Nvidia and ARM, a deal that triggered an earthquake in the semiconductor industry

Translated from - semiwiki

There has been a lot of coverage over the past few weeks regarding Nvidia’s proposed acquisition of Arm. Much of it has centered around the $31 billion that Arm’s current owner, Softbank, is paying for Arm, and whether Nvidia can afford to acquire the asset for such an eye-watering price. Herman Hauser, one of Arm’s earliest backers, has voiced his opposition, raising concerns that Arm’s fate is vital to the future of the U.K. This concern is somewhat odd, given that Softbank is a Japanese company. Let’s explore what the impact of this merger, if successful, would be on the balance of power in the computer and semiconductor industries, and whether a combined Nvidia and Arm could be a game changer.

The next strategic inflection point in computing will be the expansion of the cloud to the edge, including highly parallel computer architectures that connect hundreds of billions of IoT devices. Nvidia occupies a unique position in this ecosystem, and if it does acquire ARM as expected in the coming weeks, complete control of the ARM architecture will ensure its dominance.

Every 15 years, the computer industry undergoes a strategic inflection point, which U.S. semiconductor analyst Mark Lipacis calls a "tectonic shift" that dramatically changes the computing model and realigns industry leadership. In the 1970s, the computer industry shifted from mainframe companies dominated by IBM to minicomputers dominated by DEC (Digital Equipment Corporation). The personal computer industry changed dramatically in the mid-1980s, with Intel and Microsoft defining and controlling the ecosystem. At the turn of the century, the industry shifted again to a mobile and cloud computing model; Apple, Samsung, TSMC, and ARM benefited the most in the mobile space, while Intel remained the main beneficiary of the cloud data center relocation. As the chart below shows, Intel and Microsoft (aka "Wintel") were able to extract the majority of operating profits in the PC era.

According to research by investment bank Jefferies, in each of the previous ecosystems, the dominant company captured 80% of the profits. For example, Wintel in the PC era and Apple in the smartphone era. The emergence of these ecosystems was not accidental, but the result of a multi-pronged strategy adopted by each company in its respective era. Intel invested a lot of money and resources in developer support programs, large developer conferences, software technology, venture capital through Intel Capital, market support, etc. The result of Microsoft's duopoly is shown in the figure. Apple does many similar things, including annual developer conferences, development tools, and financial rewards. In the case of the iPhone, the App Store played an additional role in making the product so successful, in fact, that it is now the object of complaints from developers, who have played a key role in consolidating Apple's dominance in the smartphone ecosystem. The chart below shows Apple's largest share of mobile phone operating profits.

Intel's decades-long dominance in the data center market is now under threat for a number of reasons. One is that the types of software workloads generated by mobile devices are changing. The massive amounts of data these phones generate require a parallel approach to computing, and Intel's CPUs were designed for single-threaded applications. Starting a decade ago, Nvidia transformed its GPU architecture, originally designed as a graphics accelerator for 3D gaming, into a more general-purpose parallel processing engine.

Another reason Intel is under threat is that the surge in chip sales in the mobile phone market has given TSMC a competitive advantage because TSMC was able to take advantage of the learning curve and get ahead of Intel in process technology. Intel's 7nm process node is now more than a year behind schedule. Meanwhile, TSMC has shipped more than a billion chips on its 7nm process, has achieved good yields on its 5nm process, and is making a push for 3nm. Nvidia, AMD, and Intel's other competitors all produce their chips at TSMC, which gives them a big competitive advantage.

Nvidia's Domain

Parallel computing concepts are not new and have been part of computer science for decades, but they were originally used only for highly specialized tasks, such as using supercomputers to simulate nuclear bombs or weather forecasts. Programming parallel processing software is very difficult. This all changed 13 years ago when NVIDIA launched its CUDA software platform, which is now in its 11th generation. NVIDIA's proprietary CUDA software platform allows developers to use the parallel architecture of NVIDIA GPUs to complete a wide range of tasks. NVIDIA also equipped university computer science departments with GPUs and CUDA, and after many iterative improvements, the technology has developed into a leading platform for large-scale parallel computing. This has led to a tectonic shift in the AI ​​industry - from a "knowledge-based" discipline to a "data-based" discipline, which we can see in more and more AI applications. When you say "Alexa" or "Hey Siri", the speech recognition is processed and interpreted by parallel processing software, which is likely powered by NVIDIA's GPUs.

