JPN
Insight

Arm 2.0: Going Beyond Licensing to Capture Silicon Value with AGI AI CPU

0
March 26, 2026
  • The early momentum for the Arm AGI CPU looks massive, with the likes of Meta, OpenAI, Rebellions, Positron AI and F5 adopting the AGI CPU. 
  • The Arm AGI CPU addresses critical gaps in the current entrenched data center market through several strategic approaches. 
  • The most striking element of this launch is the financial trajectory it sets for the company. 


Arm has fundamentally shifted the goalposts with the launch of the Arm AGI CPU for the data center. By moving across the value chain to provide a full silicon platform rather than just IP, Arm is enabling its customers to capture more value while accelerating the industry’s transition into the AI era. 

The early momentum looks massive, with the likes of Meta, OpenAI, Rebellions, Positron AI and F5 adopting the AGI CPU. Clearly, there is a big market appetite for an alternative to the traditional x86 dominance. 

Source: Arm

Value Creation:  

The Arm AGI CPU addresses critical gaps in the current entrenched data center market through several strategic approaches covering design, performance, efficiency, scale, and years of experience developing and licensing Neoverse cores and subsystems. 

Targeting x86 starting with the ‘AI ASIC’ integration gap 

According to Counterpoint Research, AI hyperscalers are projected to ship more than 10 million AI ASICs in 2027 (AI Server Compute ASIC Shipments to Triple by 2027 as Custom Silicon Enters Hyper-Growth Phase). 

Source: Counterpoint Research

Currently, most of these ASICs are “attached” to x86 CPUs. Arm’s AGI CPU provides a purpose-built, native alternative that eliminates the inefficiencies of mixing architectures, offering a seamless Arm-on-Arm environment for AI accelerators. 

AGI CPU capabilities 

The AGI CPU isn't just an incremental update; it is a high-performance beast designed for the most demanding workloads: 

Source: Arm

  • Compute power: Up to 136 Neoverse V3 cores clocked at 3.7GHz, supported by a dedicated 2MB L2 cache per core. 
  • Advanced chiplet design: A dual-chiplet architecture that places memory and I/O on the same die to achieve sub-100ns memory latency, critical for AI training and inference. 
  • Massive throughput: Support for DDR5-8800, providing 6GB/s of memory bandwidth per core, coupled with 96 lanes of PCIe Gen6 and CXL 3.0 for memory expansion. 


Unmatched efficiency at GW scale 

As these AI hyperscalers roll out Gigawatt-scale data center deployments, power remains the primary constraint. 

Source: Arm

  •  Capex/Opex savings: Arm is claiming 2x more performance per rack, translating to $10 billion in capex savings for a 1GW deployment thanks to its DNA of power efficiency.


Source: Arm

  •  Purpose-built: Further leveraging its low-power architecture versus x86, TSMC 3nm process and a 300-watt TDP, the AGI CPU delivers maximum compute density without the thermal tax of legacy architectures. 


Experience and ecosystem advantage 

  • Arm Neoverse experience: With over 125 billion Neoverse cores already shipped, Arm has the world's most robust developer ecosystem for data centers, comprising more than 10,000 enterprises and millions of developers. This installed base ensures that applications can be ported to the AGI CPU with minimal friction. 

Source: Arm

  • Deep partnerships: Furthermore, deep optimizations with memory leaders (Micron, SK hynix, Samsung) and manufacturing partners (TSMC, Amkor) ensure that the hardware is optimized from the transistor level to the software stack. 
  • Solid ecosystem support: The number of partners and potential customers excited about the Arm AGI CPU is surreal.  


Value Capture: 

Until now, Arm was just a pure-play enabler through its IP and CSS initiatives, watching from the sidelines. However, the DC CPU market, dominated by x86, is up for some competition, also demanded by multiple Arm customers. There is a significant opportunity and upside for Arm to capture value. 

Revenue and financial upside: The path to $25 billion 

The most striking element of this launch is the financial trajectory it sets for the company: 

  • TAM capture: Arm is targeting a $15-billion chip revenue stream by FYE31. In a data center CPU market projected to hit $100 billion, this represents a 15% market share, a massive leap from its historical footprint. 
  • From cents to dollars: Traditionally, Arm could earn a few dollars in royalties per server chip. By selling the full AGI CPU, the average selling price (ASP) shifts to the hundreds or thousands of dollars per unit, representing a 10x to increase in revenue per socket. 
  • Gross profit boost: Arm is estimating earning $500 per chip in gross profits vs $50 currently through selling IP or $100 per chip through CSS as an example based on an illustrative pricing of a $1,000 chip. This is a 10x opportunity to maximize the profit $$.
  • The ramp: Arm expects FYE28 to be the first year for meaningful Arm AGI CPU revenue and its exponential growth from there. 
  • Consolidated scale: Between $10 billion in IP/CSS revenue and $15 billion in AGI CPU sales, Arm is targeting a $25-billion annual revenue by 2031. 
  • Operating leverage: The consolidated business model targets over $9 in EPS. When compared to current levels (around $0.80), this represents an 11x growth in earnings power. This is driven by the high-margin nature of the IP/CSS business (>65% margin) providing the foundation for the high-volume chip business (>30% margin).  


Source: Arm

Competition: 

  • The pressure is on the x86 camp to protect its market share and position. It remains to be seen how Intel and AMD react, whether with pricing or a better architecture approach or software, areas they planned to address with their x86 Ecosystem Advisory Group formed 18 months ago. 
  • While AI hyperscalers look to diversify with heterogeneous architecture, with some already licensing Arm Neoverse cores, those with strong internal teams may be less inclined to buy the Arm AGI CPU off-the-shelf, which could limit some TAM.   


Bottom line:  

  • By building on the Neoverse IP and CSS foundation, Arm is accelerating time-to-market for the entire ecosystem, creating additional value and aiming to capture the same.  
  • Arm no longer wants to be on the sidelines and leave money on the table. This approach to go after x86 is a prudent strategy. This might also work next in the PC market, but no certainty about the smartphone, IoT and automotive markets. 

#ArmEverywhere

You can find more research on foundry, compute, memory and manufacturing to end markets here: 

Counterpoint Insights | Technology Market Research and Industry Analysis Firm 

 Follow Counterpoint Research

Twitter
LinkedIn

Receive our insightful weekly newsletter and stay ahead of the competition.

Author

Neil Shah

twitter_icon
linkedin_icon

Neil is a sought-after frequently-quoted Industry Analyst with a wide spectrum of rich multifunctional experience. He is a knowledgeable, adept, and accomplished strategist. In the last 18 years he has offered expert strategic advice that has been highly regarded across different industries especially in telecom. Prior to Counterpoint, Neil worked at Strategy Analytics as a Senior Analyst (Telecom). Neil also had an opportunity to work with Philips Electronics in multiple roles. He is also an IEEE Certified Wireless Professional with a Master of Science (Telecommunications & Business) from the University of Maryland, College Park, USA.