--- title: "UBS Says AMD and Arm Stand to Gain as 'Agentic AI' Drives a Server-CPU Revival" description: "UBS argues that the rise of autonomous AI 'agents' is reviving demand for general-purpose server processors — not just GPUs — positioning AMD's EPYC chips and Arm's CPU designs as beneficiaries. The bank lifted its price targets on both. The thesis is one desk's analysis, not a forecast or a recommendation." category: "Tech" category_url: https://boursel.com/category/tech author: "Sofia Marchetti" published: 2026-06-24T15:32:00.000Z updated: 2026-06-24T15:32:00.000Z canonical: https://boursel.com/article/ubs-amd-arm-agentic-ai-cpu tags: ["ubs", "amd", "arm", "agentic-ai", "server-cpu"] --- # UBS Says AMD and Arm Stand to Gain as 'Agentic AI' Drives a Server-CPU Revival UBS argues that the rise of autonomous AI 'agents' is reviving demand for general-purpose server processors — not just GPUs — positioning AMD's EPYC chips and Arm's CPU designs as beneficiaries. The bank lifted its price targets on both. The thesis is one desk's analysis, not a forecast or a recommendation. Swiss bank UBS told clients this week that the next wave of artificial intelligence could lift an often-overlooked corner of the chip market: the general-purpose central processing units, or CPUs, that run alongside AI accelerators in data centers. The view, [attributed to UBS analyst Timothy Arcuri](https://www.cnbc.com/2026/06/24/these-chipmakers-are-climbing-thanks-to-sky-high-cpu-demand-ubs-sees-more-gains.html), is that the spread of "agentic AI" should benefit Advanced Micro Devices (AMD) and Arm Holdings. It is a scenario laid out by one research desk, not an established fact. ## What the thesis says For most of the AI boom, attention has centered on the GPU — the graphics processing unit, a chip built to run thousands of calculations in parallel, ideal for training and running large models. The CPU is the more general-purpose processor that manages logic, coordination and the flow of work. UBS's argument is that agentic AI shifts more of the workload back toward the CPU. Agentic AI refers to systems that do not just answer a single prompt but act as autonomous "agents," carrying out multi-step tasks: calling external tools, querying databases, running code, retrying failed steps and coordinating sub-agents. Much of that orchestration runs on CPUs rather than GPUs. As UBS frames it, when an agent pauses to call a tool or wait on a result, the GPU can sit idle while the CPU does the coordinating work. ## The numbers UBS put on it UBS estimates the total addressable market for server CPUs could grow [roughly fivefold, from about $30 billion in 2025 to around $170 billion by 2030](https://finance.yahoo.com/sectors/technology/articles/ubs-agentic-ai-shift-signals-195528399.html). It also cites expert estimates that agentic systems may require far more CPU cores — on the order of 80 to 120 cores per GPU, against 8 to 12 in older training setups. Those are forward projections that depend on how quickly agentic AI is deployed at scale. On the back of that thesis, UBS [raised its 12-month price target on AMD to $670](https://www.investing.com/news/analyst-ratings/ubs-raises-amd-stock-price-target-to-670-on-server-cpu-share-gains-93CH-4757934) and [on Arm to $470](https://www.investing.com/news/analyst-ratings/ubs-raises-arm-holdings-stock-price-target-to-470-on-ai-chip-outlook-93CH-4757942), keeping a Buy rating on both — sharp increases from prior levels. Price targets are analyst estimates of where a stock might trade, not guarantees. ## Why AMD and Arm AMD sells server CPUs under the EPYC brand — high-core-count x86 chips for data centers. UBS said it is "more constructive on AMD as standalone CPU racks gain traction," citing the company's advantage in core count, multithreading and the established x86 software ecosystem. Arm's case is different. Arm does not generally manufacture chips; it designs CPU architectures and licenses them to others, collecting upfront fees and per-chip royalties. UBS expects much of the growth in AI-rack "head nodes" — the CPUs that route work to accelerators — to be Arm-based, alongside standalone CPU-only servers split between x86 and Arm designs. ## The broader picture — and the caveats The note fits a wider rebalancing in AI hardware: GPUs and custom accelerators for training and heavy inference, CPUs for serving, orchestration and agent coordination. It echoes themes in Boursel's coverage of OpenAI's custom-chip efforts, Qualcomm's data-center ambitions and SK Hynix's memory supply. UBS is not alone — Bernstein and Bank of America have published similar agentic-AI CPU theses, a sign the idea is gaining traction on Wall Street. Still, readers should treat all of this as a scenario from one bank's analysts. The market-size and core-count figures are estimates that hinge on actual adoption, and price targets are not promises. This article is reporting on analysis, not offering investment advice.