The global semiconductor market, which grew by more than 25% in 2021, has suffered a sharp slowdown this year, with an increase in turnover close to 4%, the same percentage that is expected for next year but with a negative sign. . In this way, the world market will once again bill 600,000 million dollars in 2023, as in 2021, far from what was expected: the vast majority of consultants estimated that billing would grow until the end of the decade at the same rate as in 2021 and it would reach a trillion dollars by 2030, a bar that now seems difficult to achieve.
In the last half year, the situation of the global semiconductor market (or chips, as they are also known) has taken a spectacular turn: from significant growth, which followed in the wake of 2021, it has gone from the summer to a sharp stop, which has meant that the global turnover of semiconductors for all of 2022 is expected to be 618,000 million dollars, with a 4% increase, according to the estimate of the end of November by the consulting firm Gartner
The most serious thing, however, is that the same consultancy forecasts a drop of 3.6% for 2023, with which the turnover will be 596,000 million dollars, approximately the same as in 2021, after growing no less than the 26.3%. At the end of this July, Gartner calculated that the global turnover of semiconductors would grow 7% in 2022 and already projected a drop of 2.5% for 2023, much lower than the increase of 13.6% for 2022 that it estimated at the beginning of this year. year. This gives an idea of the rapid deterioration of the chip market that has occurred throughout 2022 and the poor prospects for next year. And neither is it very optimistic for 2024, since growth is expected, at most, in a very low digit.
“Globalization and free trade in semiconductors is all but dead” and unlikely to return, TSMC founder Morris Chang said as he laid the cornerstone for his Arizona chip plant.
The consulting firm Gartner has not been the only one that has totally erred in its forecasts of the evolution of the semiconductor industry. The World Semiconductor Trade Statistics (WSTS), the association that brings together the vast majority of chip manufacturers from around the world, estimates that the world market will grow 4.4% in 2022, standing at 580,000 million dollars, to fall the 4.1% in 2023 and thus reach 557,000 million dollars, in its forecast of November 29, in line with Gartner.
The lower graph of the WSTS shows that the global turnover of semiconductors has always fluctuated, but not as pronounced as up to now, with the exception of the 2019 crisis, where it went from around 350,000 million dollars towards 2017 to about 450,000 million two years later, to situate itself for three long years at the level of 400,000 million dollars. In 2021, after relative control of the pandemic, it rose to a maximum of 600,000 million dollars in the first quarter of this year, followed by the now-expected drop. The WSTS forecast is that revenues will continue to fall during the first half of 2023 and then start a slow recovery (grayed in the graph).
The consultancy IC Insights, very specialized in the sector and also very prestigious, makes forecasts similar to those of Gartner and WSTS. In his November 28 newsletter, he forecasts that the semiconductor market will increase 3% in 2022, compared to a 25% increase the previous year, and a 3% drop for next year. The estimated annual turnover varies slightly in the three consultancies, probably due to differences in the calculation method, although the evolution is practically the same. For IC Insights, the global turnover of 2023 will be 491,400 million dollars, and it will not reach the one that was in 2021 of 510,500 million, especially since the market for integrated circuits, the very majority within semiconductors, will fall much more, as seen in the graph below.
Although global semiconductor turnover has grown by an average of 3% this year, the breakdown of the forecast by different product categories shows a very different variation, from an increase of 16% for sensors and actuators, 17% for components analog or 13% of discrete components compared to the 17% drop in memories, which account for almost a quarter of all semiconductors. Integrated circuits (IC in the graph) are expected to increase on average 3% in 2022, like the set of semiconductors in round numbers, but the turnover of OSDs (optoelectronic components, sensors and actuators and discrete components) will grow on 8 % (additionally slowed down by stagnation of the optoelectronic components). The sharp drop in memories is partly offset by the increase in logic circuits and microcomponents, as seen in the graph below.
In the WSTS spring forecast you can see the different weight of the different types of semiconductors, whose broken down turnover in 2021 is correct (the data for 2022 and 2023 have been corrected very downwards). Integrated circuits represented 83% of all semiconductors in 2021, while discrete, optoelectronic and sensor components shared the remaining 17%. Within integrated circuits, memories and logic circuits accounted for about 33% each, and microelectronic and analog components another third, as seen in the table below.
The small increase for 2022 means that this proportion has changed little, except in the reports. As seen in the Spring 2022 WSTS table, the full-year forecasts have turned out to be dead wrong, with an expected 16% increase going to 3% in nine months. Still, the WSTS Spring 2021 table breakdown is illustrative. It can be seen, for example, that the American semiconductor market represented 22% of the total in 2021, compared to 62% in Asia or 14% in Europe and a similar proportion in Japan. By 2022, WSTS calculated that the European and American markets would grow slightly more than the Asian and Japanese, but by 2023 a similar increase was projected in all three regions.
From exultant growth to stagnation
That it has gone, in just over half a year, from exultant growth forecasts for the semiconductor market to its virtual stagnation, which may continue for another two years, may seem strange, but it is justified, as often happens when looking at past bull. First of all, the exposed figures refer to the turnover, not to the sale of semiconductors in units, which foreseeably have continued to grow at a good pace.
The main reason for the slowdown in global billing as 2022 progressed must be attributed, mainly, to the sharp drop in memories (DRAM and NAND flash types), of 17% throughout 2022, according to IC Insights, while in the spring the WSTS attributed to them a growth of 18.7%. They are 26,000 million dollars less compared to the total billing of last year, to which could be added another 29,000 million dollars foreseen by WSTS in the section of reports. A difference of 50,000 million in reports explains, almost entirely, why WSTS’s spring growth projection of 16.3% of total billing has gone down to only 4.4% in its final annual forecast. of November.
Another important reason is that, in periods of high sales growth, one tends to think that this increase will continue, if not indefinitely, then for many quarters (in times of decline, the opposite occurs, one tends to be much more pessimistic). It was thought, in mid-2021, that sales of personal computers, laptops and servers would remain high throughout 2022 and at a high unit price as in 2021 due to lack of components, when the opposite has occurred. In the first half of 2022, computer sales have fallen 10.1% (-5.1% in the first quarter and -15% in the second), estimates IDC, and Gartner calculated in October that they fell by 19.5% the third trimester.
Within four years there may be another peak in demand for chips, but it is not clear that it will be enough to cover the productive overcapacity that will then exist, with plant investments underway.
According to the IDC consultancy, 350 million computers were sold worldwide in 2021 and the current projection for 2022 is that sales will be around 305 million units, a drop of 13%. For all of 2023, if computer sales reached 290 million, it would already be a success. In the case of smartphones, another device that uses a lot of memory and sells five times more units than computers, it is estimated that this year they will drop by around 7%.
The bulging drop in sales that is expected to occur this year for smartphones and computers was not foreseeable until well into the year. It has been this fall, to a large extent, that has caused the drop in billing in memories, apart from the fact that there is less demand for units, it has caused its unit price to plummet. The consequence of both factors explains, although it does not necessarily justify, that all the consultants erred their forecasts half a year ago for all of 2022.
Another factor should be added, derived from the supply problems of some types of semiconductors during 2021 and part of 2022 (some car manufacturers are still waiting for their orders to be served and their plants are at half throttle). As the final customers anticipated that there would be problems with the supply of components, they placed orders that were higher than necessary, to see if they would receive all or a large part of what they needed.
This caused an unrealistic increase in demand and, although manufacturers were already counting on this effect of increasing orders, it masked, or at least delayed, the real scope of the strong semiconductor demand crisis in early 2022. Throughout 2023 the opposite may occur: that the demand is somewhat higher than expected and there is not enough production to meet it, as happened a couple of years ago due to the pandemic. The increase in inflation around the world and the evolution of the price of the different currencies can further complicate the market forecast for the coming year.
Foreseeable excess chip production in four years
Within four years, however, it is likely that there will be another peak in demand, another mountain like the ones that emerged in 2019 and 2022, as the WSTS historical graph above makes clear, and some 700,000 will be billed by 2026. millions of dollars. If so, the world semiconductor market would have doubled between 2016 and 2026, from around 350,000 to 700,000 million dollars, which would not be bad, although below the optimistic forecasts for 2021.
It should be borne in mind that, throughout the second half of this decade, semiconductor production capacity will be much higher than it is today (some experts estimate that there will be twice as much capacity within four years), taking into account the massive investments being made by TSMC, Samsung and Intel, as well as the rush to become more self-sufficient in which the United States, Europe, Japan, India, South Korea, Taiwan and, of course, China are embarking.
If a global agreement is not reached to rationalize future investments and specialization in semiconductor production plants, which now looks unlikely, it is possible that within two or three years we will enter a long cycle of semiconductor production overcapacity. , which would cause a drop in the price of chips and serious difficulties in amortizing the investments made. The supply and demand cycles in semiconductors have never coincided in their more than forty-year history; What happens is that now the market is much higher and the chip manufacturing plants cost much more, making it more difficult to amortize the investment if the market does not evolve minimally as expected.
