Summary
TSMC reported a weak Q1 earnings call and guidance because of a dire economy resulting in lower GPU/CPU shipments, revenue, and ASPs.
TSMC is impacted by a slowdown in constrained hyperscale budgets in the near term due to the challenging macroeconomic scenario.
Strong tailwinds will prevail in 2024 as TSMC increases capex to meet AI chip demand.
BING-JHEN HONG
On May 31, 2023, I wrote a Marketplace Semiconductor Deep Dive article for subscribers entitled TSMC: Reconciling Weak Q1 Financials With Nvidia’s Blowout Guidance Claims. In the article I noted that Nvidia $NVIDIA Corp(NVDA)$, which uses TSMC $Taiwan Semiconductor Manufacturing(TSM)$ to manufacture its GPUs for its data centers and other segments, reported a strong earnings call and guidance while TSMC reported a weak Q1 earnings call and guidance.
Tied to my analysis in the article, today in Seeking Alpha note by Ravikash, SA News Editor, that Applied Materials, other semi-cap stocks cut to Hold at Needham on AI hype. The SA Editor wrote:
“The analysts' noted that based on an analysis, they do not think that the surge in demand for AI systems will have a meaningful effect on foundry/logic and memory utilization rates or catalyze higher WFE spending in the near-term.
The firm thinks that incremental wafer demand at Taiwan Semiconductor Manufacturing Company (from NVIDIA's demand upside amounts to only a few thousand incremental wafer starts per month over the next year.”
This corroborates my analysis for capex spend in the near term, as I show in Table 1, presents my analysis that for 2023 and taking the midpoint of guidance, capex will drop 6.3%. But this drop follows a 21.0% increase in 2022 and a 74.4% increase in 2021.
The Information Network
TSMC is cutting capex for 2023 because of the overhang of slow sales to PC and Smartphones that won’t recover to 2022 levels until 2024. As a result, TSMC’s fab utilization has dropped from 107% in 2022 to 88% in Q1 2023. This means, in principle, that in Q1, 12% of fabs were idle.
While NVDA’s upside demand is only a “few thousand incremental wafer starts per month over the next year,” as Needham reports, I show that for 2024, capex spend by TSMC will increase 39.4% to $47.4 billion precisely because of demand for chips from NVDA and other AI chips at the data center, as well as a recovery in demand for consumer products.
Also, some of this will be associated with the expansion of fabs as part of the various country Incentive Programs such as the U.S. CHIPS Act.
Strong Long-Term Tailwinds for TSMC
While TSMC is impacted by a slowdown in AI-based spend and constrained hyperscale budgets in the near-term due to the challenging macroeconomic scenario, long-term tailwinds are strong.
Strong Growth at Small Nodes
TSMC's growth over the years has been remarkable, positioning the company as a key player in the semiconductor industry.
Chart 1 shows TSMC's revenue by technology node between Q1 2018 and Q1 2023. Revenues from the 3-10nm nodes have been increasing, and in 2022 represented 52% of total revenues (purple bar).
Chart 1
TSMC
Close examination of Chart 1 above shows that revenues at the 3-10nm decreased in Q1 2023 from 54% of revenues to 51%.
At these 5 and 7nm nodes, revenue is down because customers' revenue is down, inventory is high, and purchases of chips have slowed. Apple’s $Apple(AAPL)$ revenues were down 19% in the past quarter, for example, and Qualcomm’s $Qualcomm(QCOM)$ days of inventory have increased to 151 days, while Nvidia’s days of inventory are 175 days but down from 200 days the previous quarter.
TSMC’s 3nm technology is now in volume production with the N3 process and the enhanced N3E version on the way in 2023, TSMC is adding new variants to the roadmap to suit customers’ diverse needs.
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.
TSMC has consistently focused on research and development to enhance its manufacturing processes. The company's commitment to innovation has resulted in significant advancements in semiconductor technology, enabling the production of cutting-edge chips for various applications, including smartphones, automotive electronics, and artificial intelligence.
High Chip Pricing at Small Nodes
Table 2 shows estimates of TSMC’s ASP (average selling price) by node, according to The Information Network’s report entitled Global Semiconductor Equipment: Markets, Market Shares and Market Forecasts.
In 2021, ASP was $10,775 for 7nm wafers and $14,104 for 5nm wafers. In 2023 during full production, 3nm wafers are priced at $19,865. I anticipate prices will decrease 5%/year following the start of full production.
The increased wafer price at smaller nodes, as TSMC continues to migrate to N3 in 2023 and N2 in 2025 is a strong tailwind for company revenue growth.
The Information Network
TSMC Client Base of Leading Tech Customers
TSMC's growth has been driven by its ability to attract and retain top-tier clients, including industry giants such as Apple, Qualcomm, and Nvidia. By offering highly advanced and reliable chip manufacturing capabilities, TSMC has become the preferred partner for many companies seeking to leverage the latest semiconductor technologies. Its market dominance is further evidenced by its high market share and robust financial performance.
Chart 2 shows TSMC’s revenues by customer for 2022. TSMC makes chips for more than 500 customers making nearly 13,000 different chips.
