- In its latest semiconductor industry research report, Bank of America has significantly raised its forecast for the global server CPU market size in 2030 from the previous $125 billion to $170 billion, nearly five times the estimated $35.2 billion in 2025, with an expected compound annual growth rate of 37%.
- The report points out that artificial intelligence is evolving from mere content generation to the stage of autonomous task execution agents. The complex decision-making, sequential control, and task orchestration of Agentic AI significantly increase the dependency on CPUs with ultra-low latency and strong single-thread sequential execution capabilities.
- Bank of America expects that by 2030, the CPU to GPU ratio in AI data centers will narrow from the previous imbalance of 1:2 or 1:4 to above 1:1, with global computing infrastructure fully entering a new norm of dual-engine parallel drive by CPUs and GPUs.
Agent Transformation Spurs New Computing Power Increment
In the global wave of artificial intelligence over the past three years, the market focus and computing power capital expenditure have been almost entirely dominated by graphics processors, leading to the perception that central processors are somewhat marginalized. However, as generative AI evolves into autonomous agents, the underlying logic of computing infrastructure is undergoing a structural transformation. The latest industry report from Bank of America shows that this transformation does not mean that central processors will return to the market and replace graphics processors, but rather that with the comprehensive expansion of AI infrastructure demand, a new incremental space, previously underestimated by the market, is about to be unleashed. If agent applications accelerate their deployment in the commercialization process, the demand scale of the central processor market may exhibit stronger explosive power.
Subdivided Nodes Establish Three Major Functions of Central Processors
In the architecture of new AI data centers, the functions of central processors are undergoing refined reconstruction. The Bank of America report points out that future central processors will mainly undertake three core responsibilities. First, maintaining the basic business of traditional cloud computing and enterprise-level daily servers; second, acting as the head node in AI clusters, responsible for managing and coordinating the collaborative work among thousands of graphics processors; third, being entirely dedicated to the specialized nodes derived from the new era of agents. These specialized nodes will focus on managing the inference loops, state memory retention, tool software invocation, and complex task orchestration of agents, becoming the brain center of the entire system operation.
Specialized Output Value Approaches Traditional AI Clusters
According to Bank of America's quantitative forecast, by 2030, the market output value of central processors dedicated solely to autonomous agents will reach $70 billion. This figure is almost on par with the market size of central processors in traditional AI clusters. This means that future data centers, while deploying large racks of high-power graphics processors, must also configure large-scale central processor clusters tailored for running agents. This adjustment in hardware-level configuration directly reflects the marginal shift in the focus of resource allocation within the computing power network, with a new type of purely incremental market rapidly taking shape.
Architectural Ratio Reconstruction Drives Industry Chain Valuation Recovery
In response to market concerns about whether the growth of central processors will erode the share of graphics processors, Bank of America provides a clear negative conclusion. The report emphasizes that AI infrastructure is upgrading from a competition of single component specifications to a confrontation of overall system-level efficiency. As the hardware guarantee for core training and intensive inference, the demand for graphics processors will expand in sync with the central processors responsible for command and scheduling. By 2030, the configuration ratio of the two within data centers is expected to reverse from an imbalanced state to above one-to-one. If this turning trend in computing power architecture is established, the market valuation and potential growth space of central processor industry chain-related companies, which were previously under valuation pressure in the computing power wave, may face profound repricing and value recovery.
