Deep Dive: Arm Ltd
Can the global powerhouse in chip design live up to its hype?
Arm is a semiconductor IP company with an extensive influence in global chip design. Arm’s designs have long been the global standard for mobile devices. Every time you pick up your smartphone or use a connected device, there's a good chance you're engaging with Arm’s IP. It's estimated that ~70% of the world’s population interacts with Arm technology on a daily basis.
Arm has accelerated its growth investments since Softbank took ownership of the company in 2016. These efforts are now bearing fruit, with Arm gaining share in key sectors like cloud, automotive, and IoT.
As value investors, we don’t often participate in IPOs. Instead, we view them as an opportunity to study a business deeply and to build conviction, so that, when the time is right, we can act decisively. The semiconductor industry is always highly dynamic, and through all of this, we believe Arm is on the right side of change. In fact, we believe it is one of the highest quality companies in the world.
With Arm’s impending IPO this year, our goal here is to help investors stay informed and ahead.
This report delves into:
Arm’s origins and its significance in the semiconductor industry
The company’s evolution under Softbank
Arm’s licensing and royalty structure
The competitive environment, highlighting Arm’s advantages
Arm’s positioning and growth in new segments, including cloud and autos
Arm’s weaknesses, challenges, and risks
Thoughts on Arm’s valuation and where we would buy the stock
Our insights are sourced from interviews with experts, including former Arm employees, competitors, and other industry figures. We believe these discussions have yielded a fresh, in-depth perspective on Arm.
Arm’s origins
Arm was founded in November 1990 as Advanced RISC Machines Ltd. Arm began as an in-house project within a UK company called Acorn Computers to design chips utilizing a particular architecture known as RISC (Reduced Instruction Set Computing).
Acorn is not around anymore, but back in the day, it was best known for a computer called BBC Micro (below image, left) which found extensive use in the UK public schools during the 1980’s. To serve the needs of schools, Acorn aimed to build computers that were small, cheap and power efficient. The RISC chips were optimal for these criteria.
Acorn teamed up with a company in Silicon Valley called VLSI Technology, and later with Apple, to further develop RISC technology. At the time, Apple was working on its handheld computer Newton - an early form of PDA and a precursor to modern tablets (above image, right). In search of a low-power, cost-effective chip, Apple became interested in the RISC design. However, Apple was wary about relying on a competing firm for technology. Hence, the three companies - Acorn, VLSI, and Apple - agreed to establish Arm as a separate entity, to be carved out of Acorn, and jointly owned by the three parties.
Arm became a listed company in 1998 on the London Stock Exchange. It enjoyed modest success in the handset market, which was still nascent at the time, but struggled to gain traction in the much larger PC market which Intel came to dominate. Arm was a niche player in the early years of its existence.
The company’s trajectory shifted dramatically with the rise of smartphones. The release of the iPhone in 2007, which utilized Arm-based processors, kickstarted a new era. By the 2010’s Arm had essentially monopolized the smartphone market.
In 2016, Softbank bought out Arm for £24 bn (or $32 bn) and took it private.
Arm’s role in the industry
Arm is not a chip manufacturer. It creates intellectual property (IP) in the form of technology designs, which are licensed to chip designers.
One way to understand Arm's contribution to the semiconductor industry is through its role in standardization. Its architecture serves a role akin to being a common language in a device ecosystem. A key insight on the industry is that when the amount of software written for a device starts to meaningfully increase, the industry has the tendency to standardize around a common architecture. Standardization promotes efficiency in software development and allows for faster innovation.
“With the complexity of CPU design increasing exponentially, over the past decade no company has successfully designed a modern CPU from scratch.”
- Arm
Similar to chip designers like Nvidia and Qualcomm, Arm operates as a fabless, R&D-focused organization. However, unlike these firms which do end-product designs, Arm's focus is on creating designs that can be considered the 'highest common denominator' among various design companies. It is about providing customizability while ensuring standardization for efficiency and time-to-market in the industry.
