Server CPUs regain data center dominance in AI

A research report released on June 17 by analyst Dai and his team forecasts the global server CPU market will reach 223 billion U.S. dollars by 2030, up sharply from an expected 37 billion U.S. dollars in 2025. The projection reflects a major shift in artificial intelligence architecture, where CPUs regain importance alongside GPUs as AI systems evolve beyond chatbot-style models into more autonomous “agentic” systems.

The report argues that CPUs will take on a larger role in managing complex workflows and logic control, reshaping how data centers are designed and how computing resources are distributed.

Ai architecture shift drives CPU demand

As AI systems become more autonomous, the balance between GPUs and CPUs is expected to move toward parity. The ratio of GPUs to CPUs in cloud inference clusters is projected to narrow from 8 to 1 in 2025 to 1 to 1 by 2029.

This change significantly alters computing workloads. CPUs are expected to handle about 50 percent of processing in these environments, up from roughly 14 percent in traditional large language model tasks. The shift reflects increasing demand for coordination, reasoning, and task execution rather than pure parallel computation.

Hardware makers adjust strategies

Major semiconductor companies are already adapting their designs to reflect this transition. AMD, Nvidia, and Google are building systems that pair fewer GPUs per CPU compared with earlier generations.

Nvidia is integrating its Grace CPU with Blackwell GPUs, while AMD continues to expand its EPYC server CPU lineup alongside its accelerators. Intel is also positioning its Xeon processors to capture rising demand tied to AI infrastructure.

The report estimates that by 2030, AI data center capacity will reach 70 gigawatts, supported by 3.5 trillion U.S. dollars in cumulative AI capital spending and a 1.6 trillion U.S. dollar accelerator market. Server CPUs alone are projected to grow at a compound annual rate of 43 percent, with 174 billion U.S. dollars tied specifically to agentic AI workloads.

Arm seen as key beneficiary

ARM was identified as a major structural winner due to its energy efficiency advantage. Systems based on ARM architecture, such as AWS Graviton, can deliver about 40 percent lower cost and 60 percent lower power consumption compared with x86 alternatives.

The company plans to begin manufacturing its own CPUs by 2026, targeting 15 billion U.S. dollars in chip revenue by 2030. Meta has been named as its first confirmed customer. The report raised ARM’s projected fiscal 2030 earnings per share to 11.79 U.S. dollars and increased its price target to 500 U.S. dollars.

SoftBank, which holds roughly 90 percent of ARM, also saw its valuation target lifted to 11,200 yen from 8,200 yen, reflecting stronger asset value and improved portfolio performance.

Rising targets across chipmakers

Other semiconductor companies are also expected to benefit from increased CPU demand:

• AMD target price raised to 600 U.S. dollars on expectations of continued x86 leadership
• Intel target increased from 65 to 100 U.S. dollars as server CPU demand strengthens
• China-based Hygon target lifted from 280 to 450 yuan, with domestic market share projected above 35 percent by 2030

Supply constraints and risks remain

Despite strong growth projections, the report highlights significant supply-side risks. Foundry capacity at advanced nodes such as 3 nanometers and 5 nanometers may not be sufficient to meet demand, while memory supply could also limit production.

The model implies the need for roughly 30 billion U.S. dollars in additional annual CPU output, raising concerns about whether manufacturing can scale quickly enough.

Energy availability is another constraint. Data center electricity consumption is expected to rise sharply, with global usage projected to more than double by 2030. Power supply limitations could slow the expansion of AI infrastructure.

Competitive and structural uncertainties

The report also points to potential competitive pressure from Nvidia’s development of its own ARM-based Vera CPU, which could both partner with and compete against ARM. This creates uncertainty around ARM’s projected 54 percent market share.

Overall, the analysis concludes that if agentic AI systems are widely adopted, the industry will move away from GPU-dominated architectures toward a more balanced CPU-GPU model. This shift is expected to redirect capital flows and reshape the semiconductor sector, provided supply chains and investment levels can keep pace with demand.