Arm-based server CPUs

Why the concentration exists

Supply concentration stems from Arm's intellectual property position as the licensor of the Arm architecture for server-class processors. Arm licenses custom and semi-custom versions of its architecture to clients including Apple, Marvell, Nvidia, Samsung, and Qualcomm. This IP gatekeeper role means any server CPU using the Arm ISA must flow through Arm's licensing framework, creating a bottleneck at the architecture level.[13]

What the evidence shows

Who supplies it

Hyperscalers dominate Arm server CPU supply through in-house custom designs. AWS stated that more than half of the new CPU capacity it brought online over the last two years was based on its Arm-based Graviton chips. Google entered the Arm server CPU market with its Axion design announced at Google Cloud Next 2024 in Las Vegas, using Neoverse V2 cores. Microsoft has announced intentions to design its own Arm-based processors for Azure servers and its Surface line of PCs.[15][20][14]

Merchant suppliers serve the broader market outside hyperscaler internal use. Ampere Computing is estimated to hold between 5 and 10 percent of the Arm server CPU market, having grown from essentially nothing to account for 0.6 percent of all CPU instances available from cloud providers globally by mid-2022. NVIDIA supplies Grace CPUs, which are shipped in the company's Blackwell NVL72 AI rack platforms and drove nearly doubling of Arm server CPU shipments year-over-year in Q1 2026.[1][9]

Arm itself has entered the supply chain directly with its AGI CPU, marking the company's first production silicon designed specifically for agentic AI workloads. The AGI CPU was co-designed with Meta, which served as the lead partner on the project and was the first customer to receive samples. Commercial commitments have also come from Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom.[2][8][16]

What it depends on, and what depends on it

Arm server CPUs are built on Arm's Neoverse core designs, which licensees customize for their specific workloads. AWS Graviton2 uses Neoverse N1 cores, while Graviton3 uses Neoverse V1 cores. The Fujitsu A64FX is an Armv8.2-A compatible server CPU used in high-performance computing. Arm's AGI CPU packs up to 136 Neoverse V3 cores based on the Armv9.2 instruction set, with 2MB L2 cache per core and 6GBps memory bandwidth per core.[10][11][19][23]

The CPUs are manufactured at leading foundries on advanced process nodes. Arm's AGI CPU is etched in one of TSMC's N3 processes with transistors in the 3-nanometer range and a thermal design point of 300 watts. Intel has announced a deal to build a 64-core Arm Neoverse SoC using its Intel 18A process, potentially diversifying the manufacturing base beyond TSMC.[18][19]

Arm server CPUs are deployed in systems built by major OEM and ODM partners. Arm partners with ASRock Rack, Lenovo, Quanta Computer, and Supermicro for system integration. Early systems are available now with broader availability expected in the second half of the year. These systems serve data center workloads including AI inference, data analytics, web hosting, and cloud-native applications.[5][6][17]

Where it sits in the stack

What would break it

Arm's entry into direct chip sales with the AGI CPU creates potential conflict with its traditional licensing model. Arm now competes directly with its own licensees, including Ampere Computing, which SoftBank was reported to be in advanced talks to acquire. Qualcomm has signaled plans to enter the server market with the hiring of Intel Xeon chief architect Sailesh Kottapalli, adding another competitor to the Arm server CPU space.[4][20]

Legal disputes threaten supply stability for some Arm-based designs. Arm lost its lawsuit against Qualcomm over the Nuvia acquisition, with the jury unable to reach a verdict on whether Nuvia breached their agreement. Arm intends to seek a retrial due to the jury's deadlock. This litigation uncertainty could affect Qualcomm's ability to bring Arm server CPUs to market.[12]

x86 competitors are advancing their own process technologies to challenge Arm's efficiency advantage. AMD became the first company to ship 2nm server CPUs with EPYC Venice, announced at Computex 2026. Arm confirmed 50% of hyperscaler CPU compute at Computex 2026, marking a turning point in x86 versus Arm data center competition. This competitive pressure from x86 could slow Arm adoption if performance-per-watt advantages narrow.[3]

What to watch

Arm projects full production availability of the AGI CPU in the second half of 2026. Arm announced the AGI CPU on March 24, 2026. Arm CEO Rene Haas predicted the AGI CPU would bring in $15 billion in revenue by 2031, with total company revenue reaching $25 billion by that year.[16][23]

Server CPU demand growth is expected to top 50% year over year in 2027 across x86 and Arm-based chips. This surge spans Nvidia, Intel, AMD, and cloud-service-provider chips. Arm server CPUs are projected to capture over 45% of data center market revenue, up from around 25% in 2025, driven by accelerating rollout of in-house Arm CPU programs across major hyperscalers.[7][21]

Infrastructure investments will expand manufacturing capacity later in the decade. Micron's newest major memory fab in Singapore is scheduled to begin production in the second half of 2028. Amkor's new Arizona packaging facility is not expected to begin production until 2028.[22]

Related nodes

Sources

1. theregister.com · 2025-04-01T04:00:30
2. arm.com
3. techtimes.com · 2026-06-02T17:52:20
4. forbes.com · 2026-03-24T20:02:32
5. electronicsweekly.com · 2026-03-28T12:45:57
6. newsroom.arm.com · 2026-03-27T13:22:46
7. tomshardware.com
8. tomshardware.com
9. theregister.com · Q1 2026
10. en.wikipedia.org
11. en.wikipedia.org
12. arstechnica.com
13. zdnet.com · December 2018
14. jumpcloud.com · February 2021
15. nextplatform.com · 2024-04-09
16. wired.com · 2026-03-24
17. techtarget.com
18. extremetech.com
19. nextplatform.com
20. crn.com · December
21. winbuzzer.com
22. randtech.com
23. hpcwire.com · March 26, 2026