Arm’s first CPU ever will plug into Meta’s AI data centers later this year
Foto: Stylized image showing the edge of a CPU
Up to 136 cores in a single processor and twice the performance-per-watt compared to x86 architecture – this is the Arm AGI CPU, the first-ever proprietary chip produced directly by the British giant. After decades of limiting itself solely to licensing designs to other entities, Arm is entering the market as a manufacturer, with Meta becoming its first key customer and co-creator of the solution. Despite attempts to develop his own chips, Mark Zuckerberg decided on close cooperation with Arm to power data centers handling advanced AI inference and autonomous agents. The new unit is based on the Neoverse platform and allows for the placement of up to 64 processors in a single air-cooled server rack. For the global ecosystem of creative technologies and cloud services, this means a drastic reduction in memory throughput bottlenecks and lower maintenance costs for AI infrastructure. Players such as OpenAI, Cloudflare, and SAP are already on the waiting list for the new chips. This move directly challenges the dominance of traditional Intel and AMD processors, offering companies without their own fabs a ready-made, high-performance solution for handling the most demanding language models. Arm is ceasing to be merely the architect of others' successes, becoming a real player in the race for hardware dominance over artificial intelligence.
For decades, the British giant Arm was associated exclusively with designing architecture, which it then licensed to other entities. A business model based on selling intellectual property rights defined the mobile and server markets, but the era of artificial intelligence is forcing a paradigm shift. On March 24, 2026, Arm announced a historic pivot: the company has produced its first fully proprietary processor, and the first customer to integrate it into its data centers is Meta.
The new chip, named the Arm AGI CPU, is not just another computing unit in the company's portfolio. It is a product designed from the ground up for inference processes—the key stage of AI operation where trained models generate responses to user queries. In an era of growing popularity for AI agents that can autonomously generate and manage multiple tasks simultaneously, the demand for cloud computing power is reaching levels that traditional solutions are beginning to fail to meet.
Neoverse Architecture and 136 Cores in a Single Chip
The foundation of the Arm AGI CPU is the proven Neoverse platform, which served as the basis for successes such as AWS Graviton and Nvidia Vera. However, Arm did not limit itself to replicating known patterns. The new processor offers a powerful configuration of up to 136 cores per single CPU unit. Such computational density is intended to translate directly into performance in multitasking environments, where it is not just the speed of a single operation that counts, but the throughput of the entire system.
Read also
What sets the Arm AGI CPU apart from the competition is a drastic improvement in energy efficiency. The manufacturer claims that its new processor offers twice the performance per watt compared to traditional x86 architecture processors. On the scale of the massive data centers managed by Meta, such a difference means not only financial savings running into billions of dollars but, above all, the ability to pack more computing power into the same physical space. These systems are optimized for standard air-cooled server racks, fitting up to 64 processors in a single rack.
Meta as a Strategic Partner and Co-creator
The choice of Meta as the first recipient and partner is no coincidence. The social media giant has long struggled with challenges when attempting to create its own proprietary AI chips. Collaborating with Arm allows Mark Zuckerberg to bypass these difficulties through a "co-developer" type partnership. Meta announced that it plans to work on "multiple generations" of these processors, suggesting a long-term strategy of moving away from total dependence on external suppliers of ready-made solutions.
However, it is worth noting that Meta does not intend to abandon its existing partnerships. The Arm AGI CPU is intended to work side-by-side with accelerators from Nvidia and AMD. This strategy shows that modern AI data centers are hybrid ecosystems where Arm processors take on the management role and handle inference, freeing up GPU resources for the most demanding tasks related to model training.
Global Support Chain and the Absence of a Giant
Arm's announcement triggered a wave of enthusiasm among tech industry leaders. Congratulations and declarations of support poured in from the biggest players, including Amazon AWS, Microsoft, Google, Marvell, Samsung, and even OpenAI. For many of these companies, the Arm AGI CPU is becoming an attractive alternative in situations where building their own processor from scratch is too costly or technologically risky. Mohamed Awad, head of Cloud AI at Arm, explicitly admitted that the company's goal is to fill the gap for entities that need dedicated silicon but do not want to go through the design process themselves.
The list of customers and partners also included companies such as:
- Cerebras and Rebellions (AI acceleration specialists)
- Cloudflare and F5 (security and network infrastructure)
- SAP (enterprise solutions)
- SK Telecom (telecommunications)
However, the absence of Qualcomm from the list of well-wishers is striking. This is a direct result of an ongoing legal conflict regarding the terms of licensing agreements. Although Qualcomm announced "total victory" over Arm in court last fall, the current move by the British company—entering the market with finished products—could completely change the balance of power in this rivalry.
Breaking Memory Bottlenecks
One of the most pressing problems in modern AI architecture is so-called memory bottlenecks. Data cannot be delivered to the processor fast enough, wasting the chip's computational potential. Arm claims that the Arm AGI CPU was designed in a way that radically reduces this phenomenon. By optimizing data flow within the chip, the unit is said to be able to maintain high efficiency even with the most complex queries generated by advanced large language models.
The decision by Arm, currently owned by SoftBank, to produce its own hardware is a risky but necessary step toward direct monetization of the AI revolution. Instead of collecting fractions of a cent for every core sold under license, the company is becoming a full-fledged infrastructure provider. For Meta, this is an opportunity to stabilize supply and optimize operational costs. Arm's entry into the ready-made server processor market marks the end of an era in which the company was merely a "silent architect"—now it is becoming a front-line player in the race for dominance over the infrastructure of global intelligence.
