Deep Dive into Arm's Prospectus: Can the Dominant Player in the Mobile Era Sustain its Myth in the AI Era?

Arm is impacting Nasdaq and becoming one of the most anticipated topics in the tech industry. Arm will officially debut on Nasdaq on September 14th with a valuation of up to $50 billion, making it the largest IPO in the world this year and potentially the third-largest IPO in the history of the US stock market, following Alibaba and Facebook. As a leader in chip design, Arm has a firm grip on the "neck" of global tech giants. The Arm architecture is the most widely used CPU architecture worldwide, having nurtured over hundreds of billions of chips, covering over 99% of the global smartphone market. Arm's listing has stirred up waves in this year's stagnant IPO market, while the booming AI trend continues to drive up Arm's value. However, despite holding the architecture dominance, Arm is not making big profits, and its "AI capabilities" are also lacking. Can it support such a high valuation? Next, let's look beyond the surface noise and delve into the prospectus to see what kind of company Arm really is. Industry Cornerstone: Holding the "Neck" of Global Tech Giants What kind of chip company is Arm? And how did it rise and grow into an industry leader? From the following introduction, we can understand Arm's industry position and the products it provides. Arm is undoubtedly the leader in the CPU industry. ARM specializes in designing, developing, and licensing high-performance, low-cost, and energy-efficient CPU products and related technologies. Its architectural advantages have made Apple, NVIDIA, Google, and other giants its customers. Arm's products have a wide reach, including smartphones, tablets, navigation systems, and even network devices. Almost all of them adopt Arm architecture CPUs. Approximately 70% of the global population uses Arm-based products. In terms of market share and shipments, as of December 31, 2022, Arm's CPUs accounted for over 99% of the global smartphone market, with cumulative shipments reaching 250 billion. By the end of the fiscal year on March 31, 2023, the shipment volume of chips based on Arm architecture exceeded 30 billion. 1. Rising with the RISC Approach Historically, Arm's rise is essentially due to its correct bet on the RISC (Reduced Instruction Set Computing) approach. Arm was founded in 1990 as a joint venture between Acorn, Apple, and VLSI. At its inception, ARM set the goal of developing "high-performance, low-power, programmable, and scalable processors." Over the past thirty years, this mission and market positioning have remained unchanged, and high performance and low power consumption are essentially the path of RISC. Arm's CPU architecture can be divided into two sets: RISC (Reduced Instruction Set Computing) and CISC (Complex Instruction Set Computing). The difference between the two lies in the fact that CISC has higher performance and power consumption, while RISC has lower performance but lower power consumption. Intel and Arm represent each of these architectures, respectively. In the PC era, the pursuit was for computing speed and performance, so Intel's X86 instruction set and architecture almost monopolized the entire computer chip market. However, in the mobile era, the focus shifted to a balance between computing power and power consumption. As a result, Arm, which adopts a low-power, high-performance RISC architecture, quickly rose to prominence. Arm CPUs also played a key role in the smartphone revolution. "Licensing Fees + Royalties" as Two Revenue Streams Furthermore, Arm occupies a special position in the industry chain and has a unique business model. The entire chip industry chain can be roughly divided into chip design, chip manufacturing, and chip packaging. Before chip design, there is a need for instruction sets and architectures, which can be likened to "design blueprints." In simple terms, Arm develops the "design blueprints" for CPUs, and companies like Apple and NVIDIA purchase these blueprints and refine them according to their needs to design chips. Finally, they turn to TSMC to manufacture the chips based on the design. Without the "design blueprints," it is difficult to design chips, and Arm dominates in terms of "design blueprints" for mobile devices. In its prospectus, Arm points out that with the exponential growth in CPU design complexity, no company has successfully designed a modern CPU from scratch in the past decade. Understanding Arm's foundational position in the industry chain, its business model becomes easier to comprehend. Arm earns money by selling "design blueprints" (intellectual property), mainly through two aspects: License fees: Customers need to pay a fee to obtain Arm products ("design blueprints") and develop processors based on the Arm architecture. Royalties: When products based on the Arm architecture are designed and manufactured, customers need to pay royalties to Arm based on the selling price of each chip, typically ranging from 1% to 2%. In short, Arm not only sells "design sketches," but also collects taxes from the "finished products" made from those "design sketches," which generate continuous revenue. In fact, "design sketches" are also divided into different levels. Chip manufacturers with strong capabilities can design chips with simplified sketches, while manufacturers with slightly lower capabilities require more precise sketches. At the same time, there are differences in architecture, corresponding to different licensing restrictions. Traditionally, most customers authorize Arm's products based on the terms of the TLA. In 2019 and 2021, Arm introduced the Arm Flexible Access and Arm Total Access agreements, respectively, to allow as many customers as possible to access Arm products in the most suitable way for their specific business needs. Holding Architectural Dominance but Not Making Big Money Although Arm holds architectural dominance, the actual profits are not as substantial. 