ARM IPO Breakdown
The Long Awaited Softbank IPO looks expensive. Here's my thoughts,
ARM prices this Wednesday and trades on Thursday. I have a lot of thoughts. This write-up will be structured with a pretty significant paywall. Below is my outline. This will be a great post if you want to learn about ARM. For details about the stock, that will be behind the paywall.
ARM History and Overview
ISAs and What ARM Does
IP Industry Overview
Competitive Analysis (Paywall)
Business Overview (Paywall)
Valuation (Paywall)
IPO Tactics (Paywall)
So, without further adieu, let’s dive into the most important IP company in the world.
A Brief History of ARM
ARM started as a joint venture between Acorn Computers, Apple, and VLSI Technology. Apple invested cash, VLSI provided tools, and Acorn provided engineers. With this, ARM was born. ARM’s humble beginnings start in a barn.
ARM’s big break came from Texas Instruments in 1993, when they finally managed to license their RISC-based ISA to large semiconductor manufacturers. Other customers, such as Nokia and Samsung, joined, and ARM managed to ride the mobile wave quickly. Mobile companies adopted ARM because of its power efficiency and reduced instruction set.
ARM went public in 1997, and shares soared along with the general ebullience in the tech market. ARM hit its stride in the mobile market in the early 2000s and began to ship billions of units in 2004. This coincided with the rise of the mobile phone, and by the time Apple launched its iPhone, ARM was firmly in the incumbency position in the mobile IP market.
ARM would quietly chug along for a few years until Softbank took them private in a $32bn deal in 2016. Softbank would invest heavily in ARM, letting operating profit drop to reinvest into the business. In its form F-1, they call this their strategic evolution and list some key elements in their changing business.
Creating a Market-Led Business with Products Optimized for Specific Verticals (Specialized Compute IP)
Building the New Armv9 Architecture
Introducing New Licensing Models with an Increased Focus on Maximizing Royalty Opportunities (Increasing types of License types)
Interestingly, while creating a meaningful amount of progress, most of these initiatives all dwarf compared to its mobile platform. ARM CPUs are finally starting to be adopted in the data center from Amazon’s Graviton and Ampere. But don’t kid yourself; most of ARM's chips are mobile-related, with a few IoT platforms and the data center. Below is their cohort graph of products by year.
Softbank heavily invested in the business, and while data center adoption has increased, meaningful revenue acceleration at ARM hasn’t happened. ARM is the most important ISA globally, but a core push and pull exists. They need to price their IP at a price low enough to lower friction and reduce the likelihood that customers choose another product but also raise prices enough to have a viable business model.
Besides the Softbank buyout and investment period, a few other important events happened: ARM China going rogue and Nvidia’s failed buyout. Let’s talk about those now.
ARM China, like many foreign companies, is structured as a locally owned business that does business with ARM Holdings at the global level. Founded in 2018, ARM China is the exclusive platform in China and sells its designs. ARM China’s independence and ownership are strategic for China’s semiconductor independence, so it makes sense that the government would demand this structure.
There was the matter of the rogue CEO. Allen Wu, the CEO of ARM China, went rogue and created a veto vote to the board that gave him defacto control. He created a competing subsidiary, Alphatecture, to compete with current ARM China customers. ARM, the parent company, clearly didn’t like this, and Softbank tried to fire Allen Wu in 2020. Wu didn’t attend the meetings or hand over the official seal and important control aspects.
This continued for two years. Wu played chicken with ARM, and while ARM China couldn’t license new IP, they could continue to do business with their existing IP rights. Wu hired a security team at ARM China’s Shanghai office to continue his coup. Wu responded against the eventual push for him to step down by the Shenzhen regulator. He had to eventually be kicked from the internal IT systems of the company in May 2022. On his last day, he emailed employees not to listen to new managers until Wu's lawsuit against ARM China was resolved.
In the end, he was kicked out, and ARM China finally wrestled control back from the rogue entity. It’s been a bit of a saga. Oh, and did I mention that Nvidia tried to buy them simultaneously?
Concurrent with the rogue subsidiary, Nvidia tried to buy ARM. In September 2020, Nvidia announced its intention to buy ARM for 40 billion dollars. At the time, this looked like a mediocre deal for Nvidia, but a great deal regarding strategic relevance. Jensen would effectively wrest control of the most important ISA in the world and force the industry to conform to his roadmap.
The entire acquisition attempt was plagued by regulatory scrutiny, and eventually, in December 2021, the FTC blocked the acquisition. ARM was effectively frozen the entire time, and pay raises, hiring, and other important aspects were put into limbo for over a year as the deal stalled. The deal was officially scrapped in February of 2022.
Softbank has wanted to offload the global IP company, given that they were willing to sell the business at ~40 billion dollars in primarily stock to Nvidia (which would have been a home run). Now, that brings us to today, the IPO. ARM is attempting to list between 50-55 billion.
What is the ARM ISA, and What are ISAs?
I mentioned ARM’s RISC-based ISA, but I want to explain what an ISA even is and how ARM’s differs. ISA stands for Instruction Set Architecture and is a set of rules and instructions that interface the hardware and software. A semiconductor without an ISA is just a piece of fancy silicon. The ISA is the interface between the code and the semiconductor.
