BofA: Agentic AI could expand the server CPU opportunity to $170B, lifting focus on AMD, ARM, Intel and Nvidia

A new take from Bank of America is arguing that the AI compute cycle is set to broaden beyond today’s chatbot-style workloads. In a note highlighted by Yahoo Finance, the bank estimates that agentic AI, meaning systems that can take actions and carry out multi-step tasks, could expand the server CPU market to as much as $170 billion over the coming year.

The framing matters because server CPUs sit at the center of most AI training and inference stacks. Even as GPUs and specialized accelerators have taken most of the headlines, CPUs are still used for core orchestration, data movement, and general compute around AI workloads. BofA’s view suggests that the amount of CPU capacity demanded by agentic systems could be larger than investors and planners previously modeled.

The market opportunity referenced in the report implicitly includes a wide range of chip suppliers. The highlighted group includes AMD, ARM, Intel and Nvidia, reflecting the way AI datacenters rely on both CPU architectures and the companies that supply server hardware and platforms built around them. ARM, for example, is an architecture licensor whose designs show up in many server CPUs, while Intel and AMD sell chips directly, and Nvidia sells GPUs that pair with CPU-driven server systems.

For AMD, the company is often associated with high-performance server processors that can be deployed across hyperscale and enterprise datacenters. For Intel and ARM, the competitive angle is typically tied to architecture performance, power efficiency, and the ecosystem of platforms that datacenter customers can buy. For Nvidia, the relevance is more indirect, since Nvidia’s AI business is GPU-focused, but server CPUs remain a necessary complement in AI infrastructure. The report’s emphasis on “agentic AI” therefore targets the broader compute stack rather than just one component.

What BofA appears to be changing is the expected intensity and duration of compute needs. Agentic AI is commonly described as AI that can plan steps and execute actions, which can translate into more frequent job scheduling, more diverse data access patterns, and potentially more compute across longer task chains. If that translates into additional CPU demand in server environments, the bank’s $170 billion estimate indicates that chips used alongside accelerators may see a sustained tailwind.

Still, the public item does not provide enough detail to verify how the $170 billion figure is derived, what time horizon it assumes beyond “one year,” or which portion of the total is CPU-only versus “server compute” more broadly. It also does not spell out assumptions about specific agentic AI deployments, customer capex timing, or how much of the incremental workload would be CPU-bound versus accelerator-bound.

Company disclosures were not cited in the available post, and no segment-level guidance changes were included. That means it remains unclear whether chipmakers have already booked demand indicates consistent with the bank’s forecast, or whether this is more of an expectations reset for where the next incremental compute budgets may land.

Investors will likely be watching for indicators that agentic workloads are translating into measurable datacenter CPU demand. Next steps to watch include how chip suppliers describe AI server orders, platform adoption with agent-like applications, and any commentary on whether agentic workloads shift performance and power requirements in ways that benefit particular architectures and product lines.