1. Small Scale of Revenue and Net Profit Due to the downturn in the smartphone market, Arm's performance has been poor. For the fiscal year ending on March 31, 2023, the operating revenue was $2.679 billion, lower than the $2.703 billion in the same period last year, and the revenue for the 2021 fiscal year was $2.027 billion. Specifically, Arm's revenue mainly comes from license fees and royalties, with license fees accounting for 40% and royalties accounting for approximately 60% of the revenue. Among them, royalties are the most profitable business for Arm. However, most Arm-based products are microcontrollers (MCUs), which are relatively cheap. On the other hand, Arm collects fees from chip prices rather than charging for end products, resulting in relatively low revenue. In the 2023 fiscal year, Arm shipped over 30 billion chips, collecting an average of around $0.1 per chip. In 2022, the total value of Arm-based chips reached $98.9 billion, accounting for nearly half of the market share. However, Arm's royalty revenue was only $1.68 billion, accounting for 1.7% of the chip value. However, the advantage of royalties lies in continuous revenue from one-time development. Old chips can generate ongoing income. In the 2023 fiscal year, approximately 46% of Arm's royalty revenue came from products released between 1990 and 2012. For example, the ARM7 TDMI architecture chip sold to Nokia and Texas Instruments in 1997 is still being shipped in 2020, making continuous profits for over 20 years. In terms of expenses, as a chip design company, Arm's R&D costs are not low, averaging around 40% of its revenue. This has resulted in Arm not making big money, with net profits for the fiscal years 2021-2023 being $388 million, $549 million, and $524 million, respectively. In the past three years, Arm has accumulated $1.4 billion in profits, which pales in comparison to the profits of its customers. Qualcomm had a net profit of $13 billion last year, and NVIDIA made over $4 billion. According to the prospectus, the improvement in net profit margin in the past two years is mainly due to the layoff and restructuring plan implemented in March 2022. From the implementation results, Arm has basically maintained the same number of engineers while significantly reducing the number of management personnel at its UK headquarters. 2. Insufficient Industry Influence Fundamentally, Arm's inability to make big money may be due to its insufficient industry influence. Some analysts point out that the most profitable companies in the entire industry chain are the chip companies that control the software ecosystem, as they directly determine the competitiveness of a chip in the market. According to a report by Wall Street analyst Robert Castellano, NVIDIA makes $40,000 from its GPU chip H100, TSMC makes $1,000, and SK Hynix, which provides memory chips, makes $2,000, leaving NVIDIA with almost 90% of the revenue. In addition, ARM has always maintained a neutral position, opening up its architecture standards to major companies, which has led to its unstable position. In contrast, its competitor Intel not only designs IP but also owns its own chip factories and monopolizes the ecosystem by forming the Wintel alliance with Windows, thereby controlling the industry chain. How Much "AI Value" Does It Have? Arm's listing coincides with the AI boom, and the frenzy over AI has driven up NVIDIA's stock price. To boost Arm's valuation, Masayoshi Son has been riding the AI wave. Masayoshi Son believes that ARM can achieve "exponential growth" through AI and has told investors that Arm is in the central area of AI-related companies, which can generate synergies. 1. Insufficient AI Value However, from the prospectus, Arm's AI value is insufficient. Arm mentions in its prospectus that with the rapid development of artificial intelligence (AI) and machine learning (ML), Arm plays a crucial role in these fields and collaborates with multiple leading companies to handle AI work. "CPU is crucial in all artificial intelligence systems, whether it is fully handling AI workloads or working in conjunction with co-processors such as GPUs or NPUs," however, Arm focuses on the architecture foundation of CPUs rather than creating architectures for GPUs and AI-specific chips required for large models. In the risk disclosure section, Arm admits that emerging technologies such as AI and ML may adopt algorithms that are not suitable for general-purpose CPUs (such as Arm's processors). Therefore, the importance of its processors may diminish in chips based on the Arm architecture. "2. Upgrades around AI and ML" In other aspects, Arm's latest Armv9 architecture has made certain upgrades around AI and ML. In May of this year, Arm officially launched the latest Armv9 architecture, marking the first major adjustment and transformation of its CPU architecture since the introduction of the Armv8 architecture more than ten years ago in October 2011. "For developers and users, the most significant feature of the brand-new Armv9-compatible CPUs is the introduction of SVE2 as the new baseline after ARM NEON technology." SVE2 was released in April 2019 with the aim of accelerating high-performance computing and providing significant gains in handling 5G, VR and AR, as well as ML workloads. Arm believes that in the coming years, ML workloads will become increasingly common. Accordingly, any device that is performance- or power-centric will need to run ML workloads on dedicated accelerators, but most of them will still choose to adopt smaller-scale ML workloads running on CPUs. "3. Key drawback: Insufficient computational power" In fact, Arm's weakness in the field of AI