Arm pushes its battle lines deep into the x86 stronghold

On September 8, at the Arm Everywhere China conference held in Shanghai, Arm simultaneously updated its three product lines covering mobile, cloud and Physical AI: it launched the second-generation computing subsystem CSS for Mobile 2 for the mobile segment, released the Neoverse CSS N4 for the cloud, continued to advance the AGI CPU unveiled in March this year, and further expanded the Total Design ecosystem for Physical AI.

The more noteworthy changes are mainly taking place in the cloud segment.

In the past, Arm mainly made profits through architecture and IP licensing, with customers including Qualcomm, Apple, Amazon, and Google completing chip design on their own. Now, ranging from IP and CSS to the directly deliverable AGI CPU, Arm is gradually incorporating the links that were previously completed by customers into its own product portfolio.

Some Chinese companies are also joining the Arm camp. Volcano Engine plans to introduce Arm's AGI CPU into its data centers for next-generation AI services, Lenovo is developing AI infrastructure around the AGI CPU, and Arm's cooperation with Unigroup covers both AGI CPU and CSS at the same time.

As the Arm server ecosystem continues to expand, the x86 camp that has long been dominated by Intel and AMD is facing greater share pressure. For Arm, the new challenge has evolved from "getting more chips to adopt Arm architecture" to "having more chips sold directly by Arm itself".

Part One

Three-pronged Development

The core business Arm has always been best at is letting others pan for gold.

Over the past years, companies including Qualcomm, MediaTek, Apple, as well as later cloud players Amazon and Google, have all been able to design their own chips based on Arm's architecture and IP. Arm does not need to judge who will eventually win the market, as long as more chips adopt its technology, it can share the growth through licensing fees and royalties.

This fundamental business model is still highly profitable today. In the first quarter of fiscal year 2027, Arm posted revenue of 1.289 billion US dollars, a year-on-year increase of 22.4%; among which licensing revenue reached 574 million US dollars, royalty revenue hit 715 million US dollars, and the non-GAAP operating profit margin reached 41.2%.

Behind this model, mobile phones have always been Arm's largest revenue base. In fiscal year 2025, smartphone application processors still contributed about 45% of Arm's royalty revenue, corresponding to a market share of 99% with limited room for further growth.

It is against this background that Arm has been continuously exploring new growth opportunities.

AI is obviously the largest incremental market at present. According to Arm's long-term plan released this year, by fiscal year 2031, the total market size corresponding to Edge AI, Physical AI and Cloud AI will expand from the current about 535 billion US dollars to more than 1.5 trillion US dollars, among which cloud AI is one of the most important sources of increment.

This has also prompted Arm to extend its business that used to stay at the upstream of the industrial chain further downstream.

From IP to CSS that has completed a large number of system integration work for customers, and then to the mass-production-ready AGI CPU first launched in March this year, the products Arm delivers to customers are getting more and more complete.

Ami Badani, Chief Marketing Officer of Arm, said in an exchange with media including Wall Street See · All Weather Tech that this roadmap largely comes from customer demand. After Arm launched the Compute Subsystem (CSS), more and more customers hope Arm can go a step further to provide products that are closer to finished chips.

"That is one of the main reasons why we launched the Arm AGI CPU," Badani noted, representing a continued extension of Arm's business model.

At the Arm Everywhere China conference on September 8, Arm almost fully demonstrated this business boundary expansion to the public.

In the mobile market that Arm is most familiar with, the company launched the second-generation mobile terminal computing subsystem CSS for Mobile 2, integrating the brand new C2 CPU cluster, Mali G2-Ultra NX GPU, system IP, physical implementation solutions and software support into the same platform.

Among them, Mali G2-Ultra NX integrates a dedicated neural network accelerator for the first time in Mali GPUs. Arm also continues to extend its cooperation scope to the application end: vivo will introduce relevant technologies into its flagship phones, NetEase's *Where Winds Meet* will promote the implementation of neural super sampling, while Tencent Games and Unity China will participate in the adaptation from the game middle platform and engine layer respectively.

Changes in the cloud segment are even more notable.

The newly released Neoverse CSS N4 is the cloud CSS with the highest configurability in Arm's product line so far. A single die can carry up to 128 cores, support LPDDR6 and PCIe Gen 7, which is expected to further shorten chip development time.

The AGI CPU directly brings Arm to a position closer to end products. This data center-oriented CPU adopts 3nm process, currently features a 136-core design, and supports PCIe Gen 6 and CXL 3.0.

Up to this point, Arm has formed three product lines with different levels of depth:

Customers with full chip design capabilities who want to retain more room for differentiation can continue to purchase IP; customers who hope to design chips independently while reducing underlying integration work and shortening development cycles can adopt CSS; customers who prioritize rapid deployment and have no need to design CPUs on their own can directly use the AGI CPU.

The three product forms correspond to different trade-offs made by customers between independent design, development cycle and deployment speed, and further expand the scope of customers that Arm can cover.

However, as Arm extends its business to finished chips, potential competition has begun to emerge in the previously simple upstream-downstream relationship.

AWS already has Graviton, Google has Axion, Microsoft has Cobalt, and NVIDIA has also launched the Arm architecture-based Vera.

In the past, these companies' self-developed CPUs based on Arm architecture meant Arm could continue to obtain IP licensing and royalty revenue. But with Arm's AGI CPU entering the market, competition arises between the two sides.

In response to this, Badani said in an exchange with media including All Weather Tech that the AGI CPU is defined as an expansion of the market space, and emphasized that IP, CSS and AGI CPU will still coexist for a long time in the future.

Badani took SAP as an example: part of the company's cloud business is currently running on AWS Graviton servers based on Arm architecture, but a large number of on-premise deployments for enterprises are still running on chips of other architectures. For Arm, entering AWS through IP in the past does not naturally grant access to all SAP's computing environments.

The AGI CPU provides a path to directly enter these enterprise on-premise deployment scenarios. Arm hopes to extend the coverage of Arm architecture to markets that were previously difficult for IP and CSS to reach, and promote more software to complete Arm adaptation at the same time.

Part Two

The Chinese Mainland Camp Is Gradually Taking Shape

With the arrival of the Agent boom, the CPU market is expected to expand.

Arm's management judges that with the increase of Agentic AI workloads, the configuration ratio of CPUs and GPUs may gradually approach 1:1 in the future.

This is unlocking incremental space for CPUs. JPMorgan Chase predicts that the shipment of dedicated CPUs for Agentic AI will increase from 1.6 million units in 2025 to 27 million units in 2028, with a compound annual growth rate of 155%.

At present, a number of Chinese mainland companies have clearly planned to introduce Arm's self-developed chips.

Arm disclosed this time that Volcano Engine is building next-generation cloud services based on the Arm platform, and plans to launch an agent sandbox solution based on its AGI CPU.

When All Weather Tech further inquired about the progress of the cooperation between the two parties on site, Eddie Ramirez, Vice President of Cloud AI Business Expansion at Arm, said that the two parties are currently working closely to introduce the AGI CPU into Volcano Engine's data centers and run next-generation AI services on it. The specific deployment timeline will be announced later by Volcano Engine.

Lenovo is also developing next-generation AI infrastructure around the AGI CPU. The cooperation between Arm and the new Unigroup covers both AGI CPU and CSS, and further extends to agent orchestration, CXL memory pooling, rack-level system architecture and edge AI.

However, while the server CPU market is expanding, the original competition landscape is also being reshaped.

The first to bear the brunt is the x86 camp that has long occupied a dominant position.

In the past, the server CPU market was mainly controlled by Intel and AMD. Now, server CPUs based on Arm architecture are entering data centers from more directions.

This means that x86 is facing the expansion of the entire Arm server ecosystem.

JPMorgan Chase predicts that the share of Arm architecture in the server CPU market may rise from about 22% in 2025 to 43% in 2028. Even if the market expansion brought by Agentic AI can still allow x86 to maintain growth, the share pressure faced by Intel and AMD may continue to rise.

The competition within the x86 camp is also fierce. AMD is continuously taking market share from Intel. JPMorgan Chase estimates that AMD's server CPU share reached about 44% in the first quarter of 2026, and it continues to promote new generation products such as Venice and Verano to compete for new demand in AI servers.

For the x86 camp, the current problem is how to defend its market share; for Arm, the bigger problem is how to further translate the architecture dividend into chip revenue.

Agentic AI is bringing server CPUs back to the center of the stage. Arm has already had a growing architecture base, and the next step is to prove that it can build a sufficiently large chip business on this foundation.

This article is from the WeChat official account "All Weather Tech" (ID: iawtmt), written by Zheng Minfang, authorized for release by 36Kr.