Arm leaders used a company event to outline a multi-business strategy built around higher-performance CPU platforms, expanded software support, and a new move into selling data center CPU chips, arguing that AI workloads are reshaping demand from edge devices through cloud infrastructure.
Edge AI: performance push and a growing role for CSS
Chris Bergey, executive vice president of Edge AI, said the Edge business unit—described as representing much of Arm’s legacy footprint—expects a roughly 40% increase in total addressable market (TAM) over the next five years, driven primarily by AI workloads moving onto more devices. Bergey said customer conversations at events including Mobile World Congress and Embedded World centered on a need for higher compute, more advanced process nodes, more powerful CPU complexes, and more memory bandwidth.
As evidence of traction, Bergey said that as of the quarter ending this month, Arm expects 25% of mobile royalties to come from CSS. He also pointed to new device categories where Arm expects additional content growth, including XR glasses and “personal AI computing,” citing NVIDIA’s “GB10” as an example of a CSS-based product category he described as in high demand.
On software, Bergey highlighted Armv9 adoption and AI-related instruction support. He said SME2 (Scalable Matrix Extensions) is shipping in leading handsets across iOS and Android, and that Arm is approaching about 50% value share on Armv9 penetration, projecting it will rise to 85% in the next two years. He also described KleidiAI, an Arm library aimed at helping AI workloads take advantage of Arm CPU instructions through integration into AI frameworks.
Physical AI: latency-driven systems and a long-range robotics thesis
Drew Henry, executive vice president of Physical AI, defined “physical AI” as AI embodied in machines that must “sense, decide, act safely” in the physical world, with latency from sensor input to actuation described as a key metric. Henry estimated the Physical AI TAM at about $25 billion per year today, growing to roughly $50 billion, driven less by unit growth and more by rising compute content in applications such as advanced automotive platforms.
Henry said Arm recently created the Physical AI business unit but has been in automotive and related markets for decades. He said Arm shipped more than 2 billion devices into the Physical AI space in the last 12 months through its ecosystem, reflecting how many compute elements are embedded across vehicle subsystems.
He also argued that the largest inflection in physical AI could occur after fiscal 2031, citing humanoid robotics as a potential “hockey stick” driver. Henry said he believes the market could reach $200 billion in TAM and suggested it could be larger, though he did not specify timing. He described humanoid robotics computing as exceptionally complex due to the need to manage multiple actuation systems alongside perception and interaction workloads.
Henry tied Arm’s business model evolution—moving from Armv8 to Armv9 and then to CSS—to increased value per design win. He said Arm has “doubled” royalty rates moving from Armv8 to Armv9 and described moving into CSS as another “doubling,” tying the increases to accelerated time-to-market and reduced engineering effort for customers.
Cloud AI: expanding from IP and CSS to Arm’s own CPU chip
Mohamed Awad, executive vice president of Cloud AI, said the data center CPU market is expected to exceed $100 billion by fiscal year ending 2031, and argued that accelerator growth is also increasing CPU demand as systems become more “agentic.” He described Arm’s go-to-market as spanning direct engagement with hyperscalers, channel-heavy engagement in enterprise via semiconductor partners, and a hybrid approach in wireless and edge (including telco infrastructure).
Awad said Arm has reduced customer barriers over time, first through Neoverse (launched in 2019) and then through CSS. He cited examples of CSS accelerating customer schedules, including a customer estimate of saving “about 80 man-years” of engineering effort and another customer going from receiving CSS to silicon running Linux in “less than 18 months,” including fab time.
Awad positioned Arm’s newly announced “AGI CPU” as a way to address customers that either do not build silicon or are not at sufficient scale to justify custom designs. He said Arm’s strategic advantage is being able to offer IP, CSS, or off-the-shelf silicon depending on customer needs, with a shared software foundation across those options.
Software: breadth, depth, and AI framework work
Shibani Roy, vice president of AI Services, said Arm has “over 2,100 teammates” focused on software and characterized software as essential to unlocking hardware value. She cited Arm’s ecosystem scale, including “22+ million developers,” more than “1,300 open source and upstream projects,” and “over 50,000 partners.”
Roy said Arm’s software efforts span firmware through operating systems to applications, and that the organization prioritizes three KPIs: performance per watt, ease of development on Arm, and speed of delivering innovation to market. As an example of partner work, she described collaboration with Meta related to PyTorch and ExecuTorch, including Arm’s contributions intended to help deploy models more efficiently on Arm-based devices.
