Lessons from History: The 1980s Semiconductor Cycle(s)
Japan and overcapacity. A new series on cycles to gain insight into today's semiconductor market.
I have been diving into the history of semiconductor cycles. Given that the biggest debate today for investors is where exactly we are in the cycle I thought that it would be worth it to get some context of the past. In my personal experience looking back is always a good idea when the future is uncertain.
What I found is an industry that’s totally different than the semiconductor industry of today, but I still think there are a lot of lessons to learn.
This series is an attempt to try to characterize what happened and give context to investors today. Think of it as a semiconductor investor’s almanac. I wish I could go back further into the 1970s, but that is just outside of the sources I had. Instead, I should start at 1980 and walk through one of the most pivotal decades in the industry’s history. The decade pivots around one country: Japan.
1980-1983 - Memory Price Cycle
Let’s start with 1980. I think this is a good place to start because despite the big cycle most investors know about (1985), 1980 had a cycle onto itself. From the Intel 1981 Annual report.
“Revenues grew by 29% from 1979 to 1980 but decreased by 8% from 1980 to 1981 in spite of an increase in unit shipments during this latter period. Typically semiconductor components decline in both sales value and cost to manufacture. During 1979 and the first half of 1980 sales prices remained uncharacteristically level while cost to manufacture declined. This price stability was due to an excess of demand over industry capacity. Commencing in the second half of 1980 and throughout 1981 prices have declined at unprecedented rates. These rapid price reductions are primarily related to industry overcapacity which occurred as a result of industry-wide capital expansion programs and the effects of recessionary economic conditions throughout the world. The current industry outlook is for a continuation of existing conditions well into 1982, at a minimum.”
Prices stayed flat in 1979-1980 which of course drew a meaningful amount of investment and then when capacity was added at a rapid rate the entire cycle flipped.
But note despite knowing this, Intel was not immune to the capital additions. Despite 1981 being a brutal down year in the market it was the highest capital spending Intel had ever pursued as a company. This is clearly “different” today, when Capex is much more reactive to cycles and cycles are fewer and less intense.
Intel at the time was mostly a memory company and that boom-bust cycle is definitely shown here. If you recall - “The Innovators Dilemma” was written focused on memory companies, because of the fruit-fly-like cycles of competition and destruction. That same intensity and pace applies for Intel in the 1980s, and despite overcapacity, companies invested in the fear of losing market share. 1982 would be another brutal year - partially driven by a recession. Some context is below.
The 1982 recession was one of the worst recessionary periods pre-2007. What drove this recession is something readers today can appreciate today, inflation. Shortly after Volcker’s appointment to the chairman of the Fed in 1979, Volcker pursued a tight monetary policy and flipped the business cycle. Unemployment peaked at 11% in 1982 and was the worst post-WW2 recession the country had seen to this point. This context makes the 1982 semiconductor cycle make sense. But while 1983 had better economic strength to come, the real threat to come was Japan.
Quietly in this period, Japan surpassed the United States in global market share in memory during 1982, and this trend would only accelerate. The real driver here was manufacturing yields, as American semiconductor companies were shocked to learn from HP that Japanese companies consistently had better quality and yield that American chip companies. Not only were they beating them to the punch in process, but in quality. That’s hard to compete then.
1982 was the year the fateful cross in market share happened, and the United States would never truly recover from this leadership loss for manufacturing. The United States peaked its global manufacturing share somewhere in the 1980s. The baton for manufacturing was then handed to Taiwan from Japan in the late 1990s.
More context to consider is the varying economies because compared to the United State’s inflationary concerns, Japan’s rising sun was at it’s zenith. Inflation was not crushing the Japanese economy, and Japanese companies buzzed strongly ahead as they sold chips to both a booming domestic economy and exported chips abroad.
Now back to the financials. National Semiconductor’s 1983 Annual Report (fiscal year ending March 31) paints a brutal picture for 1982. However, a recovery was widely expected by the end of 1982. And that it did! Take a careful look at 1983 (FY 1982 is effectively 1983) because I want to highlight something that was truly odd about this cycle. Revenue grew but losses increased, and this was mostly driven by unit growth against ASP declines.
Both 1981 and 1982 (remember March 31 fiscal year) had higher units, but lower prices, which lead to extending net losses. Also notable is that despite the economic brutality, National would continue to invest at higher rates of capex despite declining profitability.
The outlook was strong at the point the market was a lot smaller and would grow meaningfully faster. The end of the annual report pointed to better times - which would come. Let’s talk about the brief boom of 1984 and the subsequent misery of 1985.
The Good Times Never Last (1984-1985) - Japan Rising
1984 was a strong year, with the entire industry growing at meaningful rates. But 1985 was horrendous. This was the worst year the industry had ever faced. What happened?
Let’s just begin this section with a blurb from a 20-year overview from the 1988 Intel Annual report. I love the phrasing.
