SoftBank’s proposed sale of Arm to NVIDIA should be blocked. While there are major UK national interest concerns about this transaction, the threat and likelihood of severe anticompetitive effects are global and more fundamental. NVIDIA’s control over Arm will throttle the advance of other Arm licensees in cloud datacentres. NVIDIA will extend its ascendancy there into edge computing—as is essential in emerging 5G networks and applications like Augmented Reality (AR), Virtual Reality (VR) and the Industrial Internet of Things (IIoT)—and stifle the potential for heterogeneous computing—including Central Processing Unit (CPU), Graphics Processing Unit (GPU) and Digital Signal Processor (DSP)1—in the wider ecosystem including smartphones and IoT devices.
A series of revolutions and hegemonies
Various tech firms have risen to dominance, ingraining themselves and taking the vast majority of their market’s profits for many years, while at the same time stifling their rivals’ abilities to innovate and compete. They invariably attempt to retain such positions indefinitely. Some have only been dislodged after decades, if and when major technological changes and shifts in market demand make disruption possible. In other cases, as with Microsoft and Intel a couple of decades ago, and with Facebook and Google right now, antitrust interventions after the fact have tried to limit the harms stemming from this abusive dominance and entrenchment. However, by the time the antitrust interventions have started, it is often too late to undo the damage.
In computing, IBM’s longstanding dominance with centralised mainframes was eroded by the introduction of minicomputers, first from Digital Equipment Corporation in the 1970s, including lower cost technology and with more widespread applications and usage. Introduction of highly decentralised personal computing based on low-cost microprocessors in the 1980s was initially to IBM’s benefit, but its strategic missteps soon enabled the “Wintel” vertical duopoly of Microsoft Windows software and Intel CPU chips to prevail there. In contrast, many Arm licensees have significantly shared in supply of the CPU chips for mobile phones, consumer electronic products and industrial devices. Smartphones have become the predominant personal computing devices in the last five years2. Mass adoption of broadband Internet—on PCs since 2000 and on Apple and Android smartphones with 4G LTE since 2012—has cultivated demand for this decentralised computing. While around 95% of smartphone CPUs are based on Arm’s instruction set architecture and processor designs, Apple has taken the majority of mobile device profits throughout the last decade and with more than 60 percent of these in Q3 2020. This shift to decentralised computing and broadband connectivity has also significantly fostered the rise of cloud computing in centralised datacentres as devices are continuously online and increasingly seek interaction with large troves of continuously updated data including web content and video. Intel was the main beneficiary in chips with its x86 CPUs dominating in the servers there, as well as in PCs for 40 years, but it is increasingly being contested as workloads change to include the graphics and Artificial Intelligence (AI) for which parallel processing by GPUs is best suited.
Consequently, NVIDIA’s fortunes have improved substantially over the last few years. While it initially flourished with its historic pre-eminence in GPU-based 3D graphics acceleration in the large and fast-growing market for high-performance gaming on PCs, it has significantly grown its revenues by adapting its GPU technology architecture for general-purpose parallel processing and scaling it up for use in datacentre servers. By introducing CUDA, NVIDIA’s proprietary programming environment for its GPUs, NVIDIA has grown its developer network in support of its GPU dominance. As more developers use the CUDA platform, graphics and “visualisation” capabilities are being harnessed by NVIDIA’s GPUs in numerous applications: in media and entertainment, engineering and architectural construction, product design and manufacturing, surgeon training, and vaccine development. GPUs are also particularly well suited to AI processing, as used in intelligent agents including Alexa and Siri, which are fuelling rapid growth in cloud datacentre demand.
Recent demand growth for datacentre computing is substantially for NVIDIA’s GPUs—from public cloud services providers such as Amazon Web Services—while demand for CPU market leader Intel’s processors is moderating there and has “collapsed” in enterprise and government datacentres.
The next shift in computing technology supply is underway as the cloud expands from centralised datacentres to the edge, with increasingly parallel processing soon to connect tens of billions of personal and IoT devices. While definitions for “edge” may differ and even encompass terminal devices, my definition—in the context of 5G—is close to the air interface on the network side. Powerful computing with low latency is essential for 5G signal processing and network operation (e.g. with massive MIMO radios and AI-based automated network configuration) and for real-time 5G applications including those in AR, VR and in IoT, such as with AI in Cellular Vehicle to Everything (C-V2x)-based autonomous driving and in factories (i.e. Industry 4.0 and IIoT). According to Intel’s incoming CEO Pat Gelsinger: “5G is going to represent a platform that is redefining edge computing; it will open up smart cities, smart factories, it will displace Wi-Fi. This is a powerful technology. It will also be deployed in private 5G environments as well”.
