AINEWS
Personal AI agents like Meta's Muse and OpenAI's Dots run for hours on their own, work that fits server CPUs better than GPUs, and the market is taking notice.
AMD is up 32% in a month. Intel is up 21%. Both have beaten every tech megacap over that stretch, and AMD has crossed a trillion-dollar market cap for the first time. The lift came from personal AI agents, the assistants that run for hours on their own, and the CPU-shaped work those products actually run on.
Meta's Muse debuted in early September and tells users it runs on an AMD-powered computer. OpenAI's Dots, released last week, tells users it runs on a virtual machine powered by AMD's EPYC-branded server CPU. Both are personal AI agents that open browsers, call tools, and wait on results before answering. That work sits on a long-running virtual machine, not a batched training cluster. It is CPU-shaped by design.
Ryan Shrout, president of Signal65, which consults on AI hardware and costs, said the workload fit is what matters. "As more agents are created and developed, and more people start to use them for more tasks, it's going to start to shift the workload away from GPUs and onto CPUs," Shrout said. It is a CPU-rehabilitation story the AI trade has not seen in a while.
The CPU never left. The biggest cloud companies, Amazon, Google, Meta, and Microsoft, run their hyperscaler servers on Intel and AMD CPUs as the default. Every large cloud provider is also building custom silicon, typically based on Arm's designs, on top of that base. Arm announced its own CPU for agents in March, with Meta as its debut customer, and the stock has more than doubled since. Nvidia, the company that defined the GPU era, released Vera, a fully redesigned CPU, alongside an entire rack filled only with those CPUs. Even the GPU king is building a CPU story for agents.
Muse caught on fast enough to top the Apple App Store in under two weeks. That kind of uptake tells the chip market there is a real consumer surface for long-running agents, and the demand curve for server CPUs follows.
The hedges still matter. A Meta spokesperson said the company "take[s] a diverse approach to our hardware and are largely CPU-agnostic by design, which gives us the most flexibility in acquiring capacity." OpenAI uses multiple CPU providers too. AMD and Intel are not walking away with the whole market. They are the visible winners of a rebalancing across a stack that is finally diversifying. That is the better story. A wider chip market running a wider mix of AI workloads gives the agent era more competition and more capacity than a single-vendor GPU stack ever offered.
The training market still runs on GPUs, and Nvidia still owns most of it. The persistent agent-runtime market is showing up in server-CPU revenue and in the tape. AMD and Intel are the cleanest public expression of that shift. Watch the next round of hyperscaler earnings calls, where the CPU line item is finally getting the airtime it used to lose to the GPU.