The global semiconductor market is expected to hit bottom in 2023 and begin a comprehensive recovery starting in 2024.

From 2021 to 2022, the global semiconductor market grew rapidly due to the special demand triggered by the outbreak of the COVID-19 pandemic. However, as the special demand caused by the pandemic ended in the second half of 2022, the semiconductor industry fell into one of the most severe recessions in its history in 2023. Nevertheless, the great recession is expected to bottom out in 2023, with a full recovery anticipated this year (2024).

The author believes that the comprehensive recovery of the semiconductor market depends on the trend of memory devices.

In fact, looking at the quarterly semiconductor shipment volumes of various types, logic chips have already surpassed the peak demand caused by the COVID-19 pandemic and have set a new historical high. Moreover, Mos Micro (microprocessors) and Analog (analog devices) are likely to set historical highs in 2024, as the decline due to the end of the COVID-19 special demand is not significant.

Advertisement

Among them, Mos Memory bottomed out in the first quarter of 2023 (Q1) after a significant drop and is recovering, but it seems to still require a considerable amount of time to reach the peak demand post-pandemic. However, if Mos Memory exceeds this peak, the total semiconductor shipment volume will definitely set a new historical high. In that case, it can be said that the semiconductor market has fully recovered.

However, looking at the changes in semiconductor shipments, it is clear that this view is incorrect. This is because, although the shipments of Mos Memory have basically recovered, the logic chips that have set a historical high in shipment value are still at an extremely low level. In other words, for the global semiconductor market to fully recover, the shipment volume of logic chips must increase significantly.

Therefore, in this article, the author will analyze the shipment volumes and quantities of various semiconductors and the total semiconductor volume. Next, the author will use the difference between the shipment volume and quantity of logic chips as an example to demonstrate how TSMC's shipments are rapidly recovering, yet the wafer shipments remain sluggish.Further speculation is made on why such differences may arise, and it is discussed that the possibility of a comprehensive recovery of the global semiconductor market is likely to be postponed until 2025.

The conclusion is that the semiconductor market appears to be recovering now, which is a "illusion" caused by Nvidia GPUs being traded at very high prices. Therefore, it seems that the semiconductor market will not fully recover until foundries such as TSMC reach full capacity and the shipment of logic chips sets a new historical high.

Analysis of Semiconductor Shipment Value and Quantity

Trends in the shipment value and quantity of various types of semiconductors and the entire semiconductor industry.

The shipment value of Mos Micro peaked in Q4 2021, reached a bottom in Q1 2023, and is now recovering. On the other hand, the number of shipments has not changed significantly, remaining relatively flat with a slight decrease from Q3 to Q4 2023.

The shipment value of Mos Memory dropped significantly from Q2 2022, bottomed out in Q1 2023, and turned upward, but only recovered to about 40% of the peak value in Q4 of the same year. On the other hand, the number of shipments recovered to 94% of the peak. That is to say, it can be considered that the factory operation rate of memory manufacturers is close to full capacity. The question is how much the prices of DRAM and NAND flash memory will rise.

The shipment value of logic chips peaked in Q2 2022, bottomed out in Q1 2023, and began to rise, setting a new historical high in Q4 of the same year. On the other hand, the number of shipments peaked in Q2 2022 and dropped to about 65% of the peak value in Q3 2023, remaining flat in Q4. In other words, there is a significant difference in the behavior between the shipment value and the number of units shipped in logic chips.

The shipment value of analog devices peaked in Q3 2022, bottomed out in Q2 2023, and then remained flat. On the other hand, the number of shipments has been on a continuous downward trend since peaking in Q3 2022, until Q4 2023.Finally, after a significant decline in the overall semiconductor shipment volume from Q2 2022, it bottomed out in Q1 2023 and turned upward, recovering to about 96% of its peak level by Q4 of the same year. On the other hand, the number of shipments plummeted from Q2 2022, bottomed out in Q1 2023, but has since remained flat at about 75% of its peak.

From the above, if one looks solely at shipment volumes, Mos Memory is the issue, currently only recovering to about 40% of its peak. However, when viewed from a broader perspective, it can be seen that logic chips are a major problem, with shipments reaching an all-time high, yet the volume of shipments has stagnated at around 65% of the peak. The impact of this discrepancy between the shipment volume and the number of shipments of logic chips seems to extend to the entire semiconductor sector.

In short, the recovery of the global semiconductor market hinges on the price increase of Mos Memory and whether there will be a significant increase in the shipment volume of logic chips. Moreover, with DRAM and NAND prices on the rise, the biggest issue is the increase in the number of logic chip shipments.

