• Home
• Technology

AMD is regaining momentum in the artificial intelligence chip race as AI agents increasingly plan and execute tasks for hours with limited human intervention, potentially putting central processing units back in focus after years of GPU dominance led by Nvidia.

AI Agents Expand the Role of CPUs

The generative AI boom that followed the launch of ChatGPT has relied heavily on graphics processing units to train and run advanced models. AI agents, however, are changing the nature of computing workloads. GPUs remain critical for running AI models and inference, while CPUs handle the practical steps required by agents, including executing tasks, running applications and managing different processes.

The shift gives AMD and Intel an opportunity to strengthen their positions as new AI agents emerge, including Meta’s Muse and OpenAI’s Dots. User tests and inquiries have indicated that the services rely on AMD processors to run their virtual machines. Meta said its computing infrastructure is not tied to one specific processor type and that Muse is designed to work flexibly with different CPUs. OpenAI also uses processors from multiple suppliers, leaving AMD to compete with Intel and other chipmakers.

Investor Expectations Rise for AMD

AMD shares have gained about 32% over the past month, while Intel stock has risen roughly 21%, reflecting growing investor expectations for demand for CPUs used in data centers. AMD’s market value has also surpassed $1 trillion, supported by its server processors, GPUs and rising demand for AI-related computing solutions.

Meta and OpenAI use virtual machines that divide a physical server into multiple virtual computers capable of serving many users simultaneously. Some AMD EPYC processors feature as many as 192 cores. Tests indicate that the virtual machine running Muse uses two CPU cores, while Dots relies on about nine.

CPUs also offer a major cost advantage over advanced GPUs. Market estimates indicate that some AMD EPYC processors available on the secondary market cost less than $3,000, while some Nvidia GPUs can cost more than ten times as much. The gap becomes increasingly important when AI agents operate continuously in the background.

Data Centers and a Potential $220 Billion Market

AMD’s data-center revenue more than doubled to $6.7 billion in the quarter ended in June, accounting for about 60% of total revenue. Muse has surpassed five million downloads, while estimates put the cost of running the agent at between $3 and $130 per user per month, depending on inference usage.

AI agents could account for about 20% of AMD’s chip sales in 2026. AMD expects the overall CPU market to reach about $220 billion by 2030, compared with an earlier estimate of $60 billion. Separate forecasts put CPU sales at $118 billion in 2027. AMD aims to capture more than half of the market by 2030, compared with its current roughly 46% share of the x86-compatible CPU market.

Nvidia is also moving into the CPU market with Vera processors and servers designed for AI agents, targeting a CPU opportunity it expects to reach about $200 billion by 2030. Arm is also expanding its presence, while major cloud companies are developing their own chips, often based on Arm architecture, to reduce data-center costs.

Amazon, Google, Meta and Microsoft use AMD and Intel processors across significant portions of their server infrastructure while developing proprietary chips. As AI agents become more widely used, the chip industry could shift from a race centered mainly on GPU performance toward a broader competition that puts CPUs back at the center of AI infrastructure.