The CPU is back and Arm wants to lead the next chapter

For decades, the CPU was the most important component in everything from laptops to data centers. Then the AI boom took off. GPUs, particularly Nvidia's graphics processors which can cost over $30,000 each, proved superior at training and running large AI models. The CPU lost its leading role.

However, the situation is changing. The emergence of agentic AI-systems that act autonomously, make decisions, and manage workflows in real time - is driving a new type of demand. These AI agents require powerful CPUs for orchestration, memory management, and data movement, tasks that GPUs aren't designed for. Jeroen Kusters, semiconductor expert at Deloitte, noted that "there's a rebalancing toward CPUs" and that demand for them is now surging (WSJ, 25.03.26). Last week, Nvidia confirmed this trend last week by launching its Vera CPU as a standalone product for the first time, without an accompanying graphics card. Arm Holdings chose precisely that moment to act.

For over three decades, Arm has licensed its processor designs to some of the world's largest tech companies including: Apple, Nvidia, Amazon, Google, and Qualcomm. Over 325 billion devices have been built on Arm's platform. The business model has relied on licensing fees and royalties, giving the company a gross margin of 98%, one of the highest in the entire tech industry (Bloomberg, 25.03.26).

On March 24, 2026, CEO Rene Haas unveiled the company's first in-house processor, called the Arm AGI CPU. "People thought the CPU was dead," Haas said at the launch. "But with AI, you need more and more CPUs. Many more." (WSJ, 25.03.26). Manufactured at TSMC using 3-nanometer technology, the processor is offered in three variants with up to 136 cores, and according to Arm is twice as energy-efficient as comparable x86 systems. However, those performance figures are based on the company's own estimates and should be interpreted cautiously pending independent testing.

The development establishes Arm as a direct competitor not only t to Intel and AMD, but potentially also to several of its own customers. The stock rose 16% the day after the launch, marking the largest single-day gain in nearly a year.

Arm's current licensing model gives the company a 98% gross margin. So why switch to hardware production, which has substantially lower margins? CFO Jason Child explained his reasoning as follows: Arm earns approximately $50 on a chip sold for $1,000, if the customer only uses the company's instruction set. If the customer licenses a complete chip design, however, the profit reaches around $100. But if Arm manufactures and sells the chip itself, the gross profit lands at approximately $500 (Bloomberg, 25.03.26). While the gross margin drops from 98% to around 50%,  the profit in dollars per chip increases tenfold. Moreover, data center chips can cost tens of thousands of dollars each, which is an entirely different price tier than the smartphone processors that have until now formed the core of Arm's business. That's the calculation driving the strategic shift.

Arm has traditionally functioned has been to function as a neutral licensor, collaborating with all sides of the industry. However, that claim becomes harder to defend when the company starts selling its own chips in the same market as its licensees.

Nvidia CEO Jensen Huang congratulated Arm at the launch event, but there is clear overlap. Nvidia recently launched its standalone Vera CPU series targeting the same data center market. Analysts assess, however, that the rivalry is limited. Mark Lipacis at Evercore ISI argued that Arm's energy-efficient architecture has a structural advantage in agentic AI infrastructure, comparable to the position Nvidia built early in the AI rally (WSJ, 25.03.26). Haas emphasized that the market is "enormously large with room for multiple players." On the same day, Intel rose 7% and AMD 7.3% suggesting that investors view Arm's entry as a sign that the market is expanding rather than being redistributed.

Meta is the lead partner in developing the AGI CPU and has signed an agreement with Arm that the company describes as a "multi-generation collaboration" (FT, 24.03.26). Other confirmed customers include OpenAI, Cerebras, Cloudflare, and SAP. However, the most interesting market opportunity does not involve the major cloud providers. Amazon, Google, and Microsoft already design their own Arm-based processors for their data centers. Instead, the natural target audience is made up of  companies that are still running business-critical software on legacy Intel and AMD servers. Banks, industrial companies, and government agencies that have built their IT systems around SAP, Oracle, and similar platforms are facing a technological shift, and Arm is positioning itself as an alternative. Haas added that the chip doesn't fall under U.S. export controls and could potentially be sold in China, a market that could add significant volume in the future.

At the launch, Haas presented aggressive financial targets. The AGI CPU is expected to generate $15 billion in annual revenue by 2031, which is 50% more than the licensing business is projected to achieve by then. The company's total revenue target is $25 billion with earnings per share of $9 (Bloomberg, 25.03.26). By comparison, Arm generated approximately $4.9 billion in revenue during the fiscal year ending March 2026. That means the company is aiming for a five-fold revenue increase over five years. JP Morgan estimates that the market Arm is targeting will reach $100 billion by 2030 (Barron's, 25.03.26).

The valuation already reflects high expectations. Arm now trades at 81 times expected earnings, which is one of the highest multiples in the semiconductor sector and nearly four times that of Nvidia (WSJ, 25.03.26). HSBC analysts described the launch as a "game-changer," but at the current valuation, the company needs to deliver results that align with its forecasts to justify the share price.

Vontobel offers a wide range of leveraged products with Arm Holdings as the underlying asset, providing investors with the opportunity to take both long and short exposure based on their own market view.

The bull case for Arm is based on several converging factors. First, the semiconductor market is expected to reach around $975 billion by 2026, driven primarily by AI. Agentic AI is generating a rapidly growing market for CPUs, in which Arm's energy-efficient architecture provides a structural advantage. The Meta agreement provides revenue stability, and support from over 50 ecosystem partners, including Nvidia, TSMC, AWS, and Google Cloud, including significant industry backing. Meanwhile, the licensing business continues to grow, driven by the Armv9 architecture's higher royalties and increasing data center adoption.

However, the risk side deserves equal attention. Arm has never manufactured its own processors, so scaling up an entirely new business line poses operational risks. The $15 billion revenue target is five years away and assumes that agentic AI develops at the pace the company predicts. There is a risk that existing licensees will eventually perceive Arm's new role as competitive rather than complementary, which could erode licensing revenue. Dependence on TSMC for all manufacturing is a vulnerability and at 81 times earnings, the valuation leaves no room for error. Any misstep in execution or lower-than-expected growth could trigger a sharp downward revaluation.