Below, we will explain the behavior of TSMC's shipment volumes and the number of wafers shipped to specifically illustrate the difference between the shipment volume and the number of wafers shipped for logic chips.

TSMC's Quarterly Shipment Value and Wafer Shipment Quantity

TSMC positions 7nm and beyond as advanced nodes, with 7nm accounting for 17%, 5nm for 35%, and 3nm for 15% in Q4 2023, bringing the total for advanced nodes to 67%. Additionally, the quarterly sales of advanced nodes will grow from the first quarter of 2021, drop once from the fourth quarter of 2022, but bottom out in the second quarter of 2023 and start to rise again, reaching an all-time high in the fourth quarter of the same year.

The fluctuations in TSMC's quarterly shipment volumes and the number of chips shipped per quarter are roughly consistent. They peaked with the IT bubble in 2000, fell after the financial crisis in 2008, and also fell after the memory bubble burst in 2018.However, behavior after the peak of special demand for the epidemic in the third quarter of 2022 was different. Shipments reached a high of $20.2 billion, then dropped sharply, but began to rebound after bottoming out at $15.7 billion in the second quarter of 2023, recovering to $19.7 billion in the fourth quarter of the same year, which is 97% of the peak.

On the other hand, the chip shipments this quarter reached a peak of 3.97 million pieces in Q3 2022, then dropped sharply, reaching a peak of 2.92 million pieces in Q2 2023, but remained flat thereafter, also maintaining at 2.96 million pieces in Q4 of the same year, which is about 1 million pieces less than at the peak.

Among the semiconductors produced by TSMC, logic chips are the most. TSMC's sales in the fourth quarter of 2023 set a historical high for the front-end node, and overall, it recovered to 97% of the historical high. Despite this, the chip shipments in that quarter were more than 1 million pieces less than at the peak. That is to say, TSMC's overall factory operation rate is only about 75%.

As for the global semiconductor as a whole, the shipments of logic chips have declined to about 65% of the special demand peak of the COVID-19 period. Consistent with this, TSMC's quarterly wafer shipments have decreased by more than 1 million wafers from the peak period, and the factory operation rate is considered to be low at around 75%.

In the future, for the global semiconductor market to truly recover, the shipments of logic chips need to increase significantly, and to achieve this, the operation rate of foundries led by TSMC must be close to full capacity.

So when exactly?

Forecasting the Utilization Rate of Major Foundries

On December 14, 2023, the TrendForce seminar "Industry Focus Information" was held at the Tokyo Bay Ariake Washington Hotel. At the seminar, TrendForce analyst Joanna Chiao introduced "TSMC's Global Strategy and Outlook for the 2024 Semiconductor Foundry Market". In it, Joanna Chiao talked about forecasting the utilization rate of foundries.According to the forecast, by the second quarter of 2022, the utilization rate of all wafer foundries will exceed 90%. However, during the fourth quarter of the same year, the utilization rates of many wafer foundries will drop sharply.

Among them, only TSMC (Taiwan Semiconductor Manufacturing Company) will maintain its leading position with a utilization rate of 99%, but it will drop to 76% by the fourth quarter of 2023. This TSMC operation rate is roughly consistent with the operation rate of about 75% derived from the number of chip shipments per quarter.

Moreover, it is expected that the operation rates of major wafer foundries, including TSMC, will not reach full capacity even by the fourth quarter of 2024. Even the most efficient TSMC will only have an operation rate of 85%. Samsung Electronics, the world's second-largest wafer foundry company by sales (hereinafter referred to as "Samsung"), is expected to have a low utilization rate of 67%.

That is to say, according to this forecast, the operation rates of major wafer foundries will not reach full capacity before the end of 2024. Therefore, it is expected that by 2024, the shipment volume of logic chips will not increase significantly, and as a result, the global semiconductor market will not truly recover.

So, why are the operation rates of major wafer foundries, led by TSMC, so low?

Year-over-year growth rates of various types of electronic device shipments

Smartphone growth rate decreased by 10.6% in 2022, and in 2023 it continued to decline by 2.7%, turning positive in 2024 but only increasing by 3.0%. Server sales grew by 4.7% in 2022, decreased by 6.0% in 2023, and recovered to positive growth in 2024, but the increase was only 2.5%. Laptops will drop significantly to -24.5% in 2022, remain at -10.2% in 2023, and finally recover to +3.2% in 2024.

In addition to these three types, the growth rates for 2024 are also disappointing, with tablets decreasing by 0.1%, televisions by 0.3%, and monitors by 1.5%.

