For the past two years, the semiconductor narrative has been almost entirely about GPUs. Nvidia became the world’s most valuable company on the back of that story. But a quieter, arguably more consequential battle is now playing out in the CPU market, and it involves three companies fighting over who gets to power the next wave of AI: autonomous agents.
AMD, Arm Holdings, and Intel are each placing very different bets on agentic AI, the technology that lets AI systems act independently rather than just respond to prompts.
Why CPUs are suddenly cool again
Unlike the large language model training workloads that made GPUs indispensable, autonomous AI agents need general-purpose compute cores that can orchestrate multiple tasks simultaneously, manage memory, and handle the coordination layer between agents.
The numbers paint a striking picture. Agentic AI workloads are projected to increase CPU core requirements by as much as 4x per gigawatt of energy consumed. Multi-agent systems are expected to drive a 15-fold increase in token generation compared to single-model inference.
