Taiwan Semiconductor Makes Chips For 3 Of The Top 4 Greatest Growth Technologies

In my research, I've identified four of the greatest growth chip sectors, and Taiwan Semiconductor (TSMC) (NYSE:TSM) manufactures three of the four. This suggests the technological prowess of TSMC and the potential to outperform the market and the industry.

As shown in Chart 1, I’ve plotted growth rates for the Top 4 semiconductor growth sectors between 2021 and 2026:

1. SiC (Silicon Carbide),
2. GaN (Gallium Nitride),
3. AI Compute Processors (Artificial Intelligence), and
4. Gen AI (Generative AI).

Note in the chart that the Generative AI data, also commonly known as ChatGPT, is 1/10^th the revenues to fit on the chart.

TSMC makes chips for sectors 2-4, as I discuss below.

Chart 1

Generative AI, also known as ChatGPT, is projected to exhibit a CAGR of 67.6% between 2021 and 2026. AI Compute Processors are traditional processors (GPUs, CPUs, ASICs and FPGAs), growing at a CAGR of 27.0%, are used in traditional AI Training and Inference in servers.

Chart 1 also plots CAGR of GaN Devices, projected to exhibit a CAGR of 42.5% and SiC Devices projected to exhibit a CAGR of 53.2% during this period. Of these sectors, TSMC does not participate in SiC.

Below I discuss each of these sectors.

Traditional AI

AI servers are computers outfitted with GPUs, FPGAs or ASICs used for AI Training and Inference. Training reevaluates or adjusts the layers of the neural network based on the results. Training a leading AI algorithm can require a month of enormous computational power. Inference applies knowledge from a trained neural network model and uses it to infer a result. AI chips are mostly used to apply trained AI algorithms to real-world data inputs; this is often called “inference."

• Training Accelerators: These specialized chips are optimized to facilitate the training of deep learning models, involving intricate calculations and the processing of extensive datasets.
• Inference Accelerators: Designed for the rapid execution of trained models on fresh data, making them ideal for real-time applications such as image recognition in cameras or voice assistants in smartphones.

Nvidia (NVDA) is the leader in the data center training chip market with nearly a 99% share in 2019 but that lead is dropping to an 89% share in 2023 as ASICs gain share on its GPU's as accelerator chips.

Generative AI

Artificial Intelligence has witnessed remarkable progress over the years, and within the realm of AI, two distinct approaches have emerged: Traditional AI, discussed above, and Generative AI. These approaches differ fundamentally in their objectives, learning methodologies, data requirements, use cases, and capabilities.

One of the key features of traditional AI is its reliance on structured, labeled data for training. The quality and quantity of data are pivotal for the performance of these models. Tasks such as image recognition, sentiment analysis, and recommendation systems have benefited greatly from traditional AI techniques.

Generative AI, in contrast, is a paradigm that aims to simulate human-like creativity and generate content autonomously. Unlike traditional AI, which is confined to specific tasks, generative AI is versatile and capable of producing diverse outputs across various domains, including text, images, music, and even entire applications. The hallmark of generative AI models is their ability to generate content that goes beyond the scope of their training data.

The various architectures of AI chips are listed below:

• GPU (Graphics Processing Unit): GPUs are widely employed for AI acceleration due to their parallel processing capabilities, making them well-suited for tasks such as deep learning and image processing.
• TPU (Tensor Processing Unit): TPUs are custom-designed chips by Google, meticulously optimized for TensorFlow-based workloads, with a particular focus on neural network inference and training.
• FPGA (Field-Programmable Gate Array): FPGAs are reconfigurable chips capable of being programmed to execute specific tasks, including AI computations. Their adaptability offers versatility, but programming expertise may be necessary.
• ASIC (Application-Specific Integrated Circuit): ASICs are custom-designed chips engineered explicitly for AI computations, delivering exceptional performance and energy efficiency for targeted tasks.
• Neuromorphic Chips: These chips are engineered to emulate the structure and functionality of biological neural networks, catering to specialized AI applications requiring brain-inspired processing.

