Discover more from Nonobvious
ARM: The biggest IPO of the year is for patent licensing
Plus! The Apple Watch, form over substance, and Edison's 1901 battery
ARM’s IPO and why it matters
Everyone is focused on Nvidia today as the marquee chip business of 2023, but ARM is the most widespread chip company in the world. They are in every mobile phone and increasingly in every edge device like cars, refrigerators, and even data centers. Japan’s Fugaku, until recently the fastest supercomputer in the world, is even based on ARM’s architecture. How has ARM gotten to such scale? In part, it’s because of the extreme power efficiency of its chips. The other reason is its business strategy: it does not make a single chip. ARM is a patent licensor. In fact, the most successful one in the world.
And starting today, you will be able to own stock in it! ARM has filed its F-1 and was priced at $55 billion, which means it is likely to be the biggest IPO of 2023. Alongside the Instacart IPO, it could contribute to the continuing creaking open of the IPO window. ARM’s business is very misunderstood, in part because of what it means for today’s AI, and in part because of its extremely unusual patent licensing structure. Read on for a deep dive into the world’s most successful patent licensing business and how it turned into a $50 billion-plus business that powers the world.
ARM is an oddity in all aspects
ARM has had one of the most circuitous histories of any company in recent memory, and generally of which I am aware.
The essence of ARM’s technology is that it is the world expert in an arcane technology called computer instruction sets. Computer chips take code written by humans and turn them into instruction sets executed by the machine. Each instruction requires a transistor to open and close, which is called a clock cycle. A complex instruction set, or CISC, has a short instruction set, where a single instruction can make a computer chip do many things utilizing multiple clock cycles in a single instruction. Reduced instruction sets, or RISCs, instead have long instruction sets where each instruction is very simple and limited to one clock cycle. CISCs, which were promoted by Intel, beat out RISCs for decades because memory was expensive, so the shorter instruction sets of CISC chips cut a lot of costs and won out. But starting in the mid-2000s, the dominant type of computer became mobile and dependent on batteries, so the more battery-efficient RISC architectures found a market and, eventually, won out.
It is not only the technology that took a winding road; ARM’s corporate history is also unusual and long. In the 1980s, a British company out of Cambridge called Acorn Computers believed that RISC processors were the future, produced a few popular desktop computers, but overall never took off and eventually went bankrupt. However, before that happened, Acorn created a joint venture with Apple and VLSI Technology, originally called Acorn RISC Machines and, eventually, Advanced RISC Machines. ARM’s chips were always in mobile products, which were for decades relatively niche. During the dark years they weren’t even Apple’s most important RISC joint venture—that was PowerPC, with IBM and Motorola! The first notable applications for ARM were the Apple Newton, a Texas Instrument/Nokia phone, and early system-on-chips. The biggest breakthrough was the next-gen Cortex line, released in 2005, which was how ARM ended up powering the iPhone. To this day, ARM still has a bizarre structure due to a weird saga with China: they created a spinoff with a CEO who staged a coup for two years in an only-in-China fascinating saga.
Patent licensing businesses are not unheard of—there are others, like RAMBUS—but it is unusual to see one at this scale. Typically, when a technology becomes so important, it is more common to see patents live in patent pools or standard setting organizations (both topics for future newsletter posts) because market participants fear the market power of the entity that promulgates the technology—in fact, fear of ARM has resulted in an open source competitor called RISC-V funded by companies like Nvidia, which tried to acquire ARM. But ultimately, for decades, ARM’s engineering prowess in RISC is so ahead and its restraint has been so great that the urgency just does not seem that great.
How the patent licensing business works
Primarily, ARM has a large workforce of 5,963 people, of which 80% are engineers who spend all day developing new chip architectures, software, and improvements. They then patent those inventions prodigiously. ARM has 6800 issued patents and 2700 outstanding patent applications globally. They have a wide range of chip architectures, including CPUs, GPUs, SoCs, and software applications. What makes ARM’s business model somewhat unusual is that they then do not produce any chips, even with an external fab; rather, they license their technology for other companies to build on top of. For this reason, ARM is frequently described as the “Switzerland” of the semiconductor world.