is also evident, namely, insufficient computational power. The AI server field is mainly dominated by x86 architecture and GPU architectures such as Nvidia Turing, Volta, and Ampere. Arm architecture is gradually gaining prominence in the server field, especially in lightweight and low-power servers, but its market share is still small. AI has very high performance requirements, and x86 architecture can adopt high-performance CPUs, coupled with large-capacity memory and high-precision graphics cards, to continuously provide computational speed. The RISC-based approach taken by Arm determines its performance limitations. If it cannot further improve its performance, the drawbacks of the Arm architecture will become increasingly apparent. Therefore, Arm seems to be on the edge of the AI boom, making it difficult to reap the benefits. "Growing increasingly challenging, competition becoming more intense" In recent years, the growth space for Arm has been shrinking while the level of competition continues to rise. 1. Increasing Difficulty in Growth In order to win over investors, Arm must find a way to generate higher revenue, but no matter which direction they turn, there are pitfalls. To boost revenue, Arm hopes to shift royalties from chip prices to terminal prices, while forcibly "bundling sales" of CPUs, GPUs, and NPUs. However, this approach faces significant pressure and may force customers to develop their own architectures. There are reports that Qualcomm will bid farewell to the Arm architecture and adopt its self-developed Nuvia architecture for the first time next year. Therefore, Arm is focusing on designing its own chip products. Some reports indicate that Arm wants to independently design processors and prioritize the use of its most advanced architecture. However, this means encroaching on the territory of customers, as chip design manufacturers previously cooperated with Arm due to its neutral position. If Arm enters chip design and produces its own chips, it would be equivalent to crossing the customers' red line. These two "paths to growth" are fraught with danger. If Arm persists in this approach, it may jeopardize the ecosystem it has built over the years and accelerate customers' shift towards competitors. 2. Increasing Competition For example, the open-source and free RISC-V architecture is eyeing Arm's existing customers and continuously attempting to penetrate the market. With its advantages of being open-source, streamlined, and modular, the RISC-V architecture has become a "rising star" in the field of chip architecture. The open-source nature of RISC-V has attracted the attention of many companies, especially small and startup enterprises, as it allows for easier access to chip design authorization and reduces costs. Arm also mentioned in its prospectus that many of our customers are major supporters of the RISC-V architecture and related technologies. If the technologies related to RISC-V continue to develop and market support for RISC-V increases, our customers may choose to use this free open-source architecture instead of our products. Furthermore, it is worth noting that nearly one-fourth of Arm's revenue comes from the Chinese market and heavily relies on Arm China. However, ARM has no direct management control over ARM China and does not have any controlling seats on the ARM China board of directors. ARM wants to maintain its position, so it must continue to invest and develop new architectures. ARM admits that research and development is the core of its business and the key to its future success. However, this is easier said than done. Further innovation becomes increasingly difficult and costly. Developing new products requires a significant amount of funding and time, usually taking five years or more, with no guarantee of the desired results. Clearly, ARM does not have an easy path ahead. Is it worth more than NVIDIA? In fact, this is not the first time ARM has gone public. Between 1998 and 2016, ARM was listed on the London Stock Exchange and NASDAQ, before being privatized by SoftBank Group for $32 billion in 2016. At the time of the acquisition agreement, SoftBank's founder, Masayoshi Son, was very excited, stating that ARM would be SoftBank's future. Seven years have passed, and due to poor investment returns from the SoftBank Vision Fund, with losses exceeding $39 billion in 2022 alone, along with the failed sale of ARM, Masayoshi Son needs to raise funds through ARM IPO and prove his investment vision. Based on a maximum price of $51 per share, ARM is valued at $54.5 billion, significantly higher than SoftBank's initial acquisition price. This could potentially be the most profitable deal for Masayoshi Son. If priced at the midpoint of the IPO price range, ARM's price-to-sales ratio is around 18 times, higher than any other constituent of the Philadelphia Semiconductor Index, including NVIDIA. However, many analysts are skeptical of SoftBank's optimistic expectations, believing that the $50 billion valuation may be too high. ARM is not NVIDIA, and from certain perspectives, ARM is almost as expensive as NVIDIA. However, its business growth rate, revenue, and price-to-earnings ratio are unlikely to support Masayoshi Son's vision or the $50 billion valuation. Investors generally compare ARM to EDA software companies. Compared to the valuations given to leading chip design companies like Synopsys and Cadence Design Systems, ARM's valuation is currently higher. If we use the price-to-sales ratio of these listed companies as a benchmark, ARM's valuation would be between $32 billion and $43 billion. Despite concerns about the high valuation, there is strong market demand, with oversubscription of 10 times on Tuesday. ARM is expected to go public on September 14th (Thursday) and will face the market's "test" at that time.