“Intel would probably as soon erase 1985 from the history books were it not for the introduction of the 386 microprocessor. No longer producing DRAMs, struggling to keep its lead in EPROMs, and in the midst of the worst downturn in history, Intel needed a winner. The 32-bit 386 microprocessor turned to be just that winner later in the decade.”
1985 was a traumatic moment for Intel and the semiconductor industry. Intel had one of the largest layoffs in its history. National Semi had a 17% decrease in revenue but moved from an operating profit of $59 million to an operating loss of -$117 million. Even Texas Instruments had a brutal period of layoffs, as revenue shrank 14% and profits went negative. 1985 was the most painful year ever in the semiconductor industry, and the reason was not the economy but Japan.
See this is the period of Japan’s superior yields (better variable costs) and conglomerates. I think one of the best explanations of the Zeitgeist of the time is this amazing article in the Washington Post.
First, it details that American companies are clearly losing to Japan, but the reasoning was fantastic as it made sense at the time, but would later become the downfall of Japanese companies. The reason given was being too narrow:
"I think it is the American companies that decide what they have to do," said Kenichi Ohmae, a Tokyo-based McKinsey & Co. consultant who works with many of Japan's leading high-technology companies. "American companies, except for a few like IBM, do not develop into a stable shape. The portfolio of the business is typically too narrow. . . ."
Japanese businesses were fully integrated, with better yield and producing better products than Intel and American companies could at the time. This led to full margin economics that American companies could not compete with. This is the logic of IDMs, but also the logic of Apple today. We likely will see more vertical semiconductor to system providers in the future.
Another excerpt from the Washington Post article felt damning:
Unlike the American Intel or a National Semiconductor, Japanese chip companies are wholly owned divisions of giant Japanese electronics companies. The multibillion-dollar Hitachi, Fujitsu and Nippon Electric Corp. are all vertically integrated electronics conglomerates that do everything from manufacturing their own chips to building their own computer and telecommunications systems to global marketing and distribution of their products.
"Japanese companies, because they are diversified into many [related] businesses, can absorb [the] peaks and valleys. American companies are not so well diversified into other areas; therefore, they have to take hard decisions. Japanese companies can make much softer decisions. They say, 'OK, bad times come, we slow down production and we absorb the losses with other businesses. . .'," Ohmae said.
This led to a complete capitulation from the West. Intel at the time started to license IP to Japan in return for process know-how, and the focus was from R&D to process. This was almost complete capitulation from the industry.
Indeed, Intel's Whittier said that Intel has started "shifting its internal resources from classic research and development to process engineering" and that "at least a third" of the old research and development budget is now devoted to manufacturing research.
Investing in the process while profits declined is a death spiral, and a race to the bottom in memory worried everyone. But the real breath of relief of course was industry working together with the government. As American semiconductor firms struggled, they rallied together to try to stop the dumping of products that Japanese firms were doing, especially in commodities like DRAM.
But before you think this was a sudden onset, Japan versus US tension was actually was a long time coming - as Japan had actively worked on taking IP for over a decade. Actions that would be illegal in the US were conducted for the consistent theft and copying of IP. The consistent financing of Japanese workers to American conferences and tracking US patents began in earnest in 1978. From “The Intel Trinity” and
Meanwhile, Japan’s Ministry of International Trade and Industry, in concert with Japan’s major banks (an arrangement illegal in the United States) helped the Japanese electronics companies embark on a massive program of tracking US patent filings in order to anticipate where they were going and beat them to the punch. The ministry provided subsidies (illegal under international law) to help Japanese companies sell their chips at artificially low prices in the United States while keeping prices high at home. In 1978, forty thousand Japanese citizens made technology-related visits, many of them subsidized, to the United States, while just five thousand American businesspeople traveled the other way.
Malone, Michael S.. Intel Trinity, The (p. 307). Harper Business. Kindle Edition.
Further, Japan had access to capital markets explicitly from the conglomerates and governments (VLSI) that US companies did not. That echoes today’s environment partially.
Japanese firms probably had easier access to capital: they are often affiliated with a large bank that could play a role in corporate governance through equity ownership (the Glass-Steagall Act prohibits such activities in the United States). Such bank ties probably allowed Japanese producers to weather industry downturns much better than their U.S. counterparts.* On the U.S. side, the high cost of capital in the early 1980s, the appreciation of the U.S. dollar, lagging adoption of new process technology, and quality control problems all hampered U.S. firms.
https://www.nber.org/system/files/chapters/c8717/c8717.pdf
Tensions were high, and the SIA and AEA, led by Robert Noyce pushed congress to try to stop Japan from dumping semiconductors on American markets. American companies could not compete on the relative access to capital combined with the dumping going on.
Prices collapsed as Japanese firms flooded the domestic and international markets, and memory prices contracted 60% in a year bankrupting and pushing everyone except Micron and Texas Instruments out of the memory business. Micron lead the charge with the anti-dumping complaints against Japanese firms, and National Semiconductor, Intel, and AMD followed suit.
