Key Takeaways:
A new Bank of America forecast projects the server CPU market will nearly triple to $125 billion by 2030, triggering a seismic shift in market share from x86 incumbents to ARM-based custom silicon.
The server CPU market is set for a 31% compound annual growth rate as agentic AI workloads move beyond GPU-centric training, with ARM-based designs projected to capture roughly 44% of the market by 2030, according to a new Bank of America report. This rapid expansion, from an estimated $43 billion in 2026, reflects a fundamental change in the role of CPUs in the data center. While GPUs dominated the AI training era, the next phase of agentic AI—where systems manage complex, multi-step tasks like planning, tool calling, and database access—is structurally more reliant on CPU-driven orchestration.
"CPUs become 'the control plane of AI inference,' handling orchestration and memory management across increasingly complex multi-step workflows," Bank of America analysts led by Vivek Arya said in a note that sent ripples through the semiconductor sector.
The forecast sees the total server CPU market growing to $125 billion by 2030. Within this, Intel's market share is expected to collapse from 54% in 2025 to 28%, while rival AMD holds steady at 28%. The most significant gains are projected for ARM-based custom chips, like those designed by AWS, Google, and Microsoft, which are forecast to grow from about 15% of the market today to a commanding 37% share.
This structural shift directly threatens Intel’s historical data center dominance and benefits ARM, whose royalty-based model is poised for significant operating leverage as its architecture scales in the high-value data center market. For investors, the report frames a clear divergence, recommending AMD and Nvidia while highlighting the severe long-term risks for Intel as its largest customers increasingly design their own chips.
For years, the primary role of a server CPU in an AI system was to feed data to the powerful GPU accelerators doing the heavy lifting of model training. As Bank of America notes, this led to AI accelerators accounting for 88% of data center compute spending by 2025, with CPUs relegated to just 12%.
However, the rise of agentic AI flips this script. Instead of a single, massive computation, these systems execute a chain of smaller tasks: retrieving data through retrieval-augmented generation (RAG), querying vector databases, calling external tools, and managing state. While a GPU may be called for specific model inference steps, the CPU is responsible for orchestrating this entire workflow. This increased demand is not for single host CPUs inside a GPU rack, but for entire CPU-only racks dedicated to orchestration, data processing, and running smaller models, a segment where ARM's efficiency and customizability provide a key advantage.
Bank of America's 2030 forecast paints a stark picture of the new competitive landscape:
Intel (INTC): The biggest loser, with its server CPU value share projected to be cut in half to 28%. The company faces a two-front war, losing cloud share to custom ARM chips and enterprise x86 share to a resurgent AMD.
AMD (AMD): A winner within the x86 camp, AMD is expected to continue taking share from Intel, peaking at around 38% in 2026 before settling at 28% by 2030 as ARM's advance accelerates. BofA maintains its preference for AMD as the strongest x86 player.
ARM Holdings (ARM): The primary long-term beneficiary. The report projects custom ARM CPUs (AWS Graviton, Google Axion, Microsoft Cobalt) will be the fastest-growing segment. This aligns with a recent Bernstein note initiating coverage with a $300 price target, citing a $2 billion-plus backlog for custom silicon and the superior economics of data center royalty rates over mobile.
Nvidia (NVDA): Nvidia is positioning itself above the fray. With its upcoming 'Vera' CPU designed to be integrated into a full-stack system with its GPUs and networking, the company is not just selling a chip but an entire architecture, insulating it from the direct x86 vs. ARM battle.
This outlook from Bank of America suggests a fundamental re-evaluation of the semiconductor landscape. While Intel faces a difficult defensive battle, AMD, ARM, and Nvidia are positioned to capture the value from a much larger and more complex AI-driven data center market.
This article is for informational purposes only and does not constitute investment advice.
