[News] ARM CEO Says Agentic AI May Drive CPU Core Counts to 512 as GPU-CPU Ratios Become Less Relevant

Haas noted that overall CPU demand is likely to increase significantly as agentic AI scales, with data centers potentially requiring more than four times today’s CPU capacity. He said this could create a data center CPU market opportunity exceeding US$100 billion by 2030.

At the same time, Haas emphasized that the industry is seeing not only an explosion in overall CPU demand, but also rapid growth in the number of cores per CPU. According to Haas, many agentic AI workloads involve independent jobs, flows, or batches running on dedicated CPU cores, increasing the need for higher-core-count processors.

Haas used Blackwell, Rubin, and other large AI accelerators as examples, noting that these chips are already approaching reticle limits, meaning their size is constrained by the maximum area a lithography mask can print. In contrast, he said CPU core counts could still double or even quadruple over the coming years.

Haas noted that the Arm AGI CPU already features 136 cores, significantly higher than many competing offerings. Looking ahead, he said the industry is likely to move toward 256-core and even 512-core CPU designs. He added that such high-core-count architectures play to Arm’s strengths, as efficiency per core becomes increasingly critical at larger scales.

Mydrivers notes that AMD and Intel are moving in a similar direction. AMD’s 2nm Zen 6-based EPYC processors are already expected to reach up to 256 cores with SMT multithreading support, while Intel’s all-E-core Xeon processors have reached 288 cores, with the next generation expected to scale to as many as 512 cores.

Regarding the Arm AGI CPU launched at the company’s Arm Everywhere event last quarter, Haas said customer response has been “very strong.” He added that customer demand across fiscal 2027 and fiscal 2028 has already exceeded US$2 billion, more than double the level projected at launch.