As mentioned before, Arm designs are based on specific type of processor architecture known as RISC (Reduced Instruction Set Computing). As the name suggests, it’s called “reduced” because the processor uses a smaller, simpler set of instructions. This is in contrast to CISC (Complex Instruction Set Computing), which is primarily represented by the X86 architecture utilized by Intel and AMD.
The choice involves trade-offs. RISC architecture excels in terms of energy efficiency and cost-effectiveness. This makes them particularly suited to mobile applications and embedded systems - dedicated computing setups integrated within larger devices to carry out specific functions (e.g. automotive systems like ECUs and navigation). In contrast, CISC architectures are better suited for high-performance applications where the emphasis is more on raw power.
Historically, Arm and X86 catered to different end-markets: X86 for PC and servers, and Arm for mobile and connected devices. However, that boundary has started to blur recently, with Arm waging its war on X86’s traditional strongholds, as highlighted by Apple’s transition from X86 to Arm processors in its MacBook lineup, and Arm’s wins in server chips including Amazon’s Graviton, Nvidia’s Grace, and Ampere’s offerings which are all built on Arm architecture.
While Arm was initially perceived as underpowered, its capabilities have improved a lot over the years. This is due to innovations like multi-core processing (which we'll discuss later). An industry expert provided a rough estimation: approximately 70-80% of computing tasks can now be managed with Arm cores. The remaining 20-30% may demand the higher throughput of a X86, but this added power also comes at a significant additional cost.
There will always be applications where that extra throughput will be desired, regardless of the cost. An example of such would be when designing a high-frequency trading system, as even the slightest deficit in performance can lead to big consequences on trading profit.
Revenue model: licensing and royalties
Arm’s revenue has two main components:
A one-time, upfront licensing fee which the partner pays to gain access to Arm’s designs. The partner then works on product development and testing, over a period of, say 3-5 years, before production commences.
Arm earns a royalty income for every chip shipped by its partners. Historically, royalty rates started at about 1% for low-end products like microcontrollers and smart cards, and went to about 2-3% for smartphone processors. In 2022, royalty payments from partners accounted for 63% of Arm’s revenue.
The chart below shows the breakdown of Arm’s royalty revenue into vintages. ~46% of Arm’s royalty revenue came from products released between 1990 to 2012. Arm is still making royalty from some designs struck more than 25 years ago! In this way, the business is highly cumulative in nature.
As of December 2022, Arm estimated that the aggregate value of chips containing its technology was ~$98.9 bn. On this base, Arm generated ~$1.78 bn in royalty revenue, representing an average royalty rate of 1.7-1.8%.
We can further breakdown the type of license that Arm offers: technology and architectural.
Technology licenses refer to Arm-designed cores, like those in the Cortex-A series or M-series (whenever you see the word ‘Cortex’ it’s referring to technology license). Essentially, customers are outsourcing the processor R&D to Arm, and are buying off-the-shelf core designs which they can readily incorporate into their own processors or SoC (System on Chip) designs. This saves customers from having to design a CPU from scratch.
For clarity, a “core” is an individual processing unit. A bunch of cores constitute a CPU, which in turn is a key component of an SoC (system on a chip). Arm’s core designs are its flagship offerings. But the company also provides more complete CPU subsystems which enables customers to outsource an even larger portion of their CPU design work to Arm.
On top of this, Arm has an array of other IPs, which customers can combine into their SoC designs. These includes GPUs, memory controllers, interconnects, physical IP, and machine learning solutions. While these are not nearly as ubiquitous as Arm’s core designs (faces a lot more competition in each), they allow Arm to offer a complete menu to customers.
In contrast to technological licenses, architectural licenses are more basic. Customers are only purchasing the rights to the most fundamental building blocks, called Instruction Set Architecture (ISA). Instead of relying on Arm's off-the-shelf core designs, companies with an architectural license opt to design their own unique cores in-house. Apple is one of the most notable architectural licensees, which it uses to build the M1 and M2 processors for its MacBooks.
