Agentic AI Craze Drives Significant Increase in CPU Demand, But Morgan Stanley Pours Cold Water: Don’t Follow the Trend to Buy Intel, AMD.

Tradingkey - Since the beginning of this year, amid the surge of Agentic AI, market demand for CPU chips has been growing exponentially, shifting the previous narrative that only GPUs would significantly benefit from AI. Current mainstream Agentic AI in the market includes Anthropic's Claude Cowork and OpenClaw, among others.

What is Agentic AI?

Unlike traditional large language models that can only generate text responses based on user input, the core characteristic of Agentic AI is its ability to autonomously complete complex multi-step tasks.

It is reported that Agentic AI is an artificial intelligence system capable of autonomous decision-making, task planning, and complex task execution, enabling it to achieve preset goals without continuous human intervention.

Its workflow involves decomposing large tasks into smaller ones and then invoking various tools for execution; the entire process follows a top-down sequence and cannot process thousands of tasks simultaneously like a GPU matrix.

This indicates that the operational efficiency of Agentic AI depends almost entirely on the multi-threaded scheduling capabilities of the CPU, rather than the parallel computing power at which GPUs excel.

NVIDIA ( NVDA) CEO Jensen Huang made a very precise remark at GTC 2026 this March: "The CPU is no longer just supporting the model; it is driving the model."

How Agentic AI is Driving Increased CPU Demand

Entering the era of Agentic AI, the positioning of the CPU has transitioned from an auxiliary role in the computing power supply chain to the central control hub of agent systems.

In the agent era, a CPU's concurrency capabilities, memory bandwidth, and scheduling efficiency directly determine system response speed and processing capacity. Once agent concurrency exceeds the CPU's capacity limits, issues such as request queuing and tool invocation failures will arise, directly impacting user experience and system stability.

The latest research findings jointly released by the Georgia Institute of Technology and Intel (INTC) clearly quantify the central role of the CPU in agent operations.

Data indicates that across three mainstream agent tasks—RAG, web-connected search agents, and intensive scientific research—CPU time accounts for more than 80% of the total processing time. Energy consumption metrics further validate this conclusion; in medium-to-large batch processing scenarios, the CPU's share of energy consumption for these three tasks peaks at approximately 60%.

That is to say, during agent operations, more than half of the power is consumed by the CPU, while the market-favored GPU remains idle most of the time, waiting for the CPU to complete task scheduling.

This structural difference has established the CPU as the primary performance bottleneck in the Agentic AI era and has directly driven explosive growth in global CPU demand.

Arm ( ARM) CEO Rene Haas provided a set of data that intuitively illustrates the scale of this growth: traditional AI workloads require approximately 30 million CPU cores per gigawatt of data center capacity, but in the Agentic AI era, this demand will surge fourfold.

Agentic AI Spurs Demand for Independent CPU Server Computing Nodes

The growth in CPU demand driven by Agentic AI primarily stems from two factors: a significant increase in the number of CPUs within individual AI servers and the rapid adoption of new architectures featuring the disaggregated deployment of GPU and CPU clusters.

In traditional AI servers, the GPU-to-CPU ratio is approximately 8:1 to 8:2. With the introduction of next-generation AI acceleration architectures like GB200 and Vera-Rubin, this ratio has improved to 2:1. During its first-quarter 2026 earnings call, Intel explicitly stated that the future CPU-to-GPU ratio will further increase from the current 8:1 to 1:1, and could potentially tilt even further toward CPUs.

On the other hand, to address the explosive growth in AI computing demand, maximize resource utilization, enhance computational efficiency, and improve system flexibility, the industry is gradually gravitating toward a technical trend of disaggregated GPU and CPU cluster deployment.

Among these, Microsoft ( MSFT )'s next-generation Fairwater data center has pioneered the adoption of this disaggregated GPU and CPU cluster architecture. Under this setup, GPUs can focus on high-load model inference, while CPUs handle complex task scheduling and logic processing, significantly boosting overall system performance. This trend is being emulated by a growing number of cloud providers and will continue to drive the large-scale deployment of standalone CPU servers in the coming years.

According to Morgan Stanley's latest estimates, AI agents will create $32.5 billion to $60 billion in incremental growth for the CPU market by 2030, expanding the total market size for server-grade CPUs to between $82.5 billion and $110 billion.

Morgan Stanley Pours Cold Water! Don't Follow the Crowd to Buy Intel and AMD!

The surge in CPU demand driven by Agentic AI has exceeded market expectations. Coupled with factors such as limited upstream wafer foundry capacity and rising raw material prices, price hikes and shortages in the global CPU market have continued to intensify this year. This supply-demand imbalance is expected to persist for the next 1-2 years.

Currently, the x86 architecture remains the absolute mainstream for data center server CPUs, making AMD and Intel the primary beneficiaries of this CPU super cycle. UBS analysis indicates that as the transition from traditional AI training in 2025 moves toward agentic inference scenarios in 2026/2027, CPU workload requirements will increase to 3 to 8 times their original levels.

Despite this, Morgan Stanley holds a different view from the market consensus, which is generally bullish on CPU manufacturers. The firm does not recommend that investors directly position in the two CPU giants, Intel and AMD: although AMD's market share in the cloud CPU sector has surpassed Intel's to reach 53%, making it a direct beneficiary of the current CPU narrative, its stock price performance is more deeply tied to GPU business expectations; meanwhile, Intel's stock price is primarily driven by its foundry business transformation narrative, and neither is a pure-play CPU investment.

Morgan Stanley believes that the stocks in the current market that truly possess pure AI-enabled exposure are concentrated in leading memory and GPU companies, including NVIDIA, Broadcom ( AVGO ), Micron ( MU ), SanDisk ( SNDK ).

Morgan Stanley stated that the core logic for recommending these stocks is simple—attractive valuations: the forward P/E ratios for CPU manufacturers such as Intel and AMD currently range from 23 to 64 times, while NVIDIA's forecast P/E for fiscal year 2027 is only 18 times, and the forecast P/E ratios for memory stocks are as low as 5 to 9 times, offering significantly better valuation appeal.