Arm Holdings has manufactured its first proprietary central processing unit specifically for Meta’s artificial intelligence data centres, abandoning the licensing-only business model that has defined the British chip designer for more than three decades.
The company announced the AGI CPU—standing for Arm General Intelligence—will be deployed exclusively in Meta’s AI infrastructure, according to statements released by Arm and confirmed by multiple industry sources. This marks the first time Arm has moved beyond designing chip architectures for others to manufacture and sell, entering direct competition with its own licensees.
“This represents a fundamental recalibration of Arm’s position in the semiconductor value chain,” the company stated in its announcement. The chip is purpose-built for large language model training and inference workloads, addressing the specific computational demands of Meta’s AI systems including Llama models.
Strategic Implications
The move positions Arm directly against established AI chip suppliers including Nvidia, which currently dominates the data centre AI accelerator market with approximately 80% share, and AMD. More significantly, it places Arm in potential conflict with licensees such as Amazon Web Services, which manufactures its own Arm-based Graviton processors, and Microsoft, which has developed custom Arm chips for AI workloads.
Meta stands to gain reduced dependence on external chip suppliers and potentially lower costs through vertical integration. The social media company has invested heavily in AI infrastructure, with CEO Mark Zuckerberg stating in January that Meta would operate more than 350,000 Nvidia H100 GPUs by year-end, representing billions in capital expenditure.
For Arm, the shift opens a new revenue stream beyond licensing fees and royalties, though it risks alienating partners who may view the company as a competitor rather than neutral technology provider. Arm’s licensing business generated $1.68 billion in revenue for fiscal year 2024, with royalty income comprising a significant portion.
Technical Architecture
According to technical specifications released by Arm, the AGI CPU incorporates custom instruction sets optimised for transformer-based neural networks and includes dedicated silicon for matrix multiplication operations. The chip is manufactured using advanced process nodes, though Arm has not disclosed the specific fabrication partner or nanometre specification.
Industry analysts note that Arm’s energy-efficient architecture provides theoretical advantages for data centre deployment, where power consumption and cooling represent substantial operational costs. Data centres currently account for approximately 1-2% of global electricity consumption, with AI workloads driving increased demand.
Market Context
The announcement follows growing interest in custom silicon among major technology companies. Google has manufactured TPU (Tensor Processing Unit) chips since 2016, whilst Amazon, Microsoft, and Meta have all developed proprietary processors for specific workloads. This vertical integration trend reflects both the strategic importance of AI infrastructure and frustration with supply constraints and costs from traditional vendors.
Arm’s stock price rose 4.2% following the announcement, suggesting investor confidence in the strategic pivot. However, semiconductor industry observers have raised questions about manufacturing scale, supply chain management, and whether Arm possesses the operational capabilities required for chip production and distribution.
Industry Response
Nvidia declined to comment specifically on Arm’s announcement but emphasised its “comprehensive platform approach” combining hardware, software, and networking. Intel, which competes in the data centre CPU market, noted that “customer choice and open ecosystems remain paramount” in a brief statement.
The development warrants close monitoring of Arm’s relationships with existing licensees, particularly those developing competing AI chips. Whether other hyperscale cloud providers follow Meta in adopting Arm-manufactured silicon, and how this affects Arm’s licensing revenue, will determine whether this strategic shift succeeds or undermines the company’s core business model.
