For those of us who have covered Arm for decades, we remember the days when this was a $500-600M yearly revenue company, and hardly growing. How times have changed.

The evolution of Arm, now clearly framed as Arm 2.0, has been years in the making, with all the stakeholders recognizing that an IP business alone is likely to hit a revenue ceiling. Arm, fully aware of that reality, is expanding thier business model to merchant silicon with the launch of the Arm AGI CPU. The way to understand Arm’s launch of its own AGI CPU is to think about what the unit of monetization in AI infrastructure has become. For years, Arm got paid when customers used its architecture or its cores. In the AI era, the relevant product is increasingly the full compute system, and in the data center, the full orchestration stack that keeps accelerators utilized. As we detailed yesterday, in our agentic CPU report, AI changes the role of the CPU. The CPU is no longer a background component attached to accelerator clusters. It is becoming the control plane for increasingly asynchronous, tool-driven, multi-step workloads, and Arm’s own analysis suggests the required CPU footprint per gigawatt can rise from roughly 30 million cores today to roughly 120 million cores as agentic AI scales. Staying only at the ISA or core-IP layer would have left Arm exposed to the largest growth pool while others captured the actual silicon dollars.

Arm spent the last several years moving up the value chain in stages, first from ISA to cores, then from cores to Compute Subsystems, or CSS, which packages more of the design work and allows Arm to charge materially higher royalty rates. What CSS proved is: it lowered the barrier to entry for customers, saved roughly 80 person-years of engineering effort in one documented case, and got another customer from handoff to silicon booting Linux in under 18 months. It demonstrated that Arm could do more of the design work, capture more value, and make customers faster simultaneously. Merchant silicon is the logical extension of that model. When the final product becomes the data center, the economic logic pushes toward selling chips, because that is where the value is increasingly being created and priced.

The Market Opportunity
Datacenter infrastructure topology is changing in real time as AI workloads shift from training toward inference. To date, most datacenters have been built around two workload classes. The first is traditional cloud software: SaaS applications, web services, and databases. The CPUs serving those workloads sit in what we have long described as the cloud-native bucket. Arm’s customers, with NVIDIA as the main exception, have largely been building those cloud-native CPUs themselves using Neoverse and compute subsystems. The second major datacenter build-out has been for training-oriented AI workloads. Inference workloads are now beginning to reshape infrastructure decisions, and that should open a new product cycle across both CPUs and accelerators in the years ahead.

The opportunity Arm has identified is a new tier of infrastructure: dedicated CPU racks purpose-built for agentic inference. The easy comparison is to frame Arm’s AGI CPU against NVIDIA’s Grace and Vera CPUs, but we think that misses the more important distinction. NVIDIA’s ambition appears to be tightly coupling its CPU strategy to its own accelerators. Until proven otherwise, we do not think NVIDIA wants to sell Vera CPU racks broadly into competing accelerator environments such as TPU-based systems. That leaves an opening for Arm. AGI CPU racks can fill a gap that is both emerging and growing quickly, particularly for customers who want a high-performance, power-efficient control-plane CPU without being tied to a specific accelerator stack. More importantly, this is best understood not as a new workload tier for Arm, but as a new economic model, one that lets the company move beyond IP and CSS and capture a much larger share of the datacenter value pool.

The Economic Opportunity
Arm spent a significant amount of time walking investors through the economics of this opportunity, and the message was straightforward. On an illustrative $1,000 chip, a traditional IP licensing model at roughly a 5% royalty rate yields about $50 of revenue and gross profit dollars. A CSS model, at roughly double the royalty rate, yields about $100. Selling the full chip yields roughly $1,000 of revenue and about $500 of gross profit dollars. The gross margin percentage is lower, but the gross profit pool is dramatically larger. Management also framed the ceiling problem directly: in a simplified fully penetrated market scenario, the CSS model caps Arm at roughly $2.4 billion of revenue, while supplying the full chip expands that figure to roughly $24 billion. That is the core of the thesis. Merchant silicon is about escaping the revenue ceiling that licensing alone imposes, not about walking away from a high-margin business.

Importantly, the IP business continues to compound underneath that new chip opportunity. Management is guiding to roughly 20% CAGR in royalty revenue over the next five years, with about 70% of forecast revenues for FY27 through FY31 already covered by contract, cloud AI royalties about 85% covered, and physical AI royalties about 95% covered. On top of that base, Arm expects the AGI CPU business to contribute materially starting in FY28, with line of sight to more than $1 billion of chip demand over the next two years and a path to roughly $15 billion of chip revenue by FY31. The FY31 framework management laid out is roughly $10 billion of IP revenue, roughly $15 billion of chip revenue, more than $25 billion of total revenue, and more than $9 of non-GAAP EPS. The key point is that chips compound on top of IP rather than replace it.

One additional point management made, and one investors should not lose sight of, is that much of the investment required to support this expansion is already in the model. That means the incremental gross profit from chips should carry meaningful drop-through to earnings over time. In that sense, this is not only a larger revenue opportunity. It is a material change in Arm’s long-term earnings power.

There is a second angle here that should be part of any serious view of Arm’s upside. Arm is now selling commercial flexibility on top of a common software base. A hyperscaler can license IP for one workload, use CSS for another, and buy off-the-shelf Arm silicon for a third, all while staying on the same Arm Neoverse software foundation. Management was explicit that this is one of Arm’s biggest strategic advantages, and we think that point has real merit. Arm is turning software continuity into commercial leverage across more points in the infrastructure stack. Merchant silicon also reaches customers the legacy model could not fully serve. IP and CSS primarily addressed the largest builders with the scale and engineering depth to design their own silicon. The chip extends Arm into enterprises, telcos, neo-clouds, and appliance vendors that want Arm economics but cannot justify building a CPU internally. Management also noted that well over 10,000 customers already use Arm in the cloud today, which creates a bridge from existing cloud adoption to on-prem deployment. That installed base is likely one of the most important demand vectors in the launch. Arm’s model now spans revenue pools where customers can either build with Arm or buy directly from Arm.

The broader strategic read is both offensive and defensive. Offensively, Arm now participates in a much larger share of the value being created by agentic AI. Defensively, it keeps itself relevant as infrastructure becomes more system-oriented and more vertically integrated. The implication for the investment debate is straightforward: the IP and chip businesses have to be modeled together. The ceiling on IP growth is what makes the chip business necessary, while the visibility and contractual coverage in the royalty model are what help de-risk the chip ramp. That is also why this should be viewed as an expansion of Arm’s earnings architecture, not just a new product launch.

Join the newsletter and stay up to date