Taiwan Semiconductor: Growing Profit Centers Driving The Company In 2023 And Beyond

Taiwan Semiconductor’s (TSMC) (NYSE:TSM) second quarter 2023 earnings conference will be held on Thursday, July 20, 2023, at 14:00 Taiwan time/2:00 Eastern Time.

Based on the company’s current business outlook, management expects the overall performance for second quarter 2023 to be as follows:

• Revenue is expected to be between US$15.2 billion and US$16.0 billion.

• Gross profit margin is expected to be between 52% and 54%.

• Operating profit margin is expected to be between 39.5% and 41.5%.

Based on new revenue data, TSMC handily beat its revenue guidance for 2Q and 1H 2023, which I discuss in this article, and which the company will report on Thursday. Based on my analysis of these revenues, I expect gross profit margin of 54% and operating margin of 41.5% will be reported.

While TSMC attributed its business being impacted by weakening macroeconomic conditions and softening end market demand in its Q1 call, the end market demand is not improving.

• Global smartphone shipments fell to 287 million units in Q2 2022, the lowest quarterly figure since Q2 2020, when the pandemic first hit.
• The worldwide PC market decline slowed in Q2 2023, with total shipments of desktops and notebooks down 11.5% year on year to 62.1 million units. This follows two consecutive quarters where shipments declined by over 30%. The second quarter volume represents a sequential increase in shipments by 11.9%.

Neverthless, TSMC’s technology prowess is mitigating the drop in end-market demand, as well as the overall dour economy.

• At sub-7nm nodes, TSMC’s market share reached nearly 90%.
• Nearly 90% of its entire 3nm wafer supply is sold out to Apple (AAPL).
• 3nm wafer output will reportedly reach up to 100,000 monthly units by the end of 2023.

In this article, I address the various revenue streams for TSMC, detailing specific metrics in each, to show how TSMC is my top semiconductor manufacturing company.

TSMC Revenue Analysis 2023

TSMC’s CFO Wendell Huang noted in the Q1 2023 earning call:

“First quarter revenue decreased 18.7% sequentially in NT or 16.1% in U.S. dollar as our first quarter business was impacted by weakening macroeconomic conditions and softening end market demand, which led customers to adjust their demand accordingly.

We expect our business in the second quarter to continue to be impacted by customers' further inventory adjustment. Based on the current business outlook, we expect our second quarter revenue to be between to USD 15.2 billion and USD 16 billion, which represents a 6.7% sequential decline at the midpoint.

Moving into second quarter 2023, we expect our business to continue to be impacted by customers for the inventory adjustment. We now expect our revenue in the first half of 2023 to decline by about 10% over the same period last year in U.S. dollar terms as compared to mid- to high single-digit percent decline previously.”

On a consolidated basis, revenue for June 2023 was approximately NT$156.40 billion (~US$4.98 billion), a decrease of 11.4 percent from May 2023 and a decrease of 11.1 percent from June 2022. Revenue for January through June 2023 totaled NT$989.47 billion (~US$31.51 billion), a decrease of 3.5 percent compared to the same period in 2022.

Table 1 shows actual Q2 revenues, which decreased 5.5% in NT or 8.0% in USD QoQ. Thus, revenues grew 1.3% above guidance of -6.7% in USD.

For the first half of 2023 revenues of 989,474 mNT were generated, down 3.5% from 1,025,216 in New Taiwan Dollars, and down 6.0% in USD. Thus, revenues grew 4% above than guidance of -10% in USD.

Chart 1 shows total revenue growth by quarter between January 2022 and June 2023 (blue bars). It also shows MoM change in revenue (orange line).

One of the advantages of TSMC’s dominance at the <7nm node, as mentioned above is the higher wafer prices that increase as nodes decrease. According to the entitled Global Semiconductor Equipment: Markets, Market Shares and Market Forecasts, prices of wafers range between:

• $9,000 for 7nm and $18,000 for 3nm wafers.
• $5,000 for 65nm and $8,000 for 16nm wafers.

These higher prices mitigate the weakening macroeconomic conditions and softening end market demand largely in consumer electronics products.

Chart 1

TSMC CEO C. C. Wei noted in Q1 2023 earnings call

“For the full year of 2023, we lower our forecast for the semiconductor market excluding memory to decline mid-single-digit percent, while foundry industry is forecast to decline high single-digit percent. We now expect our full year 2023 revenue to decline low to mid-single-digit percent in U.S. dollar terms, and our business to do better than both semiconductor ex-memory and foundry industries, supported by our strong technology leadership and differentiation.”

