Softbank confirmed, in August, that it was either accelerating the IPO or selling its subsidiary, Arm Holdings, Inc., causing rumors of a potential buyer to flourish. Nvidia was quick to rise to the top as a likely buyer, and on September 13th, Nvidia confirmed that it was making an offer of $40 billion USD for the Arm intellectual property divisions, the bulk of Arm’s business which excludes Arm’s IoT Services Group.
Nvidia states, “This is a highly complementary combination. It unites NVIDIA, the leading AI computing company, and Arm, the world’s most popular computing platform.” Rational for this acquisition was given as follows:
- The combination will create the world’s premier computer company for the age of AI.
- Together, we will bring together NVIDIA’s leadership in AI with Arm’s vast ecosystem to drive innovation for all customers and advance computing from the cloud to the edge, including smartphones, PCs, self-driving cars, robotics, 5G, and IoT.
- Our R&D scale will turbocharge Arm’s roadmap pace and accelerate data center, edge AI, and IoT opportunities.
There is a lot to unpack here, both in what is stated, and what has been left unsaid. This offer is a very big deal even beyond the historic size. Nvidia is a technology leader, but the scope of their markets has been mostly limited to highly parallel and computationally complex applications such as computer graphics, AI training and high-performance inferencing. Acquiring this new ‘Arm’ will provide Nvidia with a very long reach into markets they have yet to seriously penetrate but could also erect barriers to fend off aggressors in the graphics and AI markets. This new reach will have significant repercussions. This article breaks these down with stated or potential ways Nvidia may deal with them.
Competing with current Arm clients
By far, one of the most difficult parts of this process will be convincing Arm’s current diverse client portfolio that Nvidia’s acquisition of Arm is clearly a win for them. Arm licenses to well over a hundred semiconductor suppliers and system integrators. Nvidia says that the Arm acquisition is very complementary, and there should be no concern with competing because they will continue to provide and develop the ARM open license agreements. But it is not so much the overlap of current product markets that could be concerning, it is the inevitable overlap of product markets based on Nvidia’s own vision of its combined resources moving forward.
The largest revenue stream for Nvidia is by far its GPUs. Most of these are sold into graphics cards for PC gaming. Over half of Nvidia’s $11.7 billion in 2019 revenue was Gaming, over twice that of its next largest division, Data Center. For GPUs, Nvidia only competes ‘directly’ with AMD, each with their proprietary graphics solutions. Intel is expected to join this arena soon.
Arm supplies Mali graphics IP. Arm’s market for Mali targets the resource constrained processing environments such as portable devices like smartphones and tablets, large markets where Nvidia is not a player, and virtually any other non-computer that has a screen for human machine interface. Part of the value proposition of Mali is the ease of design using proven Arm Cortex and Mali integration to accelerate the design process, but many of the application core customers develop alternative graphics solutions in their SoC designs, Nvidia being one of those clients.
Nvidia is a licensee of Arm applications core IP, combined with its own graphics solutions to produce premium heterogeneous SoCs targeting high-performance embedded devices such as Automotive, Consumer and Industrial devices where high-performance vision and AI processing are critical to market success. It is arguable how much current SoCs compete with other Arm licensees because they generally attract premium equipment designs that do not compete with the bulk of mainstream Arm SoC solutions.
Again however, Nvidia has been aggressively pushed SoCs with Cuda graphics and more importantly, AI accelerators to edge computing environments. Jetson Nano, Xavier and upcoming Orin SoCs have clearly put Nvidia in the path to compete with Arm clients targeting vision and AI applications at the edge.
According to Arm’s announcement of the deal, “Simon Segars and his team at Arm have built an extraordinary company that is contributing to nearly every technology market in the world. Uniting NVIDIA’s AI computing capabilities with the vast ecosystem of Arm’s CPU, we can advance computing from the cloud, smartphones, PCs, self-driving cars and robotics, to edge IoT, and expand AI computing to every corner of the globe.”. Virtually every major Arm applications processor licensee currently has a product portfolio strategy to focus on one or more of these applications or technologies utilizing AI acceleration. Even those licensees that only use the Arm application core IP and develop their own AI extensions must consider this acquisition and whether their core architecture supplier, Nvidia, would be competing with them for edge AI applications.
Given the weight of the Arm ecosystem, there is unlikely to be any mass exodus from Arm architecture, but Arm is not the only core architecture supplier. Nvidia would not be the first supplier of a dominant ecosystem, to successfully compete in a limited scope with their clients – look at Microsoft. However, without a careful strategy, it could be just the nudge to push some of those suppliers to consider MIPS or the fast-emerging RISC V IP for alternative core IP suppliers. ASIC design houses like Cadence with Tensilica cores or Synopsis with ARC cores are increasingly supplying turnkey ASIC solutions.