At the 5nm node, Apple is the largest customer of TSMC with about a 50% share, followed by Qualcomm at 25%, as NVDA, AMD $Advanced Micro Devices(AMD)$, and others make up the rest.
At the 7nm node, AMD (25%), Nvidia (20%), QCOM (10%), MediaTek (10%) are the top 7nm customers of TSMC.
Nvidia’s A100 is manufactured at the 7nm node and the H100 at the 5nm node by TSMC according to The Information Network’s report entitled Hot ICs: A Market Analysis of Artificial Intelligence (AI), 5G, Automotive, and Memory Chips.
Chart 2
The Information Network
Sanctions and their Impact
Given TSMC’s global reach and critical role in the semiconductor ecosystem, it has become a target of geopolitical tensions, resulting in trade sanctions. In recent years, the U.S. has imposed restrictions on certain Chinese technology companies, such as Huawei, due to national security concerns. It lost Huawei, a major client, in 2020 after the Chinese telecommunications-equipment giant was placed on a U.S. blacklist, yet sales still jumped 31% that year thanks to strong orders from Apple, Advanced Micro Devices and others.
Importantly, U.S. sanctions have imposed guardrails on any company receiving state funding under the U.S. CHIPS and Science Act will prohibit beneficiary companies from investing in their fabs located in mainland China. TSMC operates significant facilities in mainland China and is a prospective applicant for the U.S. funding.
While TSMC was given a one-year deadline of October 2023, with a possibility to extend their waivers for another year. We learned today from an article in the Wall Street Journal that the U.S. government is now willing to allow Samsung, SK Hynix, and TSMC to get American wafer fab tools for their semiconductor production facilities in China beyond October, 2024.
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.
The Information Network
Investor Takeaway
TSMC, the leading semiconductor foundry, is set to significantly expand its chip on wafer on substrate (CoWoS) advanced packaging capacity by the end of 2024. According to DigiTimes, TSMC plans to increase its CoWoS capacity from 8,000 wafers per month to 11,000 wafers per month by the end of this year and aims to reach around 20,000 wafers per month by the end of 2024. However, even with this expansion, it is anticipated that Nvidia will consume approximately half of the capacity available at that time. AMD is also reportedly seeking additional CoWoS capacity for next year.
The growing demand for technologies such as 5G, artificial intelligence (AI), and high-performance computing (HPC) is driving the adoption of complex multi-chiplet designs, exemplified by products like AMD's Instinct MI300 and Nvidia's H100.
Presently, TSMC has the capacity to process approximately 8,000 CoWoS wafers per month, with Nvidia and AMD utilizing about 70% to 80% of this capacity. Following them, Broadcom $Broadcom(AVGO)$ represents the third-largest user, accounting for around 10% of the available CoWoS wafer processing capacity. The remaining capacity is divided among 20 other fabless chip designers.
Based on the demand for TSMC for chips made for NVDA and AMD, Chart 3 shows the disconnect of share price for the three companies for the 3-year period. TSMC makes the chips for both companies, yet its share price growth is 66% of that of AMD and 24% of that of NVDA.
Chart 3
YCharts
According to Simply Wall St:
TSM ($106.8) is trading above our estimate of fair value ($56.5)
NVDA ($410.22) is trading above our estimate of fair value ($161.92)
AMD ($124.53) is trading above our estimate of fair value ($90.71)
Chart 4 shows P/E ratio for the three companies. AMD’s P/E ratio is 553.8x, ahead of NVDA at 224.2x and 16.6x for TSM. These compare to 15.3x for the overall U.S. market.
Chart 4
YCharts
Chart 5 shows Gross Profit Margin with TSMC outperforming NVDA and AMD despite its low share price growth.
Chart 5
YCharts
U.S. stock returns year-to-date have been dominated by mega-cap technology and AI-related stocks. Stocks that have exposure to generative AI and LLMs have seen their stock prices elevated by investors/traders to new 52-week highs and, in certain cases, all-time highs.
Needham downgraded several Semi-Cap stocks noting that they have been overbought due to the AI hype.
Needham noted that with only ~10% or less upside over the next 1 year, it does not think there is enough reward to justify buying these stocks given the risks which include — 1) P/E multiples contracting over the next 1-2 years from near peak levels; 2) WFE not reaching the firm's forecast of $110B by 2026; and 3) the companies failing to reach their target operating models by 2026.
As a result, Needham downgraded semi caps Applied Materials $Applied Materials(AMAT)$, KLA $KLA-Tencor(KLAC)$, Advanced Energy Industries $Advanced(AEIS)$, Onto Innovation $Onto Innovation Inc.(ONTO)$, Nova $Nova Measuring Instruments(NVMI)$ and Axcelis Technologies $Axcelis(ACLS)$ to Hold from Buy.
I disagree on KLAC. The company has nearly a 55% share in the metrology/inspection semiconductor equipment market. This equipment is demanded at smaller nodes because the high price of a wafer (Table 2) would be a significant loss to the semi manufacturer if defects or processing errors would reject the chips. In China, where KLAC has a large financial exposure, China's low yields demand these systems to increase yields.
TSMC is my #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 with a 60% share. I rate the company a Buy.
Source: Seeking Alpha
Comments