The patent licensing business is mighty and unusual. On the one hand, it has many advantages. ARM’s gross margin in FY2023 was 96% on approximately $2.7 billion in revenue. Furthermore, by being a patent licensor rather than producing competing chips, ARM is able to be everywhere. Its software platform has over 15 million developers and there were 35 billion ARM chips produced last year—all without ARM having to ship a thing. At the same time, there are downsides to the patent licensing model. Read this paragraph again: There were 35 billion ARM chips shipped last year, which includes the iPhone, and they didn’t even make $3 billion. This is because ARM is only able to extract a 1-2% royalty rate based on the value of the chip, according to analysts. There are rumors that ARM is starting to up its prices by charging by the value of the whole device. This is hugely controversial, but it is allowed. Qualcomm does the same thing, and this practice was upheld on antitrust grounds in the blockbuster Qualcomm v. Apple case series, resulting in an eye-popping $6 billion final settlement.
As a business case, ARM’s challenge is one of the winner’s curse. Their revenue actually shrank by 2% this year. When ARM was purchased by SoftBank in 2019, they published a strategic report announcing their goal to grow in key areas, including cloud computing, network equipment, automotive, and consumer electronics. Respectively, their market shares in those markets are 10.1%, 25.5%, 40.8% and 32.3%. Those are all improvements relative to their goals! The problem is that these it is not clear these are actually the areas you would want to win; ARM has won in places where companies want cheap chips. ARM failed to even realize that the place where they needed to invest the most was GPUs, which has been a historic weak point for them. They will likely continue to burrow deeper into the data center as cost concerns mount for hyperscalers, but today they are more concerned about accessing H100s than anything else. Why go public now? Their parent company, SoftBank, desperately needs the cash, and the IP issues with ARM China held the IPO back for years. SoftBank only needs the cash more now, not less. The bull case is difficult to see.
On one last philosophical note, ARM makes one reconsider what a non-practicing entity, or NPE, really is. Typically, NPE is the nice word the legal profession uses for “patent troll,” which are opportunistic firms that target operating companies with shakedown suits. There is a fair amount of evidence that NPEs do behave this way, including a Harvard Business School paper, thus earning their negative reputation. But companies like ARM and RAMBUS do make one wonder if perhaps this is too narrow of an academic definition of NPE. ARM doesn’t make a single chip, but it did invent every architecture it licenses. Should it count as a type of NPE? What is the impact of firms like these? It would be great to see a study contrasting firms like ARMs with true patent trolls.
When it comes to ARM’s business, it is one for the ages. They have shown decades of grit. Whether it’s a buy is for you to decide. But they show how powerful the patent licensing business model can be.
Weekly Novelties
Gripping Gazette entries
US D998,610 S: An Apple Watch design patent, just in time for the annual Apple event
US 11,751,743 B2: The newest autonomous vacuum cleaner by Bissell
US 11,753,654 B2: A new globin gene therapy for treating hemoglobinopathies out of Sloan Kettering
Latter-day litigation
Apple Inc. v. Corephotonics, Ltd., No. 2022-1350 (Fed. Cir. Sept. 11, 2023): PTAB failed to distinguish between “a” vs “the” in a claim, in a reminder that patents are form over substance
Netflix v. DivX, No. 22-1138, — F.4th — (Fed. Cir. Sept. 11, 2023): An important ruling on prior art and analogousness relating to whether the prior art was solving the same problem
Notable news items
Huawei reached a global patent licensing deal with rival Xiaomi (Reuters)
A battery inventor is using Edison’s 1901 battery patent to design a new patent (Recharge News)
University of California tops the list for most patents of any university (Forbes)
Thank you