Nvidia value proposition to ARM licensees
Nvidia proposed Arm acquisition announcement sends an appeal to Arm clients first by assuring that Arm will continue the current business model of open licensing of all their current core technologies and the development of new generations of core technology that have served their clients historically. To sweeten this deal, Nvidia offers a strategy of introducing their own graphics (Cuda) IP and accelerated computing platforms (AI) and maybe even networking acceleration acquired from Mellanox, in the form of open licensing. This is a very strong gesture in a positive direction, but the devil is in the details.
If Nvidia tries to continue to serve Arm licensees designing for next generation vision- and AI-enhanced ecosystem, while still trying to expand their own SoC platforms, Nvidia cannot help but increase their competition with Arm licensees that have competing solutions. One option still open to Nvidia is to shift its SoC strategy to entirely partnering with SoC designers by transitioning their SoC market to partnering completely with SoC suppliers and shifting Jetson, Xavier, Orin and other SoC designs to available IP and turnkey solutions for their licensee partners, that would eliminate the conflict of competing with the SoC suppliers. It could even maintain its status quo in its primary business of GPUs with AI acceleration and high-speed networking in high-performance computing without significant additional conflict.
Extending Arm’s reach in data center
Server and cloud computing capacities are on the top of Nvidia priorities. They have a strong position which is growing rapidly. August 2020, Nvidia published annual data center revenue of $1.75B, up 167% from a year earlier.
The server market was also one of the priorities of ARM, according to their last annual report. Today they are very marginal in the server market (less than 5%), but Arm was expecting to grow to 25% by 2028. Since 2010, several Arm licensees (both start-ups and big players) have tried to develop and sell ARM-based CPU and server products. Until 2019, it was not a big success, with few examples below:
- Calxeda was founded in 2008, and bankrupt in 2013.
- AMD silenced its K12 in 2018.
- Qualcomm announced the end of its Centriq series in December 2018.
- Amazon’s 2018 Graviton processor has small run
As Arm core technology continues to push upward on performance per watt, ARM-based server solutions continue to gain traction:
- Ampere eMag and Altra family
- Nuvia announced in August 2020 Phoenix CPU, included in Orion SoC.
- Amazon unveiled Graviton2 in 2020 and includes it in its AWS cloud offer
- Marvel switches from MIPS to ARM in Thunder X for generations 2 and 3
- SiPearl with the European Processor Initiative
Arm has also broken the myth that they just cannot provide sufficient performance when the HPC tracker Top500 crowned the Arm-based Fugaku supercomputer as the top system in June of this year.
ARM-based servers could have several assets such as price/performances, energy optimization, allowing more competition and more product diversity. But ARM-based servers have also several drawbacks and entry barriers, such as missing software support, and the largest barrier being a tiny ecosystem in the very strongly x86 dominated market.
Nvidia’s value proposition in data center
Here Nvidia can have a role to play. As they have a strong position in the server market and they already succeed in creating a popular software environment with CUDA which makes GPU programming easy and popular, they could help to do the same thing with programming on ARM-based servers. Since November 2019 they have already brought full support for its High-Performance Computing (HPC) stack to ARM CPUs. Moreover, the strong Nvidia position in the server market could be a strong asset to enhance the development of an ARM-compatible ecosystem in the server market. If Nvidia can parley its expertise with graphics, AI and networking acceleration into Arm-based turnkey solutions in the server market, it could create a very competitive and diverse market competing with both Intel and AMD in the next generation accelerated server market.
The combination of Nvidia and ARM should complement both companies’ stance toward the datacenter. In 2019 ARM realized $1.6B of design IP revenue, and this has run stable under SoftBank’s ownership tenure, as the ARM business model has been one of maintenance of an already-massive mobile & embedded SoC customer base. Over the same time however, the x86-dominated datacenter segment has seen significant growth, a space in which ARM has, so far, limited participation. “If Nvidia invest enough to develop the right products (ARM CPU) for datacenter, they could have a solution excluding X86 CPU, integrating ARM CPU (to be developed) + Nvidia GPU (known to be the best on the market for datacenter)”, says Dr. Eric Esteve of IPnest.
