For more than three decades, Arm has thrived by licensing its computing architecture to the world's largest technology companies. Apple, Nvidia, Amazon, Google and Microsoft have all built processors based on Arm's designs, paying royalties on every chip produced. It was a remarkably successful formula: Arm reaped steady revenue without bearing the cost and complexity of manufacturing.
That model is now fundamentally changing. On March 24, 2026, Arm Holdings announced a landmark expansion of its compute platform into production silicon, unveiling its first Arm-designed data centre processor, the Arm AGI CPU, built specifically for agentic AI workloads.
The move represents a calculated gamble that the massive and growing investment in AI infrastructure will justify a departure from what made Arm profitable for decades. Meta is the lead partner and customer, co-developing the chip to optimise gigawatt-scale infrastructure for its services and alongside its own custom accelerators, while other launch partners include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom.
The Arm AGI CPU features up to 136 Neoverse V3 cores, 6GB/s memory bandwidth per core at sub-100ns latency, and a 300-watt power draw, enabling performance improvements that can more than double output per data centre rack compared with x86 servers. Like other fabless chipmakers, Arm manufactures the CPU at Taiwan Semiconductor Manufacturing Company's fabrication plants on the 3-nanometer node.
This shift raises genuine questions about what comes next. For Arm, the economics are compelling. The AI infrastructure market is expanding at extraordinary pace; Microsoft, Google, Amazon, and Meta are spending $380 billion or more on AI infrastructure in 2025. Selling its own chips allows Arm to capture a larger share of that spending rather than earning only licensing fees.
Yet the move also creates friction. It represents a major shift from only licensing chip architecture to giants like Apple, Nvidia, Amazon and Google to competing with them in physical silicon. When a company makes deals with its own customers to compete with them, relationships can strain. Some of Arm's largest licensees have invested billions in designing and manufacturing their own processors. If Arm competes aggressively, those companies might eventually choose alternative architectures such as RISC-V, fragmenting the ecosystem that has given Arm its market dominance.
For customers of Arm's new chip, the proposition is more straightforward. Companies without the resources to develop their own processors now have access to a production-ready product. While Arm has not disclosed pricing, analysts predict the CPU will cost thousands of dollars, with the company aiming for competitive pricing to serve companies that cannot afford the estimated 1,000 engineers and $500 million budget required to create their own processor.
The regulatory and geopolitical environment also adds complexity. Semiconductors are now recognised as critical national infrastructure, particularly for artificial intelligence. Arm's decision to manufacture through Taiwan and develop custom processors for major US technology companies will likely attract scrutiny from government bodies concerned with supply chain resilience and national security.
Arm's success will ultimately hinge on execution. The company must deliver products that outperform rivals, maintain relationships with existing licensees, and build a manufacturing operation that matches its design ambitions. The next few years will show whether Arm can balance these competing demands while exploiting the genuine opportunity that AI infrastructure presents.