Arm has announced its first own semiconductors, the Arm AGI CPU, marking the chip designer’s most significant strategic departure in its history. Rather than licensing designs to manufacturers, Arm will now produce its own chips targeting agentic AI workloads. The move puts the company in direct competition with Intel and AMD while testing relationships with some of its longest-standing customers.
CEO Rene Haas revealed the new chip at a press event in San Francisco, positioning it as the most efficient agentic CPU available. The company expects full production in the second half of 2026.
Agentic AI Hardware
The Arm AGI CPU takes its name from artificial general intelligence, reflecting the ambitious scope of tasks it aims to address. The chip is engineered to handle agentic AI workloads that require sustained operation over extended periods without human intervention.
Unlike traditional AI applications, agentic systems must maintain continuous operation while pursuing complex goals. Arm executives emphasised that the architecture prioritises energy efficiency alongside raw computational performance, claiming double the performance per watt against Intel and AMD x86 alternatives.
The design philosophy builds on Arm's mobile heritage, where power efficiency has always been a primary constraint. That experience translates naturally into data centre environments where electricity costs represent a significant operational expense.
Manufacturing Details
Taiwan Semiconductor Manufacturing Corporation fabricates the new chip using its advanced 3nm process, the cutting edge of semiconductor production. The technology allows more transistors per chip while maintaining or improving power efficiency.
Arm projects full production availability in the second half of 2026, aligning with anticipated surges in demand for AI-specific computing hardware. TSMC's involvement ensures the manufacturing partner has the capacity and expertise to deliver at scale.
The manufacturing timeline matters because several major customers have already committed to incorporation into their infrastructure. Delays in chip availability could affect customer adoption roadmaps and competitive positioning against established processors.
Customer Traction
Meta has emerged as the first major customer, having received samples of the CPU and expressing strong interest in the power efficiency advantages. The social media company's infrastructure team sees the chip as a key component in its AI hardware strategy.
Beyond Meta, OpenAI, Cerebras, Cloudflare, SAP, and Rebellions have all agreed to incorporate the Arm AGI CPU into their systems. The breadth of commitments suggests Arm has successfully positioned its chip as a viable alternative to offerings from established manufacturers.
OpenAI's vice president of science, Kevin Weil, appeared at the Arm event to discuss the company's ongoing need for computing capacity. Compute availability remains one of the primary constraints on AI development, and diversified supply chains are increasingly strategic.
The customer list includes companies that are also long-time Arm technology licensees, creating a complex competitive dynamic. Arm's move into manufacturing changes the nature of those relationships in ways that will take time to fully understand.
Competitive Landscape
Arm's entry into chip manufacturing puts it in direct competition with Intel and AMD, which produce x86-based processors for data centre applications. The company is also now competing with some of its own historical customers who have begun developing custom silicon.
Nvidia, which incorporates Arm-based CPUs in its rack systems, represents a particularly notable competitive dynamic. Nvidia has announced plans to sell stand-alone CPUs for the first time, entering a market that now includes its key design partner.
Industry analysts note that Arm risks alienating some partners by competing directly in chip manufacturing. The company's historical role as a neutral IP provider allowed it to serve virtually all major technology companies without conflicting commercial interests.
Nvidia CEO Jensen Huang appeared in a video testimonial at the Arm event without committing to purchasing the new chip. Amazon senior vice president James Hamilton and Google AI infrastructure chief Amin Vahdat also appeared, praising Arm's hardware without announcing purchase commitments.
Market Opportunity
Creative Strategies forecasts demand for AI CPUs in data centres will grow from $25 billion this year to $60 billion by 2030. When agentic AI-specific processors are included, the total addressable market could approach $100 billion by the end of the decade.
Arm's existing relationships with major technology companies provide a foundation for adoption of its new chip. The company's reputation for power efficiency in mobile applications translates directly into the data centre environment.
The addressable market expansion reflects the broader shift toward AI-specific computing architectures. Traditional CPUs designed for general-purpose workloads are increasingly being supplemented or replaced by specialised processors optimised for AI workloads.
Meta's head of infrastructure, Santosh Janardhan, emphasised at the event that power efficiency represents a critical factor in chip procurement decisions. As AI workloads have grown, electricity costs have become an increasingly significant portion of total cost of ownership for computing infrastructure.
The competitive response from established players will likely include accelerated roadmaps and increased investment in AI-specific processor development. Customers will benefit from more choices and potentially faster innovation cycles as major players compete more directly.