The situation of server CPUs has changed.

“By 2030, the semiconductor value of data center servers will reach $500 billion.” This prediction by market research firm Yole points to a rapidly expanding market segment. Mario Morales, vice president of semiconductor research at IDC, is even more straightforward in his assessment: “In the next five years, data centers will be the fastest-growing segment in the entire semiconductor industry.”

It seems that everyone is optimistic about the data center market.

As the core component of data centers, server CPUs are becoming increasingly important and complex.

 01 Server CPUs: A Landscape of Change

In the server CPU market, x86, ARM, and RISC-V are engaged in a silent architectural revolution. Each architecture is brewing unusual changes in its respective market.

x86 Architecture: Power Dynamics Shifting

Over the past decade, the x86 architecture has dominated the server CPU market. Intel, with its deep technological expertise and comprehensive ecosystem, has held a near-monopoly in this market for years, creating an unshakable industry landscape.

However, this stable situation is quietly changing.

In Q2, Intel still held the majority of the server CPU shipment market share at 72.7%, but both year-on-year and quarter-on-quarter figures declined.

In contrast, due to the upgrade of its latest EPYC platform, AMD achieved a strong year-on-year growth of 3.6% and a quarter-on-quarter growth of 2.1% in Q1 2025. In Q2, the company's market share reached 27.3%.

AMD's journey in the server CPU market is a story of building from scratch. Since 2017, AMD has gradually established itself in the market with the launch of its EPYC series of processors. At that time, AMD had no competitive products and its market share was almost zero.

After Lisa Su took over as CEO, AMD focused on the workstation market as a strategic priority and launched the EPYC Rome series of processors, which were the first to use the Zen architecture. In just a few years, the EPYC product line has enabled AMD to achieve significant market penetration. Market data shows that in 2026, AMD is expected to become the largest x86 CPU supplier in data centers.

Lisa Su said in a recent media interview that the company's influence in the server CPU market is constantly growing. It has become an important partner for tech giants such as Microsoft and Amazon, and both its market share and revenue share have reached record highs.

Lisa Su pointed out that many customers now regard AMD as a strategic CPU supplier. Its market position in the data center CPU field can be compared to NVIDIA's dominant position in the AI accelerator field.

Arm: A Growth Rate of 70.0%

Since 2018, Arm has gradually closed the performance gap with x86 server CPUs by launching multiple series of IP core products such as Neoverse N/V/E, and has shown obvious advantages in energy efficiency and cost. This trend coincides with the wave of cloud service providers developing their own Arm server CPUs.

The core advantage of Arm architecture servers lies in their low-power design, which makes them particularly suitable for high-concurrency scenarios such as AI training and edge computing.

Amazon AWS was the first cloud provider to deploy Arm CPUs on a large scale. Since the release of Graviton in 2018, it has now reached the fourth generation, with a cumulative shipment of over 2 million units.

Google Cloud and Microsoft Azure have also released their own Arm-based self-developed processors, Axion and Cobalt, respectively. They have optimized their core services through hardware-software integration, and internal evaluations show that they have a 30%-60% improvement in overall performance and power consumption compared to x86.

There are also chip design companies that specifically provide Arm server CPUs to customers, such as Ampere Computing.

In addition, the launch of large-scale rack configurations by NVIDIA and others has also significantly increased market attention on Arm servers.

Since the launch of the Grace CPU in 2022, this product based on the Arm Neoverse architecture has shown unique collaborative advantages. It forms a computing module with Hopper and Blackwell series GPUs, enabling unified memory access and task scheduling through the high-bandwidth NVLink interconnection technology. Notably, in the new NVLink 72 architecture in 2025, Grace is the only compatible platform.

The actual deployment data of the Grace CPU reflects the deep integration of the AI computing system with the Arm architecture. For example, it is estimated that about 5 million Blackwell GPUs will be shipped in 2025. According to the ratio of one Grace CPU for every two GPUs, this means that about 2.5 million Grace chips will be deployed that year. Therefore, Arm's market share will continue to grow.

