For 35 years, Arm made a simple promise to the tech industry: we design the blueprints, you build the chips. That promise ended on Tuesday.
The Cambridge-based company unveiled its first in-house processor—the Arm AGI CPU—at an event that felt less like a product launch and more like a confession. Arm, which has spent three and a half decades building the most pervasive architecture in computing by letting others do the manufacturing, is now doing the manufacturing itself. The company's CEO described it as "a natural evolution." That framing undersells what is actually a fundamental restructuring of Arm's identity.
The first customer for this new venture is Meta, which signed on as both lead partner and co-developer. According to statements from Meta's infrastructure team, the company plans to deploy the Arm AGI CPU across its data centers alongside existing hardware from Nvidia and AMD. This is not a small relationship: Meta reportedly struggled to bring its own custom AI silicon to market, and partnering with Arm gives the social media giant access to a processor designed specifically for inference workloads—the computationally intensive process of running AI models once they've been trained.
The customer roster extends beyond Meta. OpenAI, Cerebras, and Cloudflare are among the other companies that will purchase Arm's new hardware, according to Wired. This reveals the real business thesis: Arm is not trying to replace Nvidia in training workloads, where GPU dominance remains unchallenged. Instead, it's targeting the inference market, which analysts estimate will outpace training compute spending as AI applications shift from development to deployment at scale.
The strategic logic is clean. Arm knows data center operators better than almost any company on earth—its architecture powers virtually every smartphone and an increasing share of server infrastructure. By designing and selling its own chip, Arm captures margins that previously went to licensees like Apple, Qualcomm, and Amazon. It also gains direct insight into how its architecture performs in AI-specific workloads, feedback that could improve future licensing offerings.
The risks are equally clear. Arm now competes directly with customers it has served for decades. The company insists it will continue licensing, but the trust calculus changes when your supplier becomes your rival. Several large licensees have already begun diversifying their chip supply chains, a trend this move likely accelerates.
What Arm is not doing is pretending this is routine. The company acknowledged the weight of the moment internally, with executives noting that the decision to build silicon required reexamining assumptions baked into the company's DNA since its founding. The AGI CPU itself is a relatively conservative design—optimized for inference efficiency rather than raw performance breakthroughs. The real product is the precedent.
Whether Arm can execute a manufacturing business after excelling at licensing for 35 years remains an open question. What is already settled is the symbolism: the chip industry just lost one of its last neutral parties.