Arm Goes Public (Again); SoftBank’s Vision of an AI Revolution Faces Several Hurdles

SoftBank, which bought Arm in 2016, returned around 10% of Arm’s shares to other investors in a heavily oversubscribed Initial Public Offering (IPO). It attracted a great deal of attention, including from big Artificial Intelligence (AI) players like Google, Apple, Intel, and Synopsis, which were reported to have agreed to invest. This followed SoftBank’s previous failed attempt to sell its entire stake to NVIDIA, first announced in 2020, but eventually abandoned due to concerns by numerous competition regulators. Arm, a fabless chipset architecture designer based in the United Kingdom, commands a near 100% market share in the mobile phone market and licenses its RISC architecture to mainly mobile and Internet of Things (IoT) manufacturers. Given the ever-increasing demand for AI-accelerated workloads, it is unsurprising that its recent announcements have been around increasing its footprint in AI, High-Performance Compute (HPC), and data centers. Of note is also the launch of Arm Total Design, an ecosystem of partnerships for the delivery of Arm’s Neoverse Intellectual Property (IP) aimed at AI and HPC workloads—from Application-Specific Integrated Circuit ()ASIC design houses to foundries and firmware developers—all under one banner.

Zooming out, we know that SoftBank’s mission statement is an ode to AI, with a plan to be “the investment company for the AI revolution.” After SoftBank purchased Arm, it accelerated its Research and Development (R&D) and engineering capabilities in AI and computer vision, targeting applications in verticals such as automotive, industrial, and smart cities. More recently, and in the run-up to the September IPO, Arm told prospective investors of its intentions to expand into the data center AI arena, citing the opportunity presented by the growth of the cloud computing market. The partnership with NVIDIA to build an AI supercomputer using its IP for the NVIDIA Grace Central Processing Unit (CPU) reflects these goals, as well as both companies’ plans to bring generative AI applications to the telco edge. Add to that Oracle Cloud’s latest AI-optimized offering using Arm designs and the collaboration with Cadence to accelerate Arm licensees’ deployment of its aforementioned Neoverse IP.

The hurdle for completely tempting chipset players away from the legacy ecosystem looms large—Apple is one of the few examples of a company successfully bridging its processing from x86 to Arm IP. However, achieving the necessary scale to become a major competitor to legacy architecture in the HPC environment will require heavy investment in R&D to ensure the Arm ecosystem can meet the high expectations of cutting-edge AI applications. Intel and AMD collectively invest US$ billions annually in R&D, and the x86’s 40+ year legacy means there is a strong developer community and hardware flexibility across chipset suppliers. However, the increase of AI/Machine Learning (ML) applications, better served by heterogenous computing setups involving various process extensions (from scalar and vector to matrix and spatial microarchitectures) provides Arm with the opportunity to outmaneuver the established x86 monoculture found in many HPC settings, especially because this is something that Arm architecture has been doing in the smartphone and devices arena for years.

Arm’s processor IP is both complementary and substitutionary to dominant architectures like Intel’s x86: The European Union’s (EU) Jupiter supercomputer will use Arm IP (over x86) alongside NVIDIA technology. But it is unlikely that Intel would hand over cash to a company it regards solely as a threat, thereby boosting a competitor’s chances of success. The fact that Intel Foundry Services, a manufacturer of the physical chipsets, is listed as an Arm partner for its Total Design initiative contextualizes Arm’s position in the market, and demonstrates Intel’s strength as an Integrated Device Manufacturer (IDM) across the entire supply chain.

At the same time, the rise of RISC-V, an alternative open-source Instruction Set Architecture (ISA) that has provided a platform for non-legacy chipset designers to create AI-optimized IP, poses a threat to Arm’s ambitions. In August this year, Bosch, Infineon, NXP, and Qualcomm—all major Arm licensees—announced the formation of a company to spur the adoption of RISC-V in automotive (and eventually smartphone and IoT) chipsets. Separately, Google and Qualcomm announced a partnership to make chips destined for wearables using RISC-V designs. The advantages of investing in RISC-V are easy to reconcile: its implementation in semiconductor designs does not incur royalty fees, and its ecosystem, including software, grows by the day.

Furthermore, Arm’s status as a public company means it will face pressure from investors other than SoftBank to grow its market share and profits. This comes in addition to the extra reporting requirements of listed companies, which provide competitors with greater insight into its activities. The short-term profit goals of shareholders may conflict with the significant R&D investments needed to carve out space in the data center AI chipset market, and these conflicts would be exacerbated should SoftBank sell more equity to fund its AI ambitions. However, given that Arm stated, pre-IPO, that over 50% of its royalty revenue derives from the stagnating smartphone market, increasing its IP footprint in HPC and data center settings appears to be a sensible move. In fact, ABI Research forecasts that the revenue for HPC AI chipsets will grow at a Compound Annual Growth Rate (CAGR) of more than 9% between 2023 and 2028.

Several forces are at play, pulling in different directions. The first is the growth of AI and HPC workloads, which are markets worth targeting; but x86 legacy players dominate, and a significant R&D moat has been built. Intel and AMD’s R&D expenditures dwarf Arm’s, and many organizations looking to adopt Arm processors will have a strong x86 legacy. The other force is the stagnating smartphone (and other small form factors) market where Arm leads, but is now facing competition from RISC-V architecture. Qualcomm’s recent automotive and wearables RISC-V partnerships show that there is an appetite for alternative architecture solutions that do not command royalties, and these partnerships directly compete with Arm’s portfolio of IP outside of data center applications, which could eat into the latter’s revenue.

The majority of Arm’s royalties still originate from smartphone chipsets, but the plethora of announcements and partnerships clearly signal Arm’s intentions in the AI and HPC race. While it is true that AI/ML workloads are increasing and Arm IP is well-suited to heterogenous computing environments made up of various accelerators, an unrelenting drive toward AI may be what investors want to hear, but it should not come at the expense of Arm’s successful (and vast) smartphone and smaller form factor IP.