As personal AI agents spread, demand for server CPUs is rising and AMD and Intel are drawing attention again. On Oct. 6, CNBC reported that agents such as Meta's "Muse" and OpenAI's "Dots" are performing autonomous work for long periods, increasing the role of CPUs again in AI infrastructure that had been centred on GPUs.

In the early days of generative AI, Nvidia GPUs led the market. Recently, however, agents are increasingly being planned and then run in the background for hours to days. Ryan Shrout, president of Signal65, said that as more agents are developed and used, workloads will begin moving from GPUs to CPUs.

Meta unveiled "Muse" in early September, and the service reached No. 1 in Apple's App Store less than two weeks after launch. OpenAI also released its agent "Dots" last week. User queries and benchmarks showed that "Muse" ran on AMD-based computers, and "Dots" ran on virtual computers equipped with AMD EPYC CPUs.

Meta said it designed its systems so they are not tied to a specific CPU vendor. OpenAI also uses multiple CPU suppliers. Many hyperscaler servers at Amazon, Google, Meta and Microsoft use Intel or AMD CPUs, but major cloud companies are also developing their own Arm-based chips.

AMD has joined the ranks of $1 trillion companies as its CPU and GPU businesses grew together. In the quarter ended in June, data centre revenue was $6.7 billion, more than doubling from a year earlier, and accounted for nearly 60 percent of total revenue. Dan McNamara, AMD senior vice president, said discussions around the use of agents have changed sharply over the past 8 to 10 months and that CPU sales will also begin to rise in earnest.

The agent architecture also supports CPU demand. Meta and OpenAI use virtual machines that split a single server into multiple virtual computers. Daniel Newman, chief executive of the Futurum Group, explained that GPUs handle AI model inference while CPUs run the actual workflow. The AMD EPYC 9V74 CPU used for "Dots" can be found in the resale market for less than $3,000, but Nvidia GPUs often cost more than 10 times that for a single chip.

Market forecasts are also rising. The Futurum Group projected total CPU revenue would reach $118 billion in 2027. AMD estimated the CPU market would reach $220 billion by 2030. AMD holds about 46 percent share by x86 CPU shipments. Intel is also benefiting from rising demand, but AMD is cited as having strengths in hyperscaler market positioning and the x86 architecture, which has high compatibility with existing software.