The Impact of the Top 3 Strategic Imperatives on the Custom Silicon Industry

Transformative Mega Trends

Why

The use of chips has experienced exponential growth since their introduction in the 1970s. Chips pack a significant amount of processing power into a small form factor and have fueled the electronics boom. They are integrated into an increasing number of physical goods, including toys, cars, smart appliances, and industrial devices. More recently, the use of self-contained System-on-Chips (SoC), both generic and customized, has grown. These high-performance mini-computers optimize performance and energy efficiency across a wide array of applications, including artificial intelligence (AI) and data infrastructure, driving innovation and competitiveness across sectors.

Frost Perspective

The use of chips will continue its steep growth in the global technological landscape. Chip development will become more complex as their processing and storage capacity advances and their size decreases. With the increasing adoption of AI and Internet of Things (IoT) workloads, demand will continue to grow. Additionally, a significant increase in demand and production of custom SoCs is expected to meet the specific needs of various sectors, thereby fueling innovation and competitiveness in the digital economy.

Geopolitical Chaos

Why

The reduction in suppliers capable of manufacturing chips poses an increasing threat to the supply chain of an industry experiencing higher and more sophisticated demand. TSMC is the world's largest chip manufacturing company, holding 90% of the world market share for the most sophisticated made-to-order chips and headquartered on the island of Taiwan. The Chinese government represents a continuing threat to Taiwanese sovereignty and a potentially catastrophic disruption of the chip supply. South Korea, home to another major producer, Samsung, is also threatened by North Korea, its nuclear-armed neighbor. According to the Semiconductor Industry Association (SIA), East Asia accounts for 75% of global semiconductor manufacturing. Considering this, geopolitical disruptions in the region would certainly cause a profoundly negative impact on the global economy, which is increasingly dependent on microprocessors, potentially plunging it into virtual paralysis.

Frost Perspective

The geopolitical forces affecting the microchip manufacturing industry will persist in the coming years. Authorities in the United States and the European Union, concerned about losing access to the most advanced factories, will continue to push for greater manufacturing development within their own territories while also providing economic, defense, and political guarantees to Taiwan to ensure access to a stable supply of microprocessors. However, as long as challenges posed by the Chinese government persist, the struggle for control of the sector will remain challenging and is likely to have at least a moderated impact on chip supply.

Competitive Intensity

Why

The SoC design industry is a focal point of intense competition and innovation. Established chip designers, such as AMD, Nvidia, and Intel, are facing challenges from new industry participants. Technological hyperscalers like Amazon and Google, heavy users of commercial chips in their data centers, are developing their own designs. In addition, a slew of design start-ups seeks to capitalize on the demand for customized chips tailored for AI, machine learning (ML), and IoT applications.

Frost Perspective

In the coming years, competition in SoC design is expected to intensify as the demand for customized chips for specific applications grows. However, the barriers to entry in the market for custom silicon are high. The prohibitive levels of investment and expertise required, along with the complexity of the design process, make this an endeavor that only seasoned technology companies can embark on.

Segmentation

According to Purpose

Technology companies pursuing a custom silicon strategy will need to determine what type of processor will suit their business objectives best.

General-purpose Processor

• Designed to perform a wide variety of tasks and suitable for general applications. It is versatile and can execute different programs and applications. Common examples of general-purpose processors include CPUs used in personal computing devices and servers. AWS's Graviton is an example of this type of processor.

Purpose-built/ASIC (Application-specific IC)

• These processors are designed to perform specific tasks and are optimized for a particular set of functions or workloads. They are highly specialized and can offer superior performance in specific applications compared to general-purpose processors. For example, a processor specifically designed for AI applications may have specialized processing units to accelerate tasks related to deep learning (DL). Examples of custom silicon AI accelerators for training and inference workloads include Google's TPU, AWS's Trainium and Inferentia, and Baidu's Kunlun AI.

Segmentation

According to Development Process

Technology companies pursuing a custom silicon strategy will need to determine what type of processor will suit their business objectives best.

General-purpose Processor

• Designed to perform a wide variety of tasks and suitable for general applications. It is versatile and can execute different programs and applications. Common examples of general-purpose processors include CPUs used in personal computing devices and servers. AWS's Graviton is an example of this type of processor.

Purpose-built/ASIC (Application-specific IC)

• These processors are designed to perform specific tasks and are optimized for a particular set of functions or workloads. They are highly specialized and can offer superior performance in specific applications compared to generalpurpose processors. For example, a processor specifically designed for AI applications may have specialized processing units to accelerate tasks related to deep learning (DL). Examples of custom silicon AI accelerators for training and inference workloads include Google's TPU, AWS's Trainium and Inferentia, and Baidu's Kunlun AI.

According to Development Process

Technology companies that have decided to produce custom chips have two options regarding the development process.

Fully custom ASICs

• Fully customized chips. Every aspect of the chip's design, from transistor layout to interconnection topology, is developed from scratch according to the specific application requirements. This customization allows for exhaustive optimization of chip performance and efficiency for the intended application. Due to the complexity and customization level, designing and manufacturing a fully custom ASIC is usually costly and time-consuming. Nevertheless, fully custom ASICs are often the preferred choice when investors consider that achieving a competitive edge justifies the higher investment.

Semi-custom ASICs

• In a semi-custom ASIC, a standard or predefined design core is used for development, but certain parts of the design, such as interconnection or certain functional blocks, can be customized to meet specific application needs. They offer a balance between the optimized performance of fully custom ASICs and the faster, lower-cost development of other chips. Although semi-custom ASICs offer some degree of flexibility to adapt to different applications, they do not provide the same level of customization as fully custom ASICs.

