AI intelligent entities are driving the birth of a "super cycle"! Citigroup: The server CPU market will expand to over $130 billion by 2030.

Citi estimates that the server CPU industry will see significant expansion in the coming years, with the market size projected to grow from $29.3 billion in 2025 to approximately $132 billion in 2030. The agency predicts that the main driver of this growth will come from the emerging category of "Agentic CPU", which is primarily used to support increasingly autonomous artificial intelligence workloads. Citi expects the compound annual growth rate (CAGR) of the general server CPU market to reach 20%, with a market size of approximately $50.9 billion by 2030. At the same time, AI head node processors are expected to grow at a CAGR of 21% during the same period, reaching a market size of about $21.1 billion. The Agentic CPU is expected to be the fastest-growing area, with Citi forecasting an annual growth rate of 185% and a market size of around $59.4 billion by 2030. With the launch of Anthropic's Claude Cowork and similar super AI agent tools like OpenClaw in 2026, the wave of AI intelligences is rapidly sweeping across the globe. The bottleneck of AI computing architecture is shifting from GPUs centered around matrix multiplication and addition throughput to data center CPUs focused on control flow, task orchestration, and memory/IO coordination, leading to a severe imbalance between supply and demand for high-performance CPUs in large-scale AI data centers. In the past two years, the narrative of the AI industry has been dominated by GPUs, with CPUs having a weak presence in AI servers. This is because, during the training era, the core bottleneck was parallel computing capability, with GPUs handling the heaviest matrix operations while CPUs were responsible for general control and basic scheduling work. However, with the explosive growth of AI agents and reinforcement learning workloads, the strategic position of CPUs in data centers is undergoing a structural reevaluation. The essence of AI agents is not to provide longer answers to questions, but to break down a single request into a series of workflows. Models are no longer just generating an answer but executing a process. Once AI transitions from "calculating once" to "running processes", the system's dependency on CPUs will significantly increase. This is because many critical workloads are not suitable for GPUs. Tasks such as task scheduling, thread management, process management, sandbox execution, pre and post-processing, cache coordination, and state maintenance are typical CPU functions. Especially in scenarios with multiple agents collaborating, where multiple agents run concurrently, call each other's tools, share states, there are higher demands on the number of CPU cores, threads, single-core performance, and memory management capabilities. Dylan Patel, Chief Analyst of the well-known semiconductor analysis firm SemiAnalysis, candidly stated in a deep interview last month that the paradigm of AI workloads is evolving from simple text generation to complex AI agents and reinforcement learning, and CPUs are facing "extremely serious capacity shortages". Market research firm TrendForce pointed out that the current ratio of CPUs to GPUs in AI data centers is approximately 1:4 to 1:8. In the age of AI agents, this ratio is expected to evolve to 1:1 to 1:2. In terms of market size, according to Creative Strategies' prediction, the data center CPU market size is expected to grow from $25 billion in 2026 to $60 billion by 2030; if combined with the demand for AI agents, the size is expected to approach $100 billion. This structural shift has triggered a chain reaction at both ends of supply and demand. Intel Corporation and AMD raised prices for some CPU product lines at the end of the first quarter of 2026. Meanwhile, NVIDIA Corporation and Arm both announced their entry into the server CPU market in March 2026 - the fact that a GPU giant and an IP licensor made the same choice in the same month is not a coincidence, but a concentrated release of market signals. Intel Corporation's Xeon processor had long held over 95% of the data center CPU market. However, this dominant position began to loosen in 2021 - issues with Intel 7 process technology delayed the release of Xeon Sapphire Rapids by nearly two years, opening a market gap for AMD's EPYC Milan. Intel Corporation plans to launch two flagship products in 2026. One is the Xeon 6+ (Clearwater Forest) based on the Darkmont architecture, with 288 cores/288 threads, and a TDP of approximately 450W; the other is the Xeon 7 (Diamond Rapids) based on the Panther Cove-X architecture, with up to 256 cores/256 threads, and a high TDP of 650W. Both products are based on Intel Corporation's most advanced 18A process technology and introduce the Foveros Direct hybrid bonding technology for the first time. However, TrendForce pointed out that due to ongoing issues with the 18A process technology, the mass production of these two products may be delayed until 2027. In contrast, competitor AMD's pace is more steady, with its flagship product EPYC Venice in 2026 adopting the N2 process technology from Taiwan Semiconductor Manufacturing Co., Ltd., the Zen 6 architecture, and featuring CoWoS-L and SoIC advanced packaging, achieving 256 cores/512 threads through SMT technology - the highest thread count in the current market. TrendForce predicts that AMD will continue to eat away at market share from Intel Corporation in 2026. In addition to the two giants Intel Corporation and AMD, a group of non-traditional players are entering the server CPU race at an unprecedented pace, seeking to fundamentally change the competitive landscape. In March, NVIDIA Corporation announced the Vera CPU as a standalone product to meet customer demand for more flexible CPU:GPU configurations. Vera uses NVIDIA Corporation's self-developed Olympus architecture, based on the N3 process technology from Taiwan Semiconductor Manufacturing Co., Ltd. and CoWoS-R packaging, offering 88 cores/176 threads, and equipped with 1.8 TB/s NVLink-C2C interconnectivity for memory sharing with NVIDIA Corporation GPUs. NVIDIA Corporation also introduced the Vera CPU rack, integrating 256 CPUs in a single rack, totaling 22,528 cores/45,056 threads and a total memory of 400 TB. Also in March, Arm announced the launch of its first self-developed CPU product, Arm AGI CPU, ending its 35-year history as a pure licensee. This product is based on the N3 process technology and Neoverse V3 architecture from Taiwan Semiconductor Manufacturing Co., Ltd., providing 136 cores/136 threads, a TDP of 300W, supporting DDR5-8800 memory and PCIe Gen6. Arm also simultaneously introduced two rack configurations: a air-cooled version integrating 60 AGI CPUs (8,160 cores, approximately 180 TB of memory), and a liquid-cooled version supporting 336 CPUs (45,696 cores, 1 PB of memory). Major cloud service providers (CSP) are also accelerating the development of self-developed CPUs. Amazon.com, Inc.'s AWS released the Graviton5 (192 cores/192 threads) based on the N3 process technology from Taiwan Semiconductor Manufacturing Co., Ltd. in December 2025, and deployed it in coordination with the self-developed Trainium 3 AI ASIC to reduce AI computing costs. Microsoft Corporation launched the Cobalt 200 (N3 process technology, 132 cores/132 threads) in November 2025. Alphabet Inc. Class C plans to launch the Axion C4A.metal bare-metal version and the next generation Axion N4A in 2026, focusing on the best cost-performance ratio. Citi states that as hyperscalers gradually shift from massive AI training expenditures to commercial deployments of AI systems, Intel Corporation, AMD, and Arm are increasingly actively pursuing CPU-related opportunities. In Citi's view, by the end of this decade, Intel Corporation will still maintain a leading position in the global CPU market, with an estimated market share of 47%; AMD is expected to occupy a market share of 34%, while Arm and other competitors are expected to collectively hold the remaining 19% market share. Citi has raised Intel Corporation's target price from $95 to $130 and maintains a "buy" rating on the stock. At the same time, the agency has raised AMD's target price from $358 to $460 but maintains a "neutral" rating.