In this context, the high growth rates of AI servers and electric vehicles (EVs) are noteworthy. Specifically, AI servers were at zero in 2022, but grew by 36.9% in 2023 and 44.2% in 2024. These AI servers use AI semiconductors, and Nvidia's GPUs account for more than 80% of these AI semiconductors.Reasons for Discrepancy Between Logic Chip Shipment Value and Quantity

In Nvidia's GPU lineup, the H100 is currently the highest-performing GPU with a high demand, and it is transacted at an extremely high price of 5 to 6 million yen per unit. This is considered the reason for the new high in logic chip shipment value.

However, as shown in Figure 6, the growth rate of many electronic devices, except for AI servers and electric vehicles, is extremely slow. Therefore, the shipment of logic chips used in these devices will be at a low level. Consequently, the utilization rate of the main foundries producing logic chips will also be low.

So, how many AI semiconductors like Nvidia's GPUs will be shipped in the future? How much is the global logic chip shipment expected to increase? In other words, will AI semiconductors become the savior of market recovery?

Changes in AI Server Shipments

According to the forecast, the shipment of AI servers in 2022 is 877,000 units, which is expected to grow by 36.9% in 2023, exceeding 1.2 million units, and further increase by 44.2% in 2024, exceeding 1.7 million units. It is projected to ship about 2.7 million units by 2026.

The proportion of AI servers in the total number is expected to be 6% in 2022, 9% in 2023, 13% in 2024, 14% in 2025, and 16% in 2026.The proliferation of generative AI like ChatGPT is exploding globally. Consequently, the demand for AI servers appears to be substantial. However, in reality, the shipment volume of AI servers is not as large as one might think.

AI Server Bottlenecks

It can be assumed that the reason lies in the supply speed of AI semiconductors. Currently, NVIDIA holds about 80% of the AI semiconductor GPU market, which is processed through front-end and back-end operations at TSMC. In the subsequent processes, a packaging technique called CoWoS is employed, but the capacity of this CoWoS has become a bottleneck.

Additionally, within the CoWoS, HBM (High Bandwidth Memory), which consists of multiple stacked DRAM layers around the GPU, is also considered one of the bottlenecks. This is because the main supplier of HBM to NVIDIA is SK Hynix, while the DRAM king, Samsung, has fallen behind.

Thus, not only is there a shortage of CoWoS capacity, but it is also hard to say that HBM is being supplied adequately. Furthermore, the lead time for CoWoS is about half a year, and another half year is required for the assembly of AI servers. In short, it takes approximately one year to ship AI servers.

Moreover, in the right-hand side chart of Figure 8, it is indicated that the proportion of AI semiconductor chips in the world's advanced process technology (possibly beyond 7nm) is quite small. AI semiconductor chips accounted for only 4% in 2022 and are expected to double to 8% by 2026.

When will the shipment volume of logic chips increase?Is this 8% significant or not? Although it's a nuanced question, even by 2026, the remaining 92% of wafers will be consumed by non-AI semiconductor chips. The majority of these are logic chips. Therefore, for the shipment of logic chips to increase, and for major foundries led by TSMC to operate at full capacity, the demand for electronic devices such as smartphones, PCs, and servers must rise.

In summary, AI semiconductors, represented by NVIDIA GPUs, will not be the savior. Consequently, it is difficult for the global semiconductor market to truly recover in 2024 and is expected to be delayed until next year, 2025.

However, there is another (positive) possibility to reverse this forecast.

The AI semiconductors discussed so far imply those installed entirely on servers. But now, there is a movement towards AI processing on the edge, such as on PCs, smartphones, and tablets.

For instance, AI PCs advocated by Intel and AI smartphones that Samsung is about to manufacture. If they become explosively popular (i.e., innovative), the AI semiconductor market will rapidly expand. In fact, the American research firm Gartner estimates that by the end of 2024, 240 million AI smartphones and 54.5 million AI PCs will be shipped. If this forecast materializes, the demand for cutting-edge logic chips (both in terms of shipment volume and the number of units shipped) will also increase, and the utilization rate of wafer foundries like TSMC will rise. Moreover, the demand for MPUs and memory will inevitably expand dramatically.

In other words, if such a world comes to pass, AI semiconductors will become the true savior. Therefore, the author will focus on the developments in edge AI semiconductors in the future.

*Disclaimer: This article is the original creation of the author. The content of the article represents their personal views, and our reposting is solely for sharing and discussion, not an endorsement or agreement. If you have any objections, please contact the backend.