A few days ago, the solemn ceremony for the construction of a TSMC plant took place in Phoenix, in the US state of Arizona, with the assistance of President Joe Biden, the TSMC staff and various top managers of American companies that will place orders with the plant, such as Apple, NVidia, Qualcomm and AMD. Morris Chang, the founder of TSMC more than forty years ago and of the then pioneering concept of manufacturing chips for third parties, which has brought so many benefits to the Taiwanese company (after several decades of intense effort), warned that “globalization and free The semiconductor trade is almost dead” and is unlikely to return.
With this resounding statement, Chang alluded to the immense benefits of breaking down the semiconductor manufacturing chain into different parts and regions of the world, each mastering a specialty. Thus, the United States is strong in chip design, Europe in state-of-the-art lithography equipment, Japan in the production of the necessary materials, Taiwan in the manufacture of highly sophisticated custom chips, China in the mass production of less sophisticated chips that those in Taiwan and component assembly, and Southeast Asia in chip cutting and verification. This specialization in the chip production chain, although mainly concentrated in Asia, has encouraged the development of increasingly innovative, sophisticated and economical chips.
In an editorial, The Washington Post summarized the state of the semiconductor manufacturing ecosystem in the accompanying graphic: The United States controls about half of semiconductor design and intellectual property, but most chips are made in Asia, and The United States hardly produces any materials to make them, which are made in Europe and Asia. Chip assembly, verification and potting is also an Asian specialty.
According to TSMC, the cost of manufacturing a chip in its Arizona plant is between 40 and 50% higher than the equivalent in Taiwan, where the company concentrates the bulk of its plants and, of course, the most advanced. Aside from the increased cost, TSMC has found itself short of specialized chip manufacturing personnel, so it now has several hundred people training in Taiwan for a year or year and a half, by the time the Arizona plant is up and running. The company has also had a lot of trouble finding American contractors for the plant’s civil works, even though much of the equipment and interior enclosures will come assembled from Taiwan.
The higher cost of the chips will be borne by American consumers, although they will take the consolation that some of the chips will have been sunk on American soil by a wholly Taiwanese company. TSMC’s Phoenix plant is scheduled to start semiconductor production in 2024, using 5-nanometer design rules. At the opening ceremony, Morris Chang assured that there are plans to build a second plant, with three-nanometer rules, but clarified that “they are not finalized yet”, contradicting what the company statement says, which takes it for granted. The production of chips in this second plant would begin in 2026 and the joint investment would amount to 40,000 million dollars.
High-stakes chip diplomacy
It was in May 2020 when TSMC announced its intention to build a new plant in Arizona, with an investment of 12,000 million dollars, after a year of studying possible locations. Now, thanks to the generous subsidies allowed by the legislation approved last summer, the Chips and Science Act, this investment could triple with the second plant, and production could go from 20,000 wafers per month to 60,000 wafers per month, about 600,000 per year. This production of 600,000 wafers per year, White House spokesmen say, would be enough to cover all the needs of advanced chips in the United States.
The annual capacity of all the plants that TSMC operates is currently about 15 million 12-inch wafer equivalents, so the two Arizona plants, when completed, would represent about 4% of TSMC’s global production. Of course, not all of TSMC’s production is as advanced as planned for 2024 at the US plant, but the company has been making wafers with 5-nanometer design rules for more than a year now and will be making wafers with 5-nanometer design rules next year. three nanometers, which would not start in the United States, assuming the second plant is built, until 2026. By then, TSMC would already be two generations ahead.
Semiconductor manufacturing has not only ceased to be globalized and subject to free trade laws, as Morris Chang believes, but has also entered a highly protectionist and politicized phase, which experts consider highly dangerous and the European Commission is very worried. Chris Miller, who has recently published the book Chip War, for which he has just received the Financial Times Business Book of the Year award, is concerned about the “Taiwan risk”, the possibility of a confrontation between the United States and China and the implications for global trade and chip development and supply.
For Miller, the “fundamental problem” for US politicians is that China is both America’s number one customer and number one competitor, as he puts it in his book. And in an interview published this Monday in the Financial Times, Miller assures that the main challenge for the Biden Administration in chip diplomacy is to “convince allies that the balance between security and economic considerations can be maintained.” ”.
It is possible, Miller thinks, that the risks of economic “self-immolation” could make peace between China and Taiwan and the US-China peace-keeping, “but he doesn’t see much reason to be very confident” in it. What is certain is that chips will be high on the global political agenda in 2023 and that these tiny bits of silicon will continue to be talked about a lot.