplier. OpenAI also has multiple infrastructure options. Intel and Arm-based chips remain competitors.

And there is no public evidence showing that Muse alone has already generated a quantifiable increase in AMD revenue.

The more defensible conclusion is that agentic AI is expanding the market for general-purpose compute, giving AMD another avenue to participate in AI infrastructure alongside its GPU business.

Disclaimer: This article is for informational and educational purposes only and is not personalized advice to buy, sell or hold AMD, Intel, Nvidia or any other security.