Competitive Environment

• Given the complexity of the challenges the chip industry faces, the largest and most influential users, especially public cloud providers, are designing their own chips. This gives them greater control over the design and manufacture of these critical components, enabling them to scale operations, reduce costs, and expand their range of services to support new applications.
• The market driver for custom silicon lies in the understanding that AI has marked a significant shift and will require greater processing power as business increasingly adopt customized AI applications. Currently, AI applications primarily run on GPU processors. In this regard, Nvidia has pioneered accelerated computing by extending the GPU, a 3D graphics accelerator, into a parallel computing accelerator. The company now controls nearly 80% of the AI chips sector. The generative AI boom has skyrocketed the company's value in the last year. In March 2024, Nvidia became the third most valuable company in the world, with its market capitalization surpassing $2 trillion.
• According to Nvidia, if processing power fully transitions to accelerated computing, not only will processing capacity scale considerably, but also 12 trillion watt-hours of energy could be saved annually. This amount is the equivalent to the electricity required for nearly 1.7 million1 US households.

Growth Drivers

Generative AI and other complex applications require performance-optimized processors, driving a growing number of custom silicon initiatives

• With the growing realization that a one-size-fits-all approach to processor design no longer meets their needs, technology giants are increasingly venturing into designing their own chips. Custom silicon projects ensure greater performance optimization right from the chip-level. The release of ChatGPT and the subsequent generative AI gold rush have led to an explosion in demand for silicon optimized for AI workloads, particularly for large language model (LLM) training and inference tasks. Custom silicon offers a perfectly tailored, horizontally integrated solution that can precisely meet task specifications while guaranteeing better price-performance and greater control over the supply chain in the long run.

Advanced applications drive the quest for greater energy-efficiencies, bolstering investments in custom silicon

• Applications linked to AI, IoT, cloud computing, and high-performance computing (HPC) consume significant amounts of energy. Their main limitations are power consumption, cooling issues, and space requirements in data centers. Adapting processors to specific tasks performed by servers allows technology companies to achieve the necessary performance optimization from both cost and energy perspectives. As social and political pressure on hyperscalers intensifies, custom silicon can significantly reduce carbon footprints from their data center operations. For example, AWS designed its own server-grade Graviton processor to provide consumers with sufficient compute performance and capabilities while ensuring predictable power consumption. The company reports that its proprietary silicon delivers a 40% improvement in performance per watt. The launch of the Graviton chip represents a milestone among hyperscalers, who quickly joined the movement with similar initiatives.

Custom silicon allows companies to innovate while protecting their intellectual property 

•  Custom silicon allows companies to innovate and stand out in the industry by offering unique features and functionalities not available in competing solutions. For example, a company can design a custom-made chip to offer enhanced security compared to commercial varieties, addressing enterprises’ need for secure applications to safeguard sensitive data. A custom SoC facilitates the incorporation of differentiating technologies and is very difficult to replicate, providing protection for the company's intellectual property embedded within the chip's own technology. Apple serves as an example of this—the company initially partnered with Samsung to implement some custom intellectual property into iPhone application processors in the late 2000s, but then quickly established its own chip development team and started designing SoCs for its devices. This enabled Apple to achieve better performance in its products and a significant comparative advantage over other industry participants. The chips supported Apple’s goal to differentiate its products from Android competitors

Growth Restraints

Manufacturing risks and supply chain bottlenecks potentially hamper custom silicon production at scale

• While chip design is a complex and costly task, manufacturing is even more complex. Chip factories have experienced severe bottlenecks since the pandemic, as escalating demand collides with supply chain issues, including governmental strangleholds over rare earth metals required for manufacturing. Nearly all designers subcontract the manufacturing process to specialists, particularly in Asia, which accounts for 80% of chip production. Eliminating the need to own factories drastically reduces costs. However, accessing these facilities—already operating at full capacity and with limited production—is challenging even for hyperscalers. For instance, Taiwanese chip-maker TSMC, a major supplier to Nvidia, also manufactures custom chips for Apple and AWS. Despite US government efforts to secure chip supplies for domestic companies, setting up a new fabrication plant in the United States entails significantly higher costs compared to facilities in Asia, adding further complications. The escalating geopolitical conflicts in Asia and strained United States-China trade relations will continue to pose a significant threat to the industry in the short term.

The prohibitively high costs of custom silicon initiatives form a towering barrier for potential new entrants

• Custom silicon development demands significant financial investment. A successful deployment generally requires refining over several hardware generations. Although developers do not disclose their investment figures, estimates suggest that development costs can amount to hundreds of millions or even billions of dollars. Additionally, custom silicon development requires advanced technical expertise and highly specialized human capital, which are generally scarce and come at a high cost. Finally, custom silicon development can span several months or even years, with success not guaranteed, thus increasing the financial risk.

The absence of widespread adoption of custom chips diminishes the ROI and hinders the business case for continued investments in custom silicon innovation

• A key factor for custom silicon development to be economically feasible is the scalability of its projected use. Given the costs and development efforts required by current technology, achieving the required scale for financial viability is feasible only for the largest cloud providers and tech hyperscalers. Custom silicon developers must thoroughly assess scenarios where the utilization of their services, backed by their proprietary chips, increases due to enhancements in competitive edge, or where their operational efficiency enables greater profitability following the investment required for chip development.