A simple analogy that fits is that the semiconductor is the car, the software is the driver, and the ISA is the pedals and user interface between the two. Without a steering wheel, the software cannot drive the car.
Now, there are different types of ISAs. There have been several eras of ISAs, but ARM is now the most popular ISA.
1964: IBM System/360 - The first modern instruction set. This greatly influenced all subsequent ISAs.
1970s: CISC (Complex Instruction Set Computer) - This architecture had many different instructions, making programming easier and increasing complexity. x86 is a CISC ISA.
Late 1970s: RISC (Reduced Instruction Set Computer) - IBM researchers developed RISC as a response to the complexity of CISC, focusing on a smaller set of instructions for better performance and efficiency.
1987: RISC-based processors like MIPS, SPARC, HP-PA, IBM RS/6000, and PowerPC emerged, offering improved performance and energy efficiency
1990s: ARM architecture gained popularity, becoming the dominant ISA for mobile devices and embedded systems due to its low power consumption and energy efficiency.
2010: RISC-V was developed as an open-source ISA based on RISC principles at the University of California, Berkeley. It has since gained significant traction in the industry, with many companies and organizations adopting it for their products and projects
Now, notice in this timeline that ISAs take a long time to become adopted and even longer to die. x86, for example, is still a dominant ISA today. ARM’s greatest feature is that it is more power efficient, but the previously touted over 20% performance improvement is likely more like only a 7-10% improvement over ARM. And x86, to a certain extent, is still the highest-performant CPU ISA today.
ARM is the ISA that emerged and rode the crucial wave of Mobile. Today, another new ISA is gaining in popularity, RISC-V, an open-source ISA that is gaining in adoption as a reaction to ARM. I’ll write more about that in the competitive positioning segment.
ARM doesn’t just make an ISA. Making a steering wheel is a very low-value industry; instead, it makes a whole solution for controlling the semiconductor. ARM has two ways to make revenue: from licensing and royalties.
ARM charges customers money to license their products, and that means using the cores. Additionally, when the chip hits production, the customer pays ARM a small percentage of the cost over time. According to their F1 filing, this is about 5 cents a chip on average. Higher-end chips cost more, and their licensing costs easily millions to larger customers.
ARM has a few families of products. Here’s a screenshot of their entire product catalog on their website. For simplicity, I will mostly cover their bigger cortex products and Neoverse. Each product can be used off the shelf to be designed into larger semiconductors or electronic devices.
Cortex-X: The Cortex-X Custom Program is a high-end performance CPU. This is likely targeted at mobile computing, and Samsung and other customers use this for their high-end phones.
Cortex-A: consists of 32-bit and 64-bit processors used for most devices, including Cortex cores used in most mobile devices.
Cortex-R: A family of 32-bit and 64-bit RISC ARM processor cores optimized for hard real-time and safety-critical applications
Cortex-M: A group of 32-bit RISC ARM processor cores optimized for low-cost and energy-efficient integrated circuits, commonly used in microcontrollers and embedded systems
Ethos: ARM's Ethos family focuses on machine learning processors designed for AI and ML applications. These are primarily focused on inference use cases.
Mali: ARM's Mali family includes GPU (Graphics Processing Unit) IPs for graphics processing in various devices. Immortalis is their flagship GPU that is above the Mali IP portfolio.
Neoverse: ARM's Neoverse family targets infrastructure and data center applications, offering high-performance and energy-efficient solutions. For example, the GH200 by Nvidia uses the Neoverse platform.
Each of these IPs has targeted applications, and many companies design their mobile or data center chips starting with ARM’s design and then extending and designing more performance. ARM is the bedrock design for most chips that are not x-86 CPUs.
IP Industry Overview
This entire conversation implies that people use IPs like ARM to design a chip. That wasn’t always the case. In the past, almost every chip company fabbed, designed, and marketed their products. This included the ISA and support of the ISA. Intel’s creation and support of x86 is a perfect example. AMD has a special license to x86, but it’s mostly Intel’s ISA.
We have seen the unbundling of the design and fab in recent years since the fabless revolution. And even in the split of design, many companies cannot afford to support their own ISA and design every core from scratch. There must be some foundational design that can be reused repeatedly to create custom chips. ARM is that foundational design partner.
Another large reason for the accelerating importance of IP is that the cost of designing a chip is increasing. By working with ARM, you can offload some of the design needed and use off-the-shelf licenses instead. This is how to amortize the staggering cost of designing a leading-edge chip. The chart below is from the ARM F-1.
Many companies tape chips for less than $581 million at 5nm. But that’s partially because many of these companies are starting to rely on IP players to create hardened solutions to bring a design to market instead of having to recreate the wheel over and over.
A company bringing a relatively simple device to market is likely more focused on form factor and product fit than design. It can rely heavily on ARM for embedded designs and then focus on marketing. A company focused on high-end design is likely only tweaking very small parts of the chip and can reuse ARM IP for the rest of the chip as much as possible. The rising cost of designing a chip has become a huge tailwind for ARM and other IP players. Increasing complexity helps ARM meaningfully, and ARM is the single largest IP player in the market today.
I’m going to leave this post here for the free section. I want to discuss where I think ARM’s ecosystem is, the valuation, and what I think will happen with the IPO behind the paywall. I hope you enjoyed this free section.