The artificial intelligence (AI) computing market is undergoing a massive transformation, signaling a significant shift in demand from Graphics Processing Units (GPUs) to Central Processing Units (CPUs). Industry forecasts suggest that CPUs are poised to capture a much larger share of the total AI infrastructure budget, potentially reaching a market valuation of $120 billion. This growth is primarily driven by the emergence of “agentic AI,” which requires sophisticated, complex coordination that CPUs are uniquely suited to handle.
The Shift from Chatbots to Agentic AI Workloads
Historically, AI demand was dominated by non-agentic workloads, such as simple chatbot queries. These tasks are comparatively straightforward, requiring a direct input and output, often paced by human interaction. However, agentic AI represents a much more advanced capability. These AI agents are designed to operate autonomously, handling hundreds of concurrent tasks and independently reasoning through complex problems to reach a conclusion, often with minimal human direction.
This structural difference mandates a shift in the necessary hardware balance. According to research published by Intel and Georgia Tech, sophisticated, tool-driven agentic AI workloads create significant “bottlenecks” where CPUs consume up to 88% of the total end-to-end latency. The researchers further concluded that while improved GPUs are beneficial, the system bottleneck is likely to shift more heavily towards the CPUs to maintain efficiency. To scale agentic AI effectively, the CPU orchestration capacity must match the GPU reasoning capacity, thus requiring an increase in the CPU-to-GPU ratio within AI computing clusters.
Major Industry Forecasts and Market Growth
The magnitude of this shift is reflected in aggressive market forecasts issued by leading semiconductor companies. AMD recently revised its server CPU market outlook, nearly doubling its projected Compound Annual Growth Rate (CAGR) to 35%, and estimates the market will exceed $120 billion by 2030. Similarly, Arm announced in March that the total addressable market (TAM) for data center CPUs is expected to surpass $100 billion by its fiscal year 2031 (approximately calendar year 2030). This represents a projected increase of over four times from its previous TAM estimate of $24 billion, equating to a 33% CAGR.
Several financial institutions corroborate this upward trend. UBS predicts the market will expand from $31 billion in 2025 to $170 billion in 2030, citing AI CPUs as the primary driver. Bank of America forecasts a TAM expansion from $43 billion in 2026 to $125 billion in 2030, projecting a 30.6% CAGR. Citi estimates the overall market will grow from $29.3 billion in 2025 to $132 billion in 2030, with agentic CPU growth alone projected to hit $59.4 billion in 2030, representing a 185% CAGR.
The acceleration in growth rates is unusually high for the CPU market, which has historically seen single-digit annual growth. This rapid escalation suggests that the supply chain was not adequately prepared for the current pace of demand.
The Critical Role of CPU Orchestration
The primary function driving the CPU market boom is “orchestration”—the complex process of directing API calls, coordinating multiple tasks, and calling various tools among dozens of independent AI agents. While GPUs handle the core inference reasoning, CPUs are tasked with directing where, when, and how those resources are allocated. This orchestration role is essential for managing the diverse responsibilities of AI agents.
Furthermore, industry analysts are tracking the shift in metrics. MidTrendForce noted that the current CPU-to-GPU ratio in AI data centers falls between 1:4 and 1:8. For agentic AI applications, however, the ratio is projected to move significantly to between 1:1 and 1:2, leading to a substantial increase in demand for CPUs.
Arm CEO Rene Haas highlighted this potential growth, stating that agentic AI could boost CPU core demand by as much as four times, reaching 120 million cores per GW, compared to approximately 30 million cores per GW today. This increase is not just about core count but also represents a margin expansion opportunity for CPU designers through greater core density.
Supply Constraints and Vendor Strategies
The escalating demand has already created visible supply constraints. Reuters reported in February that Intel had a considerable backlog of unfulfilled CPU orders, with delivery times extending up to six months. Similarly, some AMD products were noted with delivery windows of eight to ten weeks. KeyBanc issued upgrades in January, observing that both Intel and AMD were nearly sold out of CPU servers for 2026, accompanied by anticipated Average Selling Price (ASP) increases of 10% to 15%. These supply issues suggest that CPU vendors are currently holding significant pricing power.
Vendors are responding by developing new architectures. Nvidia is aggressively entering the standalone CPU rack market with its Vera rack. This new offering, which includes 256 CPUs, allows customers to deploy a significantly higher number of CPUs (22,528 cores) compared to traditional configurations, marking a major architectural shift. Nvidia CEO Jensen Huang has framed CPUs as additive to GPUs, arguing that increased AI agent activity requires both greater orchestration (more CPUs) and more inference (more GPUs).
Arm is also capitalizing on the trend with its AGI CPU, co-developed with Meta. Arm touts the AGI CPU’s superior performance per watt, claiming it can deliver up to twice the performance per watt compared to x86 CPUs. This efficiency advantage is crucial for hyperscalers seeking to maximize compute power while managing power consumption.
Meanwhile, Intel is positioning itself to maintain its market leadership by announcing plans to deploy rack-scale CPU systems at Computex. The company’s new designs aim for high core density and are designed for both maximum density and latency-sensitive agentic AI workloads.
The competition is intensifying, with AMD, Nvidia, Arm, and Intel all targeting the same rapidly expanding market. While Nvidia projects a $200 billion CPU TAM based on the Vera rack, AMD’s $120 billion estimate shows the scale of the opportunity. The market is clearly moving toward a scenario where CPUs are not merely supporting role, but are the next major bottleneck in AI infrastructure, leading to an aggressive race among the major chip designers.