Importantly, since revenues in 1H 2023 grew 4% more than TSMC’s guidance given in Q1 2023, and since guidance at the time for CY2023 was low to mid-single digit drop, I expect that its likely YoY revenues will be flat, not down.

TSMC revenues by Nodes

According to the recent earnings call, TSMC expects N3 to be fully utilized in 2023 supported by both HPC and smartphone applications. Sizable N3 revenue contribution is expected to start in third quarter and N3 will contribute mid-single-digit percentage of our total wafer revenue in 2023. Its N2 technology development is progressing well and on track for volume production in 2025.

As of Q1 2023, TSMC doesn’t yet include N3 in its revenues as shown in Table 2.

For products already generating revenues, Table 2 shows TSMC’s revenues for the 7nm and 5nm nodes, which are the nodes for Nvidia’s (NVDA) A100 and H100, respectively, and discussed later in this article. While the 7nm decrease is expected, the sudden ramp in 5nm between 3Q22 and 1Q23 may be due to oversupply of chips and longer days of inventory at customers.

But it must be remembered that Nvidia is mostly supplied by one chip manufacturer – TSMC – but TSMC makes chips for more than 500 customers making nearly 13,000 different chips.

At the 5nm node, Apple (AAPL) is the largest customer of TSMC with about a 50% share, followed by Qualcomm (QCOM) at 25%, as NVDA, AMD (AMD), and others make up the rest.

At the 7nm node, AMD (25%), NVDA (20%), QCOM (10%), Mediatek (10%) are the top 7nm customers of TSMC.

Thus, 5 and 7nm revenue is down because customers revenue is down, inventory is high, and purchases of chips has slowed. Apple’s revenues were down 19% in the past quarter, for example, and Qualcomm’s days of inventory has increased to 151 days, while Nvidia’s days of inventory is 175 days but down from 200 days the previous quarter.

TSMC intends to start mass-producing its smallest 2-nanometer ("nm") chips in 2025, but Intel aims to reach its 2nm node by 2024 as Samsung (OTCPK:SSNLF) tries to hit that milestone by 2025.

Revenues from Artificial Intelligence

As AI continues to reshape industries across the globe, TSMC recognizes the significance of this technology and actively participates in its development. The company's advanced semiconductor manufacturing processes are vital for the production of AI-specific chips. TSMC's cutting-edge fabrication technologies enable the creation of specialized processors, including GPUs (Graphics Processing Units) and AI accelerators, which are essential for deep learning and other AI applications.

TSMC's collaboration with AI-focused companies has yielded substantial advancements. Its partnerships with leading AI hardware developers, such as Nvidia and AMD, have resulted in the production of high-performance AI chips, facilitating breakthroughs in areas like autonomous vehicles.

Table 3 shows AI chips produced by TSMC, including newly announced AMD’s MI300.

Revenues from Packaging

TSMC’s CoWoS (Chip on Wafer on Substrate) package allows the integration of multiple chips or dies onto a single package. This enables the combination of different types of chips, such as processors, memory, and graphics, into a single package, resulting in improved performance, reduced power consumption, and smaller form factors.

AMD, Broadcom (AVGO), Marvell (MRVL), and Nvidia are among the largest consumers of TSMC's CoWoS backend technology, whereas OSATs with their fan-out packaging are vying for a potential order transition if Si-interposer based CoWoS supply falls short of demand.

For example, Nvidia has recently obtained TSMC's commitment to CoWoS support for an additional 11,000 wafers/month in 2023, according to DigiTimes, while TMSC has a CoWoS capacity of just 8,000 to 9,000 wafers/month.

Nvidia and AMD utilize about 70% to 80% of this capacity, making them the dominant users of this technology. Following them, Broadcom emerges as the third largest user, accounting for about 10% of the available CoWoS wafer processing capacity. The remaining capacity is distributed between 20 other fabless chip designers.

Two drivers for expansion of CoWoS, according to TSMC, are the move by customers to plan our long-term capacity and by the expansion of applications like HPC and PC CPUs that require CoWoS and 3D stacking which is the company’s IC advanced packaging technology illustrated in Chart 2. This will result in expansion in capacity, which is being evaluated by TSMC.