This distinction is important because it directly relates to the amount of royalty income that Arm makes. As you can guess, when Arm provides a more ‘complete’ product with technology licensing the royalty is higher. A simple analogy here is that a coffee shop would much rather sell you their cinnamon-spiced caramel lattes at double the price instead of just a shot of espresso.
Now, imagine if a bunch of regular customers at the café got together and said: “let’s just buy the shot of espresso and make our own espresso drinks at home”. As you can guess, the café owner would surely not be pleased. We’ll leave it at this for now, but come back to this important issue later on.
Enter Softbank
Softbank paid $32 bn to acquire Arm in 2016 and took it private. Its financial performance since then is shown below.
What was Softbank’s playbook?
First, it ramped up Arm’s engineering resources. Prior to Softbank’s ownership, Arm was making roughly 50% in adjusted EBITDA margin (40% IFRS operating margin). The company was enjoying extremely strong levels of profit, but Masa believed the company wasn’t investing enough and lacked the next pillar of growth beyond mobile devices. To make the kind of investments that he thought were necessary, Masa took the company private.
Has the effort borne fruit? We believe the answer is a yes. Arm’s market share gains since Softbank took ownership is shown below (company’s own calculations). Arm has made significant inroads in strategic markets like cloud and automotive, and these have developed into solid second and third pillars of Arm’s business, after mobile. Based on conversation with a former Arm employee, it was revealed that mobile now constitutes around 60% of Arm’s revenue, while cloud and automotive/IoT each represents about 20%.
In its latest F-1 filing for the IPO, Arm provided updated market shares: As of December 2022, Arm’s market share was 99% in mobile application processors, 64% in IoT/embedded, 25.5% in networking equipment, 40.8% in automotive, and 10.1% in cloud compute. Again, these are the company’s own estimates, but our conversations with former employees and competitors broadly corroborate the rising share trend.
More recently, Softbank seems to have moved onto a more value extractive phase. Masa believes Arm has not extracted enough value from its IPs and Softbank seems to be testing its market power.
We don’t have the information on royalty rates by product, but one source (former employee) has mentioned that Arm has raised its royalty rates to double or triple the level in certain products compared to ten years ago. In 2021, we saw a sharp increase in Arm’s revenue which coincided with the introduction of ARMv9, Arm’s latest ISA release. Softbank used this opportunity to raise the royalty rate.
An FT article has reported on Arm apparently looking to change its royalty model from one based on chip value to one based on device value:
“The [royalty] amount will be at least several times higher than what Arm gets now,” said an executive from a leading Chinese smartphone maker which has so far refused to back the proposed plan. “We are told that they hope such changes could start from 2024.”
It’s our understanding that the royalty hikes, and discussions around any proposed ones in the future, have caught customers off guard, resulting in some upset customers. Arm’s justification through all of this is that the firm has ramped up its investments, allowing Arm to deliver more sophisticated products than ever before. Arm’s sales teams have been trying to convince customers of the added value of new products, such as highlighting new security and AI features for the ARMv9.
Competitive moat
At the turn of the century, the global processor landscape had three major players: Intel, MIPS, and Arm. Intel’s domain was predominantly in PCs. MIPS had a strong foothold in high-end computing applications like supercomputers and graphics systems. Arm saw opportunity in the mobile market which was nascent at the time. The one observation to make is that these companies tended to dominate, if not entirely monopolize, their respective spaces.
“Intel has dominated the Windows space. It’s really hard for somebody to break in because of software. MIPS, despite dying for decades, has dominated the router and set-top box space, again because of software. Arm sits on pretty much 100% monopoly when it comes to mobile, again, because of software.”
– Former Director, Strategic Business at Arm (May 2021, Stream Transcript)
Arm’s primary competitive advantage is its robust software ecosystem. This comprises of several interconnected layers, such as the OS, applications, middleware, libraries, and software tools, which are optimized to work with Arm’s architecture. For software developers, the cost of supporting multiple architectures is substantial, involving not just the upfront development work but also the ongoing burden of maintenance and updates for every architecture version. Developers are naturally inclined to focus on architectures that promise the broadest reach and highest return on investment. A new architecture therefore must demonstrate clear path towards widespread adoption, before the developer community will consider supporting it - chicken and egg!