Barriers for Nvidia to overcome in data center markets
First, some Arm licensees are already major competitors to Nvidia in certain markets. Intel’s SoC FPGAs are based on Arm Cortex A. Intel has its own major stake in data center solutions including AI acceleration. Xilinx, another Arm licensee with configurable SoCs based on Cortex applications processors and Mali Graphics have a significant stake already in data center and many other markets that Nvidia is targeting. AMD uses Arm Cortex IP in some of their SoC solutions for embedded applications. They too have a stake in the AI market and will not be eager to see Nvidia expand there. It is almost certain that at least some of these clients will reduce or completely sever their relationship with Arm if Nvidia competes directly with its own turnkey silicon server solutions rather than an IP supplied model with partners providing the silicon solutions.
Taking advantage of Arm’s reach in embedded
In the wireless mobile devices market, the largest market for Arm Cortex A applications processors and Mali Graphics and a growing market for Arm AI solutions, Nvidia seemed to imply that it does not currently compete so it has no plans to start. However, if they truly follow the proposal to enrich their Arm partnerships with Nvidia technology, it would be up to those partners to shape the future of whether Nvidia technology finds its way into next generation mobile devices and 5G solutions.
The combined R&D of Nvidia and Arm, collaborating on advancing AI, could accelerate an already rapidly growing AI ecosystem with a common development platform extending from the cloud all the way down to the low power edge. Contrary many existing myths about AI processor requirements, AI solutions can be employed even in low-power microcontrollers. Arm has already released AI accelerated Cortex M55 and U55 extensions to extend AI even to the IoT edge. If Nvidia can successfully unify the development platform from Cloud to IoT edge, it could create a very powerful force for accelerating AI development.
Further challenges for Arm embedded markets
Even more competitive is the Automotive and Industrial SoC markets where Nvidia is trying to increase their smaller market share by offering edge solutions for edge graphics and AI. NXP, Xilinx, Intel, Qualcomm, Samsung, MediaTek, Texas Instrument, STMicroelectronics, Qualcomm, Renesas Electronics Corporation, STMicroelectronics all have significant strategic investment in Automotive SoCs and are competing with Nvidia for dominance in Advanced Driver Assist Systems and Infotainment.
In Industrial applications, as Nvidia attempts to deploy ever more graphics and AI in edge systems for video surveillance, medical imaging, factory automation machine vision, robotics, military and aerospace, logistics and other industrial applications, they will be competing head-to-head with Texas Instruments, Intel, Xilinx, Socionext, HiSilicon, Ambarella, AMD, NXP, STMicroelectronics and others. These are just a few of the adjacent markets that Nvidia cannot help but target with their AI strategy. It may be very difficult for the Arm licensees in these markets to pallet licensing from an ever-growing competitor. Even Arm licensees not already licensing Arm Mali, Ethos or other coprocessing acceleration have a strategy for integrating Arm applications processors with their own AI acceleration solutions and may resent the competition enough to seek other application processor solutions.
Even in wireless, where smartphones and tablets are dominated by Arm IP, licensees such as Apple, Qualcomm, Samsung, HiSilicon and others, these major clients can’t help but be anxious about how easy it would be for Nvidia to slide into this adjacent market over time while collaborating with these current licensees. For now, it is likely that the business model does not make sense for these companies to stop licensing Arm, but each will likely already be making fallback to start developing their own IP or work with the RISC V Foundation, MIPs or other vendors to at least research a fallback plan.
According to Dr Eric Esteve of IPnest “The evolution of ARM design IP royalty market share has passed from 67.5% in 2016 to 60.6% in 2019. Nevertheless, ARM is still strong #1 in royalty, as the #2, SST, was 5.6% in 2019, or more than X1.” This dominance in the ecosystem will afford Nvidia time to develop their plan, but as can be seen by the growth of encroaching design houses, Arm’s dominance can be challenged.
Overcoming UK/EU Regulatory resistance
Four years ago, Softbank had to negotiate with the UK promising that they would leave Arm to operate as an entity with its HQ in the UK when it was part of the EU. Since Nvidia is a US company, there is concern that this is still a regulatory concern.
Nvidia value proposition
Nvidia has gone through great lengths in its announcement to assure the UK that not only will it not move Arm, that it intending “to build a world class AI lab in Cambridge because we have a great site there already with Arm. And from this from this facility, we will house a magnificent supercomputer that’s based on Arm CPU and Nvidia GPUs and Mellanox chips and it’s going to be its going to be the most advanced AI supercomputer the world’s ever known that we will build there.” . If the UK is satisfied with Nvidia’s commitment, this plan is likely to avoid UK regulatory issues around jobs.
Politics may always play a role and resentment in having the US company with ownership of one of the UK’s premier technology companies may still not sit well with some politicians. The reduction of competition in some specific markets such as graphics suppliers could entice legal battles. The EU, still in post-Brexit negotiations could use further regulatory hurdles or simply perceive antitrust issues.