IDC data shows that in 2025, the x86 architecture server market will continue to lead the trend, with an expected growth of 39.9%. The total market size will reach $283.9 billion. The x86 architecture remains the top choice for enterprises due to its compatibility and wide range of applications, especially in the cloud computing and data center fields. However, the performance of the non-x86 server market should not be underestimated. It is expected to grow by 63.7% year-on-year, with a total market size of $82 billion. Among them, the growth of Arm architecture servers is particularly significant, with an expected growth rate of 70.0%, and its share in the global total shipments will reach 21.1%.

However, this is still far below the target of the chip design company. Mohamed Awad, head of Arm's infrastructure business, claimed in April that Arm architecture servers would account for 50% of global total shipments this year. He also predicted that by the end of this year, Arm would account for half of the global data center CPU sales.

Some analysts do not agree with the prediction that Arm will capture 50% of the data center CPU market in 2025. Manoj Sukumaran of market research firm Omdia believes that by 2025, Arm-based CPUs may only account for 20%-23% of the global data center market, rather than the 50% claimed by Arm.

Nevertheless, it is foreseeable that in the data center CPU market, which has been dominated by x86 for many years, Arm's influence and market share are continuously expanding, posing an increasing threat to x86.

RISC-V: Breaking into the Mainstream

In addition to the mainstream x86 and Arm architectures, server CPUs based on the RISC-V architecture, as a new path, have also attracted more attention.

As a new architecture, RISC-V currently has less influence than Arm.

Previously, the RISC-V architecture had great success in the Internet of Things (IoT) field, which values energy efficiency. However, this does not mean that RISC-V cannot enter the PC and server markets, which have higher performance requirements.

Different from the x86 and ARM architectures, which have licensing restrictions, RISC-V has features such as simplicity and open source. It can be used to develop unique chips that better meet specific needs and breaks the convention of high licensing fees and difficult customization of the x86 and ARM architectures.

As an open-source standard, RISC-V allows any individual or organization to use, modify, and expand it freely without paying patent fees. Its design goal is to provide a simple, scalable, and flexible instruction set. The modular design enables users to freely customize different instruction subsets according to their needs, making it suitable for a wide range of applications, from embedded microcontrollers to high-performance computing.

Xuantie, a brand under Alibaba's DAMO Academy, has always been a leading player in the RISC-V market. In 2018, Xuantie released its first RISC-V architecture processor IP, the Xuantie E902. In 2019, it released the Xuantie C910 processor, which could reach a clock speed of up to 2.5 GHz, breaking the performance record of RISC-V processors at that time. In November 2023, Xuantie launched the C920, an upgraded version of the high-performance RISC-V processor C910, with a clock speed increased to 3 GHz.

Then in February this year, Xuantie announced its first server-level CPU IP core based on the open-source RISC-V architecture, the Xuantie C930, which began delivery in March.

A technical expert from DAMO Academy said that “just as x86 and Arm architectures each have their killer applications, high performance + AI could become the killer application for RISC-V, which is its differentiating advantage.”

In March this year, Shenzhen-based RISC-V leader Raisecore Technologies launched a new generation of high-performance RISC-V server chips, the Lingyu processors. These chips feature a “dual-core in one” design with a 32-core CPU + 8-core LPU, achieving significant improvements in both single-core and multi-core parallel performance. Their computing performance is comparable to that of international mainstream server chips from Intel and AMD.

It's not just these companies. The industry has been exploring RISC-V server CPUs for some time. As early as December 2022, Ventana released the world's first RISC-V architecture server CPU, the Veyron V1.

The application of RISC-V server CPUs has also attracted the attention and raised questions among many industry insiders. Zhang Xiantao, president of Alibaba Cloud's Wuying division, said that “I think after 5 - 8 years of development, large-scale application in servers should not be a problem.”

