Commentary by Anjali Bastianpillai, Senior Client Portfolio Manager, Thematic Equities, Pictet Asset Management.
Semiconductors are the backbone of technological innovation, powering everything from smartphones and laptops to servers and supercomputers. This sector is expected to become a trillion-dollar industry by 2030. McKinsey predicts about 70% of growth will be driven by the automotive (particularly EV), data storage, and wireless industries[1].
We are positive about the impact of AI on the semiconductor market and how it will affect government and enterprise supply chain resilience.
1. Key technologies behind the surge in AI
AI is particularly resource intensive, requiring vast amounts of data and processing power to create new content. Processing semiconductors naturally seem key beneficiaries of AI, but large language models (LLMs) also require adequate memory capacity and bandwidth. Memory, compute and storage semiconductors account for the majority of semiconductor sales. Memory (“DRAM”) and storage (“NAND Flash”) chips are primarily used for storing data and instructions, while processing chips (such as the core “CPU” in a computer or a complementary accelerator chip like a “GPU”) are used for performing calculations and processing data in real time.
With a historical 10% CAGR over the last 10 years, the memory semiconductor market, accounting for 26% of semiconductors’ revenues[2], is expanding as a result of rising smartphone usage, increased digitalization, and rising semiconductor usage across numerous sectors, including the automotive and information technology industries. However, the development of AI models, particularly LLMs, require substantially higher levels of memory and storage given the size of the datasets. Indeed, the amount of memory and the speed at which memory is accessed by the processing chips is becoming one of the key bottlenecks for AI model training. The deployment of new memory technologies in semiconductors help optimize this process by speeding up data transfer, improving the performance and increasing energy efficiency.
For example, one American semiconductor manufacturing company is at the forefront of these memory technologies, whose memory semiconductors are used in everything from computing, networking, and server applications, to mobile, embedded, consumer, automotive, and industrial designs. It has recently been leading the industry in the move to higher capacity and higher speed memory chips and should benefit as AI drives further demand for memory used in servers.
The market for logic processors, accounting for 42% of semiconductors’ revenues[2], also continues to rise with technological advancements driving increased productivity, lower power consumption, higher reliability and quality. This, in turn, continues to improve the overall price / performance trend of processor chips (a trend also known as “Moore’s Law”) enabling higher consumption of electronic devices overall. AI is both a new application enabled by the improvements in logic processors and also now a driver of incremental demand both for the servers which are processing the AI algorithms, but also the end devices where the interaction with the AI models occur. Such devices could range from a traditional desktop PC or a smartphone to an embedded device like a robotic arm or a self-driving car.
For example, a fabless semiconductor chip designer, has been one of the main beneficiaries of the growth in computational requirements to power AI algorithms. Its high-end datacenter accelerator chips are optimized for high speed parallel processing required in AI training, and these chips are now accounting for over half of its revenues.
2. Semiconductors’ competitive landscape in the age of AI: few winners taking it all?
Historically, semiconductor companies have benefitted from increased consolidation.
This trend is driven by a number of factors including: economies of scale and pooling of resources to tackle rising costs of resource intensive R&D, design and manufacturing, huge demand for ever greater connectivity (e.g. 5G, IoT), ensuring supply chain security, expansion of product portfolios beyond microchips (e.g. Software products), growing demand for custom chips and foundries, acquiring highly specialized talent, demand for cloud and data center management technologies.3 For example, a $38.5B acquisition in 2020, aimed to create the industry’s high performance and adaptive computing leader. The same year, an acquisition by a leader in high-speed data, created a $40B company to position it for opportunities in cloud and 5G.
With fewer semiconductor suppliers and designers in the industry, and even fewer leading-edge manufacturers, intrinsic knowledge and highly specialized equipments are key. Some U.S. firms dominate the international AI chip design market, while South Korea and Taiwan’s semiconductor manufacturing leaders remain the titans of global semiconductor fabrication4. For instance, the world’s leading provider of semiconductor manufacturing services in Taiwan, has over 50% market share of the overall foundry market and more than 90% market share in advanced process technologies used for AI and high-performance computing. It should benefit from AI leaders’ increased demand in semiconductor as it manufactures all chips for key designers and manufacturers of GPU and graphic technologies, and customized AI processors such as the tensorflow processing units (TPUs).
AI increases demand for faster and more efficient computing. It pushes the development of cutting-edge technologies, as well as the related production and design capabilities, while creating higher barriers to entry.
Semiconductor equipment makers and software companies that enable semiconductor manufacturing should also benefit from the rise of AI.
AI trends also drive demand for semiconductor equipment companies. They provide the chipmaking systems that produce smaller, faster, cheaper, more powerful and energy-efficient microchips. For example, a dominant supplier of lithography equipment to the global semiconductor industry is essential for the fabrication of semiconductor chips. Its unique position in the semiconductor value chain allows it to directly capture the growth of new leading edge semiconductor applications in high performance / AI computing. It is the only provider of systems using Extreme Ultraviolet (EUV) light which is required to make the most leading edge microchips used in 5G, AI, and other high performance computing applications.
The design of advanced integrated circuits would also not be possible without the assistance of computer aided design software, and more specifically electronic design automation (EDA) technologies. For example, the complexity of designing leading edge microchips to enable AI applications should require an increasing use of software tools which support and automate the design and verification (simulation) of microchips, as well as perform integrity and quality testing. These tools themselves are also AI -enabled allowing chip designers to reduce the time it takes to design a chip from weeks to days.
3. Answering to a global demand in a multipolar world
Semiconductors are the most important sector in terms of share in global manufacturing. As a critical component of modern computing and due to the complexity of its supply chain, semiconductors have been the subject of intense geopolitical competition. Taiwan and South Korea are the only supplier of cutting-edge chips, respectively supplying 90% and 10% of the most advanced semiconductors. The US maintains technological supremacy when it comes to the design of the chips, whereas China remains critical to the supply chain (38% of semiconductors’ assembly, packaging, and testing) [2].
Government and industry stakeholders worldwide are making significant efforts to bolster their positions within the global supply chain through subsidies.
– In Asia, Japan decided to provide ~$3bn a year to fund new subsidies for semiconductor manufacturing while South Korea announced a ~$250bn funding plan for semiconductors.
– In the West, the US Chips Act is driving $50+ billion in investments in the semiconductor sector (with 80% being manufacturing incentives[2]), while the EU chips act represents a $40+ billion targeted support to increase production capacity.
– These subsidies will likely take several years to materialize, assuming cost differences to manufacturing in Taiwan can be solved.
With the strategic re-shoring of semiconductor manufacturing, there is a sustained tailwind for spending on equipment through the remainder of this decade. Indeed, the world’s top three semiconductor manufacturers have already announced plans to invest >$300bn in global capacity through 2030.
Finally, supply and demand mismatches for semiconductors have generated production headaches across industries, with about 75% of all shortage-driven demand involving integrated circuits and discrete semiconductors. Forward-looking companies are using artificial intelligence to increase supply chain reliance, which give them near-real-time insights into pricing and demand fluctuations.[5]
Sources:
1 McKinsey 2022, What is driving the semiconductor market
2 Barclays 2023, Semis: the race for self sufficiency
3 Clifford Chance, Semiconductor M&A Trends in 2022: Sustained Demand, Supply Chain Risk and Regulatory Scrutiny
4 Center for Security and Emerging Technology report 2022
5. McKinsey 2023, Chip hunting: The semiconductor procurement solution
