"CPUs Are Essential for AI Agents
More CPUs Needed Than GPUs"

Recently, as demand for artificial intelligence (AI) agents has increased, market attention is shifting from GPUs and memory semiconductors, which had previously been in the spotlight, to stocks related to CPUs.


Kim Il-hyuk, a researcher at KB Securities, stated, "While the benefits of AI market expansion had been focused on GPUs and memory semiconductors, CPU-related stocks began to attract attention as Intel's earnings far exceeded expectations."


CPUs Surpass GPUs in Popularity... On-Device AI PCs Accelerate [Weekend Money] View original image

The driving force behind this shift is the emergence of 'AI agents.' As the use of AI agents with enhanced inference capabilities grows, the ratio of GPUs to CPUs has been changing. Kim explained, "Until now, for AI tasks, training required about one CPU per seven to eight GPUs, and inference required one CPU per three to four GPUs. However, due to the rapid surge in demand for AI agents, the GPU-to-CPU ratio is converging towards 1:1, or in some cases, even more CPUs are needed."


This is because, in the AI agent environment, CPUs are required not only for large language models (LLMs) but also for tasks such as querying local vector databases, calling external APIs, and controlling sequential flows between applications. While GPUs excel at parallel computation, CPUs are essential for sequential and logical control.


In particular, current CPU supply elasticity is extremely low. Both ARM-based and x86-based CPUs are seeing increased demand, while TSMC's manufacturing lines, which produce these chips, are already largely occupied by GPU production. Kim noted, "Intel, which can mass-produce x86-based CPUs, announced in its earnings report that its own production lines have no extra capacity and that there is tens of billions of dollars in unmet demand. After this announcement, the market became aware of the low supply elasticity of CPUs."



It is expected that if demand for on-device AI PCs increases, CPU supply will become even tighter. On-device AI PCs perform AI computations directly on the device itself, without relying on cloud servers. Kim said, "As users become more dependent on AI features provided through the cloud, there is a growing fear that they are helpless in the face of AI service instability or price increases. In this context, more users are seeking to run open-source AI models in local environments."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing