Forecast: Server CPU market to grow 5 times in 5 years! UBS: ARM will benefit most, followed by AMD, with Intel last.

In the AI era, CPUs, overshadowed by GPUs, may be quietly ushering in their own explosive growth.

According to Sui Feng Trading Desk, on May 5, the UBS Global Research team released an in-depth research report on the U.S. semiconductor industry. Facing numerous investor questions about "how agent-based AI will impact the server CPU market," analysts Timothy Arcuri and others, through a series of expert interviews and a combination of bottom-up and top-down models, arrived at a clear conclusion:

The market has seriously underestimated the value of CPUs in the AI era. The total addressable market (TAM) for server CPUs will grow from about $30 billion in 2025 to about $170 billion in 2030, an increase of nearly fivefold in five years.

In the past two years of AI frenzy, GPUs have taken all the spotlight. But as AI evolves from "simple conversational generation" to "autonomous task-executing agents," the bottleneck of computing power is quietly shifting.

Agent-based AI reshapes the computing landscape: from "GPU dominance" to "CPU resurgence"

To understand the explosion of the CPU market, one must first understand the workload differences between agent-based AI and traditional AI.

In traditional AI training and fundamental inference phases, GPUs are the absolute main force. If AI computing power is likened to a factory, the GPU is the tireless worker on the assembly line, while the CPU is the manager responsible for task allocation. In the traditional model, one manager (CPU) can easily oversee several workers (GPUs).

But agent-based AI changes the rules. Agent-based AI not only needs to generate text but also orchestrate tasks, make tool calls (such as executing code in a sandbox virtual machine), file retrieval, etc. This means the manager's workload increases exponentially.

Analysts obtained striking data from expert interviews:

Shift in workload focus: Experts point out, "In traditional AI workloads, 70-80% of computing power is consumed on inference itself (GPU); but in agent-based inference, this ratio flips, and 70-80% of workload shifts to the CPU."
Explosion in core allocation: In traditional AI training, each GPU typically only needs 8-12 CPU cores; in basic inference, 16-24 cores; but in agent-based AI, each GPU needs 80-120 CPU cores. This means, for the same GPU, the number of CPU cores needed in agent scenarios is 5-10 times that of traditional training scenarios.
Pressure of concurrent tasks: "An agent (and each sub-agent it generates) may require 1-4 CPU cores, and a complex task might require 10-100 sub-agents to be generated."

This shift in underlying logic directly breaks the past "GPU-heavy, CPU-light" computing architecture, opening up enormous incremental space for the CPU market.

A massive $170 billion market

Based on this logic, the analysts recalculated the total potential market (TAM) for server CPUs. The results show that by 2030, the market size will reach about $170 billion.

How was this huge figure calculated? Analysts conducted cross-validation using both bottom-up and top-down methods:

1. Bottom-up calculation: Analysts, based on U.S. hyperscale cloud service providers' accelerator models, predict that by 2027, the market will ship around 23 million accelerators (XPU) and about 10 million head node CPUs. With the development of agent-based AI, by 2030, accelerator shipments will reach about 40 million. More importantly, the CPU-to-GPU ratio will move from the current 1:4 to 1:2 or even 1:1. In addition, as AI applications require higher core counts and higher frequency chips, the average selling price (ASP) of AI CPUs will rise significantly. For instance, Nvidia's 144-core Grace CPU may be priced between $3,000 to $4,000. With both volume and price rising, the AI CPU market alone will reach $125 billion.

2. Top-down calculation: Analysts refer to Nvidia's prediction of a total AI potential market ($3 to $4 trillion) by 2030. They infer about 40 million XPUs shipped in 2030. Assuming each XPU's ASP rises to $3,000, combined with a 1:1 or 2:1 CPU allocation ratio, it similarly deduces the AI CPU market size between $120 to $200 billion.

Analysts divide the future CPU market into three core segments:

Traditional server market: Maintains steady growth, expected shipments of about 44 million units by 2030.
AI head nodes: Bundled with GPU racks, mainly responsible for task orchestration and optimizing GPU utilization.
Standalone AI racks: Pure CPU servers dedicated to handling agent-based AI's tool calls and sub-agent concurrent tasks.

With the market pie growing, the key is how it’s divided.

Analysts clearly state the ranking: ARM is the biggest beneficiary on the server CPU side, followed by AMD, then Intel—all three stand to benefit.

In 2025, ARM architecture's unit share in the server CPU market is about 15%. The report predicts that by 2030, this number will rise to 40-45%; in terms of revenue share, as the ASP for AI CPUs is higher, ARM's revenue share will further rise to 50-55%.

Where does ARM’s advantage lie?

The firm cites experts: ARM architecture has about 30% higher power efficiency, 20-30% higher memory efficiency, and with smaller core designs, clear advantages in latency and cost. More crucially, Nvidia Grace, AWS Graviton 5 (192 cores), and other major hyperscale cloud giants’ self-developed CPUs almost exclusively use ARM architecture.

The firm expects that by 2030, ARM will occupy more than 75% of the head node CPU market.

But ARM also has drawbacks. The report notes ARM was traditionally a single-threaded architecture, and SMT (simultaneous multithreading) capabilities have only been gradually developed recently; in high core count scenarios, core interference and software compatibility issues remain challenges; additionally, ecosystem maturity still needs improvement, with some software stack refinements possibly not reaching completion until around 2028.

Based on the above, analysts raised ARM’s 12-month target price from $175 to $245. As of the day before the report release (May 4), ARM's stock price was $203.26, with a “Buy” rating maintained.

AMD: High core count + multithreading, the best partner for AI

AMD's advantage lies in high core count and multithreading capability, which highly fits the agent AI demand for CPUs needing “more and faster.”

The report cites AMD’s analyst day statement from November 2025: AMD expects the server CPU market to grow from $26 billion in 2025 to about $60 billion in 2030, with AI-driven CPUs accounting for about 50% of the 2030 market; AMD expects its own share of the total market to exceed 50%.

Analysts currently forecast AMD’s 2030 EPS at $25.27; if the market evolves as expected, the revised 2030 EPS could reach $28.14, about +11% upside.

Intel: Solid foundation, but under catching-up pressure

Intel's situation is relatively complicated.

In the traditional server market, x86 architecture will still maintain about 85% share, and Intel retains advantages in tool calls, storage optimization, and other specific workloads. But in the AI head node market, Intel's presence is quickly being compressed by ARM.

UBS points out that Intel is pinning its hopes on the “Coral Rapids” product line to close the gap with AMD and ARM, but currently AMD and ARM are better positioned in the AI CPU market.

However, Intel also has a unique card: spillover effect from the PC side. As agent AI pushes more tasks to run on local devices (Anthropic’s Claude Code is already adopting this strategy), the PC upgrade cycle is expected to be catalyzed, from which Intel will benefit.

The firm’s upward revision for Intel’s 2030 EPS is about +7%, the lowest among the three.

Not all CPUs are the same: Tradeoff between latency and throughput

The report also deeply analyzes a detail that’s easily overlooked: Agent AI’s demand for CPUs isn’t simply “the more cores the better.”

Hyperscale cloud providers face a fundamental trade-off in hardware selection:

High core count CPUs: Higher total throughput, better energy efficiency, but low clock frequency and poorer latency, plus limited software scalability (most software cannot efficiently utilize hundreds of cores)
Low core count, high frequency CPUs: Low latency, fast response, suitable for “head node” roles (responsible for orchestration, optimizing GPU utilization)

In real-world deployments, hyperscalers tend to use a layered architecture of “head node + large-scale compute node”: the former responsible for low-latency orchestration/control, the latter for high-throughput parallel execution.

This means vendors who can offer a wide range of SKUs (covering different core counts, frequencies, and power levels) are more competitive than those who bet on a single 'strongest' configuration.

UBS also notes the core metric in hyperscaler procurement decisions is not peak performance, but transactions per watt, with memory configuration as the primary design variable.

Cloud vs edge? An unresolved variable

Analysts also highlight an uncertainty worth noting: the division of computation between cloud and edge.

Early agent deployments relied almost entirely on the cloud, but more system designs now push computation to local devices—5 to 10 parallel tasks can run directly on local files and data, reducing latency and saving cloud computing power costs.

UBS cites expert estimates: local execution’s expansion may cut the CPU capacity needed for cloud agent workloads by about 25%.

This means agent AI’s multiplier effect on datacenter CPU demand may ultimately compress from 5-8x to about 4x. Meanwhile, PC-side CPU demand will rise in tandem, benefiting both AMD and Intel.

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content is from Sui Feng Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join [Sui Feng Trading Desk ▪ Annual Membership]

Risk Warning and DisclaimerThe market involves risks; investment should be cautious. This article does not constitute personal investment advice nor does it take into account individual users’ special investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their specific circumstances. Investment based on this article is at your own risk.