Table 1 shows the list of AI processors manufactured by TSMC.

GaN Tailwind

Gallium Nitride is the third high-growth application area of TSMC. The foundry provides the industry’s largest and most advanced GaN-on-Silicon wafer manufacturing capability. It started manufacturing GaN Systems (Ottawa, Canada), which currently offers a portfolio of GaN transistors in ratings of 100 and 650 V, the latter commonly considered the upper voltage limit for GaN devices.

Some of the partnerships with TSMC are below:

Investor Takeaway

Chart 2 shows TSMC’s revenues by application between Q1 2019 and Q2 2023. In the recent Q2 2023 quarter by platform, High Performance Computing ('HPC") and Smartphone represented 44% and 33% of net revenue respectively, while IoT, Automotive, DCE, and Others each represented 8%, 8%, 3%, and 4%.

These were based on TSMC’s technology, where 5nm process technology contributed 30% of total wafer revenue in 2Q23 while 7nm accounted for 23%. Advanced technologies (7nm and below) accounted for 53% of total wafer revenue.

Chart 2

In Q1 2019 by platform, HPC and Smartphone represented 32% and 46% of net revenue respectively, while IoT, Automotive, DCE, and Others each represented 6%, 5%, 6%, and 5%.

Importantly, HPC’s percentage of revenues increased from 32% in Q1 2019 to 44% in Q2, 2023. Revenues for these periods increased from 291 billion New Taiwan Dollars to 451 billion NT$, equating to a 2X growth for HPC from 93 billion NT$ to 198 billion NT$.

In a discussion of TSMC’s HPC in the Q2 earnings call, C.C. Wei, CEO noted:

“The recent increase in AI-related demand is directionally positive for TSMC. Generative AI requires higher computing power and interconnected bandwidth, which drives increasing semiconductor content. Whether using CPUs, GPUs, or AI accelerator and related ASICs for AI and machine learning, the commonality is that it requires use of leading-edge technology and a strong foundry design ecosystem. These are all TSMC's strengths.

Today, server AI processor demand, which we define as CPUs, GPUs and AI accelerators that are performing training and inference functions accounts for approximately 6% of TSMC's total revenue. We forecasted this to grow at close to 50% CAGR in the next five years and increase to low teens percent of our revenue.

Our HPC platform is expected to be the main engine and the largest incremental contributor to TSMC's long-term growth in the next several years.”

The “close to 50% CAGR” reported by TSMC is consistent with my analysis of 47.3% for the average CAGR of Traditional AI and Generative AI shown in Chart 1. The traditional AI sector has been stymied, as enterprise spending experienced a deceleration in 2023, as it tapered off from the surge in pent-up demand that had been fulfilled in 2022. I anticipations another expansion cycle in 2024, particularly for certain cloud service providers.

But the data center industry is poised for robust growth over the next half decade, fueled by the swift adoption of new artificial intelligence applications and a fierce competition for supremacy among tech giants such as Amazon (AMZN), Google (GOOG) (GOOGL), and Microsoft (MSFT). This is anticipated to result in increased investments and sustained double-digit growth.

On the GaN sector, the industry is currently small and in the “Others” category. According to my analysis, the current global revenues for GaN devices was just $200 million in 2022 (6.3 million NT$) compared to TSMC’s 198 billion NT$ for HPC. Yet revenue should increase 5X in 2026.

While GaN offers substantial growth potential across these sectors, challenges such as manufacturing costs, scalability, and reliability still need to be addressed. As technology advances and production processes become more optimized, the growth potential for Gallium Nitride is expected to continue expanding, driving innovation and advancements in various industries.

The strong growth of these sectors will translate to strong growth for TSMC. I reiterate TSMC a Buy.

This free article presents my analysis of this semiconductor equipment sector. A more detailed analysis is available on my Marketplace newsletter site Semiconductor Deep Dive. You can learn more about it here and start a risk free 2 week trial now.