TSMC’s advanced packaging represented 7% of revenues in 2022. I forecast TSMC will generate $5.4 billion in revenue from advanced packaging in 2023 growing to $9.11 billion in 2025, representing 7%-9% of total revenue.

CoWoS revenues will grow at a CAGR (compound annual growth rate) of 19.7% between 2022 – 2025, compared to 10.0% for TSMC total revenues, according to our report entitled High-Density Packaging (MCM, MCP, SIP, 3D-TSV): Market Analysis and Technology Trends.

Chart 2

What About the Competition?

In TSMC’s Q1 earnings call, CEO Wei laid out details about its 3nm and 2nm node timeline status.

Importantly:

• N3P, scheduled to enter production in the second half of 2024, offers an additional boost to N3E with 5% more speed at the same leakage, 5-10% power reduction at the same speed, and 1.04X more chip density.
• N3X, which prioritizes performance and maximum clock frequencies for HPC applications, provides 5% more speed versus N3P at drive voltage of 1.2V, with the same improved chip density as N3P, and will enter volume production in 2025.
• N3AE, or “Auto Early,” available in 2023, offers automotive process design kits (PDKs) based on N3E, and allows customers to launch designs on the 3nm node for automotive applications, leading to the fully automotive-qualified N3A process in 2025.

But the company doesn’t speak of competition. Competition among the three companies making <7nm chips will be dependent on the technological features of its products coming from chip design and node transition roadmaps. Table 4 below shows the technology roadmap for TSMC, Samsung, and Intel between 2020 and 2025.

Samsung currently has a 60%-70% yield rate for its 3nm process, slightly higher than TSMC’s 55%. It’s:

• SF3 process will utilize Samsung’s 3nm GAP technology and will rely on GAA, or Gate All Around transistor, which the Company refers to as MBCFETs.
• SF4X, Samsung’s fourth-generation 4nm process offers a 10% performance improvement and 23% power-efficiency improvement compared to SF4, Samsung’s second-generation 4nm process. SF4X will compete with TSMC’s N4P.
• Mass production of the 2nm process for mobile applications in 2025, then expand to high-performance computing in 2026 and to automotive in 2027.

Intel (INTC) has a goal of achieving five nodes in four years. Meteor Lake is built on Intel's 7nm chip production node, otherwise known as Intel 4. Granite Rapids will be upgraded from Intel 4 to the Intel 3 process in H2 2023. Arrow Lake will follow in H1 2024 on the Intel 20A. Chips on the Intel 18A are scheduled for H2 2024.

Intel 4 and 3 are its first nodes deploying EUV and will represent a major step forward in terms of transistor performance per watt and density. The Intel 20A, due for manufacturing introduction in the first half of 2024, remains the big technological jump. It simultaneously introduces a new transistor architecture-RibbonFET (more generally called gate-all-around or nanosheet transistors)-and PowerVia backside power delivery.

Investor Takeaway

According to my analysis, strong growth for TSMC going forward also is dependent on fab construction. Importantly, the move to the smallest technology nodes at foundries wouldn't be possible without ASML's (ASML) EUV technology (extreme ultraviolet). I estimate that between 2015 and 2022, TSMC purchased 101 EUV lithography systems from ASML, while Samsung purchased 31 and Intel 26 systems.

Table 5 shows planned and potential fab construction for TSMC, Samsung, and Intel. All will benefit from either the U.S. Chips Act or the Chip Incentive Programs in different countries.

TSMC has a clear lead based on several metrics, and is six months ahead in node migration. Samsung will slowly close the technology gap with its transition to advanced logic architectures, including GAA and MBCFET at 3nm. The 50% lower power consumption and 35% overall higher performance of GAA will be a catalyst for the migration of customers to Samsung provided capacity is available. But the capacity gap will remain.

However, Samsung has been losing customers to TSMC recently. For example, Nvidia chose the Samsung 8nm node for the RTX 30 range, but Nvidia had dropped them because of low yields in favor of TSMC's 4nm process for its high-end RTX 40 series.

TSMC is my No. 1 semiconductor stock for 2023. Macro headwinds will impact the company, but less than other foundries and significantly less than memory companies. TSMC is still a solid long-term investment as it dominates the foundry model.

I rate TSMC a Strong Buy as micro factors appear to be improving and the Company continues to dominate the foundry and <7nm node sectors.

Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.

This free article presents my analysis of this semiconductor sector. A more detailed analysis is available on my Marketplace newsletter site Semiconductor Deep Dive. You can learn more about it here.