The approaching semiconductor capital expenditure supercycle
Due to growing complexity in chip designs, the semiconductor industry has witnessed increasing capital intensity. Driven by multiple factors, the sector could now face a CapEx supercycle.
Ever since the transistor and IC was invented (read Part 1 and Part 2 of my introduction to semiconductors article series) semiconductor chip designs have evolved and become increasingly advanced. However, as today’s manufacturing process nodes are approaching 3nm (nanometers) and lower, the industry is pushing the boundaries of complexity and innovation more than ever before.
Sam Altman, the CEO of OpenAI, has recently seeked trillions of dollars in investments to build semiconductor fabs dedicated to AI chip manufacturing. Regardless of how realistic his initiative is, it somehow highlights the strategic importance of semiconductor manufacturing in the age of artificial intelligence, and that the industry could be facing a significant capital expenditure (CapEx) cycle.
Another factor that likely will fuel the coming CapEx cycle is that governments worldwide have announced significant incentives to bolster their domestic semiconductor industries, recognizing its strategic importance.
The United States, for instance, passed the CHIPS Act, earmarking $76 billion to support semiconductor research, development, and manufacturing. Similarly, the European Union announced its $48 billion (€43 billion) European Chips Act, setting ambitious targets to increase the EU's share of the global semiconductor manufacturing market.
The semiconductor market size
In order to understand what is driving semiconductor CapEx, it is important to understand the underlying market and its size. Since 2001, the global semiconductor market has grown from $139 billion to a record $574 billion in 2022. This growth has been achieved through new market verticals such as smartphones, data centers, AI and also by an overall growth of semiconudctor content in many different product groups. However, due to rising interest rates and diminished consumer demand, the market contracted by 8% last year, to $527 billion.
Market research firms/organizations forecast that the global semiconductor market's value will surge to more than $1,100 billion by 2030 (some expect up to $1,400 billion). If the $1,100-1,400 billion predictions for 2030 hold true, the global semiconductor market could further grow to almost $1,500 billion by 2033, indicating a compound annual growth rate (CAGR) of 11% from 2023 to 2033.
This optimistic outlook for semiconductor market growth is shared beyond market research firms. During ASML’s capital markets day in late 2022, the company projected that the semiconductor market would grow to $1,098 billion by 2030. Their industry break-down can be seen in Chart 2 below.
ASML anticipated highest growth rates in segments such as data centers, automotive, and industrial electronics. For instance, the market for data centers (which also includes servers and storage) was projected to grow from $90 billion in 2022 to $249 billion by 2030. Importantly, these expectations were set before the rapid rise of the AI market.
Although it usually is wise to approach exponential growth charts with caution, it is crucial to understand how the surge in demand for generative AI has strengthened previous assumptions, like the one made by ASML back in 2022.
Just a year ago, market research firms predicted that the AI chip market would grow at a 30% CAGR, reaching around $230 billion by 2032. Currently, these predictions have been revised to $340 billion, or a CAGR of 35%.
Given that these revisions are based on the rapidly growing demand for generative AI, a global semiconductor market size close to $1,500 billion by 2033 appears to be a more realistic assumption today than it was one year ago. Just imagine how much computing power that will needed to generate actual video content in the future.
Semiconductor capital intensity
The expected growth of the global semiconductor market, as previously discussed, will not materialize without new investments, maintenance and replacements. Achieving such growth requires substantial increases in global manufacturing capacity, which in turn demands significant investments. Because of this, the market for semiconductor equipment could be on the brink of an unprecedented CapEx supercycle.
To forecast future semiconductor capital expenditures, let us examine the historical capital intensity of companies like TSMC, Micron Technology and SK Hynix, as well as the industry’s overall and average capital intensity.
For those curious about the term capital intensity, it is defined as the annual capital expenditure divided by annual sales. In 2023, TSMC (TSM) allocated $30 billion to capital expenditures, resulting in a capital intensity of 44%. Although this figure is below the +50% mark observed in 2021, it illustrates the increasing costs of producing today’s advanced chips. The following chart (Chart 4) detail the rise in TSMC's capital intensity since 2015.
Even though a decrease in capital intensity is forecasted for the next year or two, the company expects that its medium to long-term capital intensity will stabilize in the mid-30s. For the projections made in this article, TSMC's long-term capital intensity has been set at 37%, due to the increasing complexity of advanced chip manufacturing (and packaging).