With rapid growth for GPUs in the datacentre, NVIDIA is also positioned for growth in edge computing as workloads there will also increasingly demand GPU processing. NVIDIA’s competitive challenge is that its GPUs currently need to interoperate with Intel’s x86 CPUs, which still dominate in datacentres, the edge and PCs. Intel has competing offerings and ambitions in GPUs.
NVIDIA is now jockeying for position to dominate the ecosystem for CPUs as well as GPUs though its control of Arm. It is being encouraged by those favouring NVIDIA’s ecosystem dominance and the exceptional profits this will generate for NVIDIA.
With control of Arm’s intellectual property—as a licensing company, that is Arm’s entire being— NVIDIA will make itself the overwhelming beneficiary from Intel’s fall from dominance in datacentres due to the emergence of Arm CPUs in datacentres. While establishing itself as the leading processor supplier to the datacentre—with its Arm-based CPUs, as well as GPUs where NVIDIA already reigns supreme—NVIDIA will also be able to expand this position to the emerging cloud edge, where substantial growth is anticipated. In establishing technological and market dominance across the entire extended cloud, NVIDIA will garner the market power also to subordinate the rest of the broader ecosystem of devices including smartphones and emerging IoT devices. That is where numerous Arm licensees currently compete vigorously, without significant market share concentration and with customers benefiting greatly in many ways including constant innovation, wide choice, and low prices. It is also where the new paradigm of heterogeneous computing— with CPU, GPU and DSP for image, video, graphics and radio wave processing at low power—was developed during the revolutionary growth of smartphones since the introduction of the iPhone in 2007 and Android in 2008. This is where there is now also great potential for Machine Learning (ML) and AI processing, despite this mostly being cloud-based so far. This vibrant ecosystem will be marginalised because NVIDIA will have the incentive and means to dumb-down the workloads allocated to devices, in subservience to where NVIDIA will obtain architectural control and dominance in the enlarged cloud.
NVIDIA will surely also exploit the opportunity to extend its strengthening position into PCs, where it is already the leading supplier of GPUs for gaming. Arm-based CPUs are displacing x86 there with support from Microsoft. Similarly, Apple has switched from Intel to Arm-based CPUs for its Mac computers.
Ain’t nothing goin’ on but the rent
NVIDIA and associated stakeholders are very incentivised for the firm to seek hegemony by extracting economic rents at the expense of ARM licensees. These customers will no longer be treated even-handedly by a hitherto neutral Arm, but relegated by NVIDIA in its pursuit of significant long-term competitive advantage and dominance.
While Arm’s executives seem unperturbed by the prospect of Arm’s “Switzerland” business model being broken, they are highly motivated to complete the acquisition, including $1.5 billion in NVIDIA stock for Arm employees, equivalent to $230,000 each.
The large and rapid rise in NVIDIA’s stock price and market capitalisation around the time the proposed transaction was revealed indicates the stock market’s expectation of upcoming economic rents. With the prospect of these monopoly profits baked into the stock price, it became irresistible for NVIDIA’s management to do whatever is necessary to preserve those gains. These and other NVIDIA staff immediately found themselves with deep-in-the-money stock options, including some who may have to wait several years for those to become vested and thus exercisable to cash in.
While it is generally a target’s stock price that rises on an acquisition announcement—rather than that of the acquirer—NVIDIA’s market capitalisation rose by around $74 billion, versus only $8.6 billion for Arm from the price SoftBank paid for it in 2016 to what NVIDIA has agreed to pay now. The increase for NVIDIA would be even bigger, but for the discount in NVIDIA’s stock price reflecting the significant probability and rising expectations that the acquisition will be opposed by competition authorities. For example, Morningstar Equity Research based its November 2020 Fair Value estimate for NVIDIA on a 50 percent probability the deal will not be approved due to “regulatory risk”. Morningstar valued NVIDIA stock conservatively at $340 versus a market price of $537.61 on that date with the possibility of this acquisition at $52 billion in NVIDIA’s Fair Value, and at two times that (i.e. $104 billion) if the deal closes. A lot of economic rents for many years are required to generate that all that uplift in present value — far more than can be envisaged beyond organic growth already envisaged and also reflected in NVIDIA’s lofty stock price and market valuation.
1. The spaghetti soup of acronyms in computing from data center to device is more extensive and also includes Network Processor (NPU), Field-Programable Gate Array (FPGA) and Application-Specific Integrated Circuit (ASIC).
2. Among a base or stock of around 18 billion devices used in 2018, 17 percent of these were smartphones and 4 percent of them were tablets, but only 7 percent of them were PCs. Other (non-personal and non-computing) devices included machine-to-machine (23 percent) and non-smartphones (13 percent). Source: Cisco Annual Internet Report 2018-2023, updated March 2020.