Xun Dandan, CEO of Microcore, said that high-performance computing scenarios are the key to opening up the RISC-V market. “In the past, most RISC-V applications were in the AIoT field. The low-cost, low-performance segment has become a highly competitive market. For the RISC-V industry, as an emerging instruction set, its growth and expansion are closely tied to the development of emerging applications in high-performance areas (servers, AI computing) in data centers.”

 02 Server CPUs: More Players Enter the Arena

With the rapid growth of the server market, a new wave of competition is emerging in the server CPU field, with more chip giants joining the fray. Notably, the two chip leaders mentioned below have both placed their bets on the Arm architecture.

Qualcomm: Returning to the Server Chip Market

Qualcomm has been preparing for the server CPU market for many years.

Qualcomm launched the Centriq server CPU in 2016. However, due to the weak Arm server ecosystem at that time and Qualcomm's own challenges, such as patent lawsuits and attempted hostile takeovers, its server business was forced to be abandoned in 2018. Subsequently, Qualcomm acquired Nuvia, a startup focused on the design of Arm architecture server CPUs.

At the beginning of this year, Qualcomm revealed that its data center team is developing a “high-performance, energy-efficient server solution” to meet the needs of data center applications.

In June, Qualcomm announced an agreement to acquire all of the issued and to-be-issued common stock capital of Alphawave Semi. The enterprise value of this transaction is approximately $2.4 billion. Alphawave Semi is a leader in semiconductor IP for high-speed wired connectivity and computing technologies. It provides IP, custom chips, connectivity products, and chip sets, enabling faster and more reliable data transmission with higher performance and lower power consumption.

Qualcomm said that the acquisition of Alphawave Semi aims to further accelerate its entry into the data center server field and provide it with key assets. Qualcomm's Oryon CPU and Hexagon NPU processors can meet the growing demand for high-performance, low-power computing, which is driven by the rapid growth of AI inference and the transition of data centers to custom CPUs.

During the recent Q3 earnings conference call, Qualcomm confirmed that it is in “in-depth discussions” with a hyperscale cloud service provider regarding a new server chip, and expects to generate revenue in fiscal year 2028.

Nvidia CPUs: New Achievements

As a giant in the GPU field, NVIDIA's layout in the CPU field is also highly noticeable.

The NVIDIA Grace CPU is a breakthrough Arm CPU with outstanding performance and efficiency. It can be closely integrated with GPUs to enhance accelerated computing capabilities or deployed as a powerful and efficient standalone CPU. The NVIDIA Grace CPU is the foundation for the next generation of data centers and can be configured in various ways to meet the needs of different data centers.

At the CES exhibition in January this year, NVIDIA officially launched its first personal supercomputer, Project DIGITS, which is equipped with the NVIDIA GB10 Grace Blackwell Superchip developed in collaboration with MediaTek. This superchip consists of a Blackwell GPU and a Grace CPU, along with 128GB of LPDDR5X memory and 1TB/4TB NVMe SSDs. It can run large language models with over 200 billion parameters.

The GB10 Grace Blackwell Superchip is equipped with NVIDIA's latest generation of CUDA cores and fifth-generation Tensor cores, providing up to 1 petaflop of AI performance. It is connected to the energy-efficient NVIDIA Grace CPU through NVLink-C2C inter-chip interconnect. The Grace CPU includes 20 performance cores based on the Arm architecture, jointly designed by NVIDIA and MediaTek, a market leader in Arm-based system-on-chip design, offering top-notch performance, energy efficiency, and connectivity for the GB10 chip.

These two examples also demonstrate that Arm architecture server CPUs are highly suitable for data center applications.

As for how far Arm can go in the future and whether it can rival the x86 architecture, it's still unknown. However, considering its advantages in energy efficiency and cost control, as well as the increasing investment from more manufacturers, it is opening up new growth opportunities. With the diverse computing power requirements in scenarios such as cloud computing and edge computing, the Arm architecture is gradually taking a more important position in the server field and has become a force that cannot be ignored.

This article is from the WeChat official account “Semiconductor Industry Insights” (ID: ICViews), written by Feng Ning, and is published by 36Kr with authorization.