Looking at memory manufacturers such as Micron Technology (MU) and SK Hynix (000660.KS), we see a pattern similar to what was observed with TSMC. For these firms combined, the average capital intensity has risen from 27% in 2017 to 35% in 2023 (as displayed in Chart 5). This increase aligns with the rapid adoption of more complex designs, such as DDR5 and high-bandwidth memory (HBM). Given this trend, projecting a long-term capital intensity of 35% for memory manufacturers seems plausible.
A decade ago, Micron Technology aimed for a capital intensity target of about 20-25%. And here we are, exceeding this target by a wide margin.
Below is Micron Technology’s roadmap for its AI memory portfolio and a visualization of its most recent HBM3E.
With TSMC, Intel, Samsung, Micron Technology, and SK Hynix accounting for a significant portion (approximately 70%) of the industry’s total capital expenditure, it’s reasonable to assume that these companies will determine the future average capital intensity across the sector.
What is evident is the notable increase in capital intensity within the industry since 2015, a trend likely propelled by the adoption of ASML’s EUV technology (Extreme Ultraviolet lithography). Chart 6 illustrates that the average capital intensity across the industry was approximately 18% in 2015. It has since increased to just under 30% by 2023.
Looking ahead, projecting a long-term industry capital intensity within the range of 30-32% seems justified, unless manufacturing complexity for some unknown reason would decrease going forward.
Now that we have explored the potential size of the semiconductor market 10 years from now (estimated to reach $1,500 billion), and considering the projected future industry capital intensity in the 30-32% range, we can now delve into what these figures imply for future capital expenditure.
The mother of all CapEx supercycles in the making
In 2022, global semiconductor capital expenditure reached a value of $180 billion. However, due to macroeconomic factors and a temporary dip in demand for manufacturing capacity, it is estimated that semiconductor CapEx fell to around $165-170 billion last year. Which actually was a bit higher than most expectations just 9 months ago.
And with the explosive demand growth for AI chips, coupled with consumer electronics and automotive sectors that likely will start recover soon, the semiconductor industry may very well be on the cusp of a CapEx supercycle. Perhaps the mother of all supercycles. However, it might not be until 2025 that we see any meaningful year-on-year CapEx growth, with 2024 expected to be more or less flat.
During the past 3-4 years, announcements have been made for semiconductor fab projects totaling $650 billion (including some long-term projects). Several of these projects are scheduled to begin construction within the next 2 to 4 years.
Assuming that the total semiconductor market value will hit $1,500 billion by 2033 (which has been discussed above), the annual investment required to maintain, replace, and expand global manufacturing capacity could very well surpass $460 billion. This would imply an average industry capital intensity of 31%.
Please note that chart 7 below does not visualize annual CapEx spend as it will occur. Instead, it illustrates the growth trajectory corresponding to a 11% CAGR.
CapEx supercycle winners
So, which companies stand to gain from this potential CapEx supercycle? Clearly, most semiconductor manufacturing equipment suppliers will become winners. This includes major companies like ASML, Applied Materials, Lam Research, KLA and Tokyo Electron.
Taking ASML as an example, the company aims for its annual sales to hit between $47 and $65 billion by 2030. With last year's sales at $30 billion, this sets a target for a 7-year CAGR of between 7% and 12%.
Assuming that the semiconductor market reaches a value of $1,100 billion by 2030, with an industry capital intensity of around 31%, and ASML capturing a 16% market share of the annual semiconductor CapEx, this would equate to $55 billion in sales, in line with the company’s financial targets. Therefore, ASML could be looking at an organic CAGR of 9% from 2023 to 2030. Pretty impressive for a company of ASML’s size.
ASML’s CAGR could of course exceed 9%, if the company continues to gain shares of the CapEx market with the introduction of their upcoming High-NA EUV tool priced at more than $350 million each. Conversely, it may also be lower than 9%, if for some reason alternative technologies outpace EUV adoption over the next decade.
Additionally, it is important to note that the current market valuation of these companies already suggests that they will become winners in the coming market cycle. ASML, Applied Materials, Lam Research, KLA and Tokyo Electron all trades at NTM EV/EBIT (next twelve months EV/EBIT) multiples well above 20x, with ASML topping at 41x.