Overcoming US/China regulatory issues
Perhaps one of the gravest barriers to the success of this acquisition lies mostly outside of Nvidia’s and Arm’s control. In the current political climate, it is plausible that the US government could impose restrictions on Nvidia and its subsidiary, Arm, from supplying process technology to a Chinese company, or in an unlikely event they chose to block the acquisition. China could easily perceive the potential of sanctions and decide they wish to block approval for the acquisition to prevent the scenario in the first place. Nvidia may not have as much to say about their commitment if the US or China decides to make this a regulatory issue. The only analysis to offer here is that we have a US election and 18 months of negotiations to discover how this plays out.
Nvidia value proposition
Nvidia advises it is completely committed to supplying the same level of support to advancing processor technologies for Arm JV where it will continue to be a minor share holder. Nvidia has said it is committed to continue supplying the open architecture Arm is already famous for.
Overcoming cultural differences
Having Arm continue as a separate entity from Nvidia US will create some of its own culture challenges. There is a great divide in the current Nvidia and Arm business models. It may be hard to reconcile that Arm’s reach is so large because its pricing structure is so limited. Its clients have significant limits on what they can afford for processor integration while Nvidia offers advanced solutions with premium pricing.
Combining the business models and cultures of these seemingly dissimilar companies could be a challenge. It will be easier for microcontroller designers to find alternatives elsewhere and Nvidia may find itself in a position of devaluing support for cheap, deeply embedded technologies. It could see them as a drain on its core vision of a high-performance data center and edge device market supplier. MCU clients, of which there are many, could be concerned about a slowing of development outside of IP that Nvidia sees as synergetic to their current portfolio. Nvidia will need to prove its commitment to MCUs or it will create more opportunities for competing architectures such as RISC V.
Nvidia has not really addressed this issue. Nvidia could maintain compartmentalized business units with unique business unit commitments to avoid these potential clashes in culture, but it will walk a very fine line to push toward its unified vision of an AI computing powerhouse in doing so. It will be interesting to see how a processor supplier targeting performance applications with solutions that sometimes exceed hundreds to thousands of US dollars reacts when presented with requests for budget to develop the next generation $0.10 microcontroller architecture IP.
In summary, this is an exceptional deal in many ways, but many relevant entities learned of it at the same time Nvidia announced it to the public. Nvidia has clearly given it a great deal of consideration, but there are a few challenges to be overcome before the long Arm of Nvidia becomes a reality.
Tom Hackenberg is a Principal Analyst for Computing and Software in the Semiconductor, Memory and Computing Division at Yole Développement (Yole). Tom is engaged in developing processor market monitors and research into related technology trends. He is currently focused on low and ultralow power solutions such as MCUs. Tom is an industry leading expert with more than a decade’s experience reporting on markets for semiconductor processors including CPUs, GPUs, MPUs, MCUs, SoC ASICs & ASSPs, FPGAs and configurable processors. Tom is also well-versed in related technology trends including IoT, heterogeneous processing, chiplets, AI and edge computing.
Prior to joining Yole, Tom was a principal analyst at OMDIA, IHS Markit and began processor market research in 2006 for IMS Research. He worked with market-leading processor suppliers developing both syndicated and custom research. Tom holds a BSECE from the University of Texas at Austin specializing in Processors and FPGAs.
Eric Esteve (PhD.) is part of the DAC IP committee since 2016, has posted more than 400 blogs and articles, white papers or success stories since 2011. He has worked as ASIC designer (1983-1992) then in various sales & marketing positions with TI, Atmel, PLDA and Snowbush (1992-2008).
About IPnest: IPnest provides IP centric market surveys dedicated to emerging products like eFPGA, Interface Protocols (PCI Express, MIPI), Verification IP market or low power technology (FD-SOI, MIPI Ecosystem) since 2008. IPnest can be contracted for on demand research or specific consulting (ASIC, IP marketing and more). IPnest’s customer list counts major foundries, IP and Verification IP vendors and IDM or Fabless, totalizing more than 40 satisfied customers. IPnest also market the “Design IP Report”, covering the overall IP market by category (CPU, GPU, DSP, Interface, Mixed-signal, etc.), since May 2017.
Application Processor Quarterly Market Monitor
With SMIC, China is positionning as a competitor in foundry business and thereby catching the
dynamic of the APU market.
Related Live Market Briefing
Application Processor Market Responses to COVID and Beyond – Live Market Briefing Video
The APU market is a representative cross-section of the logic and foundry space, whose segments bear witness to varied COVID impacts.
Related Reports and Monitors
WLCSP/ Fan-In Packaging Technologies and Market 2020
Market & Technology
Glass Substrate for Semiconductor Applications 2020
Market & Technology