A leading indicator of computer architecture usage is cloud data instances. The number of these instances represents the usage demand for applications from leading CSPs (cloud service providers) such as Amazon AWS, Google Cloud Platform, Microsoft Azure, and Alibaba Cloud. The top four CSPs show that Intel's CPU market share remains flat or declining, while AMD is growing rapidly and ARM Graviton is gaining favor. One thing that is clear is that the demand for specialized accelerators is very strong and is dominated by Nvidia.

As shown in the above figure, nearly half of Nvidia's sales revenue is earned by data centers. As of June this year, Nvidia's dedicated accelerator share in the cloud data instance field was 87%. Most of the growth in data center processor revenue in the past year came from Nvidia's accelerators.

NVIDIA has created a hardware-software ecosystem comparable to Wintel in accelerators. With advanced and highly competitive development tools and ecosystem support, plus an annual GPU Technology Conference and an active investment program, Inception GPU Ventures, it has harvested superior architectural performance and created the very popular CUDA software platform.

The existence of ARM

But Nvidia still has a competitive barrier that prevents it from gaining absolute dominance in the data center ecosystem: it must interoperate in the Wintel ecosystem because the CPU architecture in the data center is still x86, whether from Intel or AMD.

ARM's server chip market share is still small, but it has been very successful. In addition, the company is rapidly surpassing Intel in markets outside of mobile phones because it is a manufacturing partner of TSMC. But ARM's weakness is that the hardware-software ecosystem is fragmented, with Apple and Amazon using mainly proprietary software, and smaller companies such as Ampere and Cavium are too small to create a large industry ecosystem comparable to Wintel.

In June this year, Nvidia and ARM announced a partnership to allow ARM's CPU to work with Nvidia's accelerators. First, this collaboration allows Nvidia to add computing power to its data center business. Second, and more importantly, it puts Nvidia in a good position to build a software and hardware ecosystem around ARM, which will pose a serious threat to Intel.

The coming shift

The reason this partnership is so important today is because the computer industry is going through its next strategic inflection point. This new structural change will have a significant impact on the industry and the competitive landscape. If historical trends continue, the market size of Nvidia and ARM combined will conservatively be at least 10 times larger than the current mobile phone or cloud computing market. The stakes are huge.

There are many factors driving this new transformation. One of them is the advent of 5G, which will support more devices. One of the key features of 5G networks is edge computing, which will put high-performance computing at the edge of the network just one step away from the end device. Today's mobile phones still rely on the descendants of the old client-server architecture established in the 1990s through networked PCs. These legacy issues lead to high latency in the network, which is why we experience those annoying delays in video calls.

The next generation of networks will have high-performance computers with parallel accelerators installed at the edge of the network. These terminals - including autonomous driving, industrial robots, 3D or holographic communications, and ubiquitous smart sensors, will require tighter integration with new protocols and software architectures. This will enable faster, extremely low-latency communications through a distributed computing architecture model. In the future, the amount of data generated and required to be processed will increase by orders of magnitude, further driving the need for parallel computing.

Nvidia's roadmap

Nvidia has made its intentions clear for cloud-to-edge computing:

“AI is exploding at the edge. AI and cloud-native applications, IoT, billions of sensors, and 5G are making AI at scale possible. But it requires a scalable, accelerated platform that can drive decisions in real time and allow every industry to deliver automated intelligence to the point of action – in stores, manufacturing, hospitals, and smart cities. It connects people, businesses, and accelerated services to make the world a smaller, more connected place.”

Last year, Nvidia also announced that it would collaborate with Microsoft on smartphone edge.

This is the strategic significance of Nvidia's acquisition of ARM, and why Nvidia is willing to pay a high price to own this technology. Acquiring ARM will give Nvidia greater control over various aspects of its ecosystem and its own destiny. It will also eliminate Nvidia's dependence on Intel's computing stack ecosystem, which will greatly improve its competitive position. By owning ARM rather than just licensing it, Nvidia can add special instructions to create tighter integration with ARM GPUs. To achieve the highest performance, the CPU and GPU need to be integrated into a single chip, and since Intel is developing its competing Xe accelerator series, Nvidia needs its own CPU.