While not impossible, transitioning between architectures is a massive undertaking. Typically only the largest players with the necessary resources, influence, and vertical integration can pull off. In this context Apple serves as a notable exception - their shift from X86 to Arm-based processors for the Mac lineup is a good example. Apple not only re-engineered the MacOS and its software, but also provided tools and incentives to encourage third-party developers to recompile their apps for the new architecture.
The moats in this business are typically so strong that companies like Arm and Intel end up as monopolies. However, ironically, such monopolization often sows the seeds of their own potential downfall. In this sense, success can be its own undoing. Consider the data center market, where Intel has historically held sway and wielded significant influence - this is where Arm has been enjoying a great deal of support from customers.
“Arm has been trying to break into that data center and cloud computing business for a long time. You can see maybe 10, 15 years Arm start trying that. First maybe 10 years was we wasn't get any traction because still Intel, AMD is dominating that space with x86. However, Intel has been obviously very successful and they charge very high price for the processors. Many of the Intel's customer actually, they hate Intel. They need technology but the cost is very high. If you don't have meaningful competition Intel can charge you whatever they want to charge you. With that, Arm started gradually breaking into this space.”
Similar forces also poses threat to Arm, where it has monopoly.
RISC-V (pronounced ‘RISC-five’) is an open-source RISC architecture. Unlike proprietary architecture like Arm, RISC-V has no licensing fees. Having been developed by researchers at UC Berkeley in 2010, it’s still a relative newcomer but has been gaining traction and supported by firms that want to avoid the high cost and lock-ins of a proprietary architecture. RISC-V seems to be doing well on the low-end side, mainly IoT applications, but has not yet made much of a dent in the smartphone market.
“Arm indeed and does see RISC-V as a threat if it’s low-end. However, at least a couple of years ago, Arm had not seen any account or business winning by RISC-V. Basically, it’s still not a face-to-face, very fierce competition yet…I would say RISC-V is good. They’ve got a promising future but still has a long way to go to really challenge Arm, especially in the mid to high-end”
It’s worth mentioning however that just because RISC-V doesn’t have licensing fee or royalty, this doesn’t make it “free”. As one industry insider explained, nothing is free in the industry! There is a lot of verification and performance optimization process involved in the adoption of open-source, which can take up significant engineering resources. Many players have been experimenting with RISC-V, although we’ve seen limited movements by Arm’s major customers when it comes to actual adoption.
Having said this, RISC-V is gaining support from many ventures and startups that possess limited funding. And Arm has definitely felt this competitive pressure, which is why it decided to remove its upfront licensing fee on the Cortex-M0 and Cortex-M3 processors in 2017. These M-series are low-power applications used in IoT applications (wearables and home electronics and appliances), which has been feeling the most competitive pressure from RISC-V.
Under the new model for these chips, Arm charges no licensing fee, but the royalty on production has been jacked up substantially. The typical royalty for Cortex-M0 was 1% under the standard licensing agreement, but under the new model, it has been increased to 4%. Therefore, while this new model might seem more friendly for startups and ventures and allows them to do free design and prototyping with Arm designs, it is by no means a pricing reduction given the much higher royalty rate.
Cortex M0 and M3 were offered under what was at the time called DesignStart program. This was rolled into a broader Flexible Access program in 2021, which broadened the scope of the affordability program. Now, entry level customers have access to either no-cost or low-cost licensing to over 80 of Arm’s products. Flexible Access has been growing its licensees at a rapid rate, with 203 licensees for the fiscal year ended March 2023, up from 129 in 2021.
In essence, while open source is a potential threat, Arm is well aware and taking the necessary steps to thwart its success while, at the same time, maximizing economics on the lower end products.