Financial model: chip revenue ramp alongside IP growth
Jason Child, Arm’s CFO, said the company’s decision to introduce the Arm AGI CPU followed three years of exploring the chip market in response to customer demand. He argued that Arm’s chip business is intended to compound rather than displace the existing IP model, and said the combined approach could materially expand revenue, profit, and EPS potential by fiscal 2031.
Child sized Arm’s semiconductor logic opportunity (CPUs and “XPUs,” excluding memory and optical) at over $500 billion today, growing to more than $1.5 trillion in fiscal 2031. He broke the opportunity into Cloud AI, Edge AI, and Physical AI, with Cloud AI described as a $330 billion market today growing at over 30% annually to around $1.2 trillion in fiscal 2031. Edge AI was described as $180 billion today, growing to $250 billion over the next five years, and Physical AI as $25 billion today, doubling over the same period.
Within Cloud AI, Child said the company is focusing on a non-accelerator CPU TAM he described as $455 billion, including “$100 billion plus” in the data center CPU TAM addressed by the AGI CPU. He said Arm believes inference and agentic AI will increase CPU performance needs and drive volume and ASP increases, adding that Arm has “visibility into our customers’ roadmaps.”
Child outlined how Arm views value capture across models, using an illustrative $1,000 chip example:
- CPU IP: about a 5% royalty rate (about $50 royalty/gross profit dollars per $1,000 of chip sales).
- Compute subsystems (CSS): about twice the royalty rate (about $100 royalty/gross profit dollars per $1,000 of chip sales).
- Chips: about $500 of gross profit dollars per $1,000 of chip revenue in the simplified example.
He said the company has “line of sight to more than $1 billion in chip demand over the next 2 years,” with the “vast majority” expected in fiscal 2028, and projected “material revenue” from AGI CPU starting in fiscal 2028 with a ramp to around $15 billion in fiscal 2031. Child said the company’s “biggest challenge” is not customer demand but “memory shortages limiting our customers’ ability to deploy our chips.”
On the IP side, Child said Arm’s royalty revenue has grown about 14% CAGR over the past five years and over 20% in the past two years as Armv9 and CSS have begun to ramp. He projected royalty revenue CAGR of 20% over the next five years, while cautioning that annual results may vary due to market downticks or inventory corrections. He also said Arm is seeing core counts per chip increase by about 20% per year based on customer history and plans.
Child said Arm has high contract visibility for royalties, stating that for fiscal years 2027–2031, 70% of forecasted revenues are already covered with royalty rates set in contract, and that even by fiscal 2031, the contracted base is around 60%. He also provided segment-specific contract coverage figures, including 85% of expected cloud royalties under contract over the next five years and 95% of Physical AI royalties under contract through fiscal 2031.
For licensing revenue, Child said growth has accelerated beyond what Arm outlined at its IPO, citing recent growth “over 20% per year.” He attributed drivers to the AI cycle, subscription licenses, compute subsystem agreements, and expansion of license and design service agreements with SoftBank, and projected SoftBank licensing growth around high single digits.
Looking to fiscal 2031, Child said Arm expects the IP business to reach about $10 billion of revenue with a 99% gross margin and over 65% non-GAAP operating margin, raising the company’s long-term operating margin target by 500 basis points from 60%. He said the AGI CPU chip business is expected to reach about $15 billion of revenue with gross margin of at least 50% and non-GAAP operating margin over 30%. Combined, he projected $25 billion of revenue and “more than $9” of non-GAAP EPS in fiscal 2031, and said Arm is affirming its previously issued Q4 guidance.
In Q&A, executives said they have committed customers through the first two generations of AGI CPU and are in definition for the third generation. They also said the SoftBank work referenced in prior periods is separate from the AGI CPU initiative. On competitive dynamics, Awad said the CPU market is large enough for multiple suppliers and emphasized that AGI CPU targets workloads such as orchestration and distribution in AI systems, distinct from accelerators that generate tokens. Arm leaders also said they see no issue selling AGI CPU into China from an export-control standpoint based on current rules, though they did not announce China customers for the product at the event.
About ARM (NASDAQ:ARM)
Arm Limited (NASDAQ: ARM) is a global semiconductor IP company best known for designing energy-efficient processor architectures and related technologies that underpin a wide range of computing devices. Founded in 1990 as a joint venture between Acorn Computers, Apple and VLSI Technology and headquartered in Cambridge, England, Arm develops the ARM instruction set architectures and core processor designs that chipmakers license and integrate into custom system-on-chip (SoC) products. The company operates a licensing and royalty business model rather than manufacturing chips itself.
Arm’s product portfolio includes CPU core families (such as Cortex and Neoverse lines), GPU and multimedia IP (Mali), neural processing units (Ethos) and a suite of system and physical IP blocks.