Growth in strategic end-markets
The levers to growth are: mobile, automotive, cloud, IoT. If Arm can sustain a growth rate in the 10-20% range in its new, strategic end-markets of cloud and automotive/IoT, then overall a 5-10% growth on a consolidated basis should be very doable. With handset devices, where volume growth has been flat and Arm has already saturated its market share, growth will largely depend on pricing. While we do think there is a chance that Arm has more pricing levers to pull, we do not factor this into our base case assumption.
Cloud
Currently, the most significant growth driver in the business is cloud.
Historically, x86 architectures had a monopoly in datacenters. Within this, Intel has a monopoly position with AMD in a distant second position. However, as the industry transitioned from on-premise to cloud services, Arm found itself increasingly relevant in this market.
First, hyperscalers are in the business of selling compute, and efficiency is paramount. They needed to provide services at a reduced energy consumption and cost, and Arm's design is suited for this. According to Softbank, between 2010 to 2040, electricity usage by cloud computing is expected to increase 365-fold! Second, there's a growing trend of hyperscalers developing custom chips to cater to their unique needs. As x86 does not license out its architecture, Arm saw the business opportunity and capitalized on the gap in the market.
As a result, Arm designs have been gaining share in cloud computing. This was through either partner companies supplying the hyperscalers (e.g. Ampere providing Arm-based processors to Microsoft Azure) or hyperscalers directly engaging Arm through their in-house chip design (e.g. Amazon and Alibaba). In both cases, Arm provided customers a greater degree of flexibility and choice beyond the effective Intel monopoly.
Industry experts point out that the cost advantages of Arm chips is significant. According to former Arm employee, AWS's Arm-based server costs are approximately a third less than those based on x86 architecture (side note: Amazon’s own claim is 40% better price performance against comparable x86). Another industry expert here talks about how the entire industry “‘will be forced to provide an answer to cheaper, more economical compute”:
“If you look at Broadcom’s acquisition of VMware, Broadcom is a large producer specifically for ARM-based computing. You can see the logical path where they’re going with their product roadmap. Given VMware owns the lion’s share of the market for virtualization, you can see that’s just inevitability of where product is going… The compute power and cost efficiencies that you’ll be able to avail yourself to be just undeniable…once it’s in the virtualization layer, I think the competition, if they haven’t answered that already, will be forced to provide an answer to cheaper, more economical compute”
With its roots in low-power devices, for a long time Arm was regarded by the industry as being inadequate for higher performance applications. But over time, Arm recognized the industry’s need for higher throughput and has started to balance its efficiency with added performance improvements.
ARM v7, introduced in 2011, marked the start of multi-core processing for Arm. And under Softbank, they have doubled down on this high core count strategy. This is possible with Arm, because while the performance per core of Arm chips may never be as good as x86, Arm’s cores are much more power efficient that they can afford to economically “cram” more cores than x86. For instance, Arm-based server chips had achieved core counts of up to 144 (Nvidia Grace), with 192 (Ampere Siryn) and 256 (Ampere Polaris) core count now in the roadmap (see below).
Now, this doesn’t mean cramming more and more cores solves all problems, as there are applications where high single-core performance is more important than having many cores. Having said this, hyperscalers have been gravitating towards the benefits of high core counts, finding them particularly aligned with their needs, as the expert below (who is interestingly, an ex-Intel employee) explains:
“One is the hyperscalers, obviously, that are providing data center compute capability at scale. That is all about having more cores and being able to deploy enough workloads on different cores so that the business model of compute for dollars continues and they can provide that at scale.”
We’ve mentioned earlier how software ecosystem serves as a moat in this business, favoring the incumbents. In the server market, much of the software had been written exclusively for the x86. This is the reason why it took Arm as long as it did to gain traction in the industry - despite having entered the market more than a decade ago (and throughout all this time, customer discontent towards Intel which many have labelled as an ‘arrogant bully’!)
Let’s also not forget that Arm had reportedly dedicated a 200-engineer team working with partners like AWS to support the migration, such as software porting. In addition, the partners themselves have an internal team likely much larger than the Arm team. Challenging the x86 stronghold has certainly been no small feat.
But Arm has come a long way, and the following expert thinks the industry may be on the brink of a significant shift.