Part of the problem was Japanese firm’s primary demand was domestic, and the self-consumption and attitude of “Buy Japan” attitudes created domestic profit pools, while Japan in turn was cut off to United States producers. The US industry could never find a smoking gun of anti-competitive action, but the entire situation was detrimental to both sides of the market.
This all cumulated in the Semiconductor trade agreement of 1986, in particular, the encouragement of American companies to be able to sell meaningfully into Japanese markets. This essentially encouraged a 20% market share in Japan for US firms, while limiting dumping from Japanese companies. This massively benefited the two remaining US firms in the DRAM market, Texas Instruments and Micron in the latter half of the 1980s.
The reality is that the government helped partially, opening the Japanese market and slowing the price race to the bottom, but additionally yield meaningfully improved by the end of the decade for most US companies. In a mix of both the businesses adapting to crisis and incrementally better regulation - the semiconductor companies of the US improved and adapted. This was the birth of “copy exact” at Intel for example.
This crisis would eventually pass and times would be plenty again. Intel’s 1988 results are indicative of what plenty looks like.
Times were once again good - and I think this is a good place to end this decade of review. 1989 and 1990 were both strong years in the market and compared to the earlier losses in market share Japan would start to cede market share to US firms, especially as the Japanese asset bubble burst in 1991.
Conclusions and Takeaways: It’s Always Been Geopolitical
Things that I find surprising I guess are that Semiconductors in particular seem to have always been struggles between countries’ domestic industries. Japan and the US were the first round of fights in this long saga, but next time we will discuss Taiwan and other emerging Asian economies.
China is just another emerging semiconductor economy in a long line of national semiconductor industries. What is maybe different than Japan and China is that Japan actually leads the way with better yields and process technology than the United States. It’s pretty easy to destroy your competitor’s profits if you’re making 10-20% better yields, which leads to much better gross margins. You simply cannot keep up.
That is not the same today for China, where yields are not quite to the leading producers today. But something I think is also interesting in this entire conversation is how important Japan’s domestic economy is in this story - Japan in this entire decade was the emerging power in the global scene. A mix of strong economics leading to the eventual 1986-1991 bubble in asset prices helped buoy their fates.
If the economy is great, and you mostly service a domestic economy, this is a huge tailwind for Japan. What is interesting to me is that the timing of the bubble actually was when Japan’s semiconductor companies started to revert, but I think that if Japan didn’t have the bubble and subsequent lost decades we would have a very different mix of manufacturing and semiconductor leaders today.
I would like to believe given Japan’s largely internal market and continued strength, Japanese companies would continue to flourish. Japan would have acted as a strong foil to the United States, creating a two-horse race that would eventually be bolstered by Taiwan. Instead, Japan’s semiconductor industry is definitely one tinged in past glory. I can point to many companies whose peak relevance was likely at the nadir of Japan’s economy - like Nikon and Canon.
A pattern recognition I can’t help but notice is that strong domestic support pushed international markets into turmoil. I think today that could once again happen in a time where every major geography is pushing a semiconductor subsidy bill.
Investing During Declines
Additionally something I can’t help but notice that is not geopolitically driven is the differences in investing during the cycle. The industry really was different, and the staggering capex increases in the face of declining profits is something we would never see today. I couldn’t believe my eyes at the old Intel and National annual reports.
You can point to a part of this being the maturity of the industry, but another part is I believe that there are way fewer firms than there were in the 1980s, particularly in the likes of memory production. The On semiconductor chart of consolidation likely has a trend line to goes back to the 1980s that would list 1000s of companies.
The boom-bust of the memory cycle, in particular, is massively dampened as the continued consolidation and intense capex management of the firms dampens the cycle. Now the memory oligopoly in DRAM and partially in NAND is so different from the DRAM dumping from Japan. The industry dynamics have definitely improved.
I believe that a demand lead memory cycle should happen again, but the days of memory dumping are likely behind us. That is certainly different, and growing capex 25% amid negative earnings is a trend I don’t expect to quite see again.
The industry has matured, the 1980s is another lifetime in the semiconductor industry. It was the wild west, and those days of massive unit growth are gone. Next time we will talk about the 1990s and cumulating in the 2000 bubble.
Links and various sources
http://smithsonianchips.si.edu/ice/cd/CEICM/SECTION1.pdf
http://thf_media.s3.amazonaws.com/1991/pdf/bg805.pdf
https://irle.berkeley.edu/files/2005/Offshoring-in-the-Semiconductor-Industry.pdf
https://www.files.ethz.ch/isn/100101/2002_05_E-Business.pdf
https://www.eib.org/attachments/pj/semiconductor_industry_en.pdf
Until next time - I hope to have a 1990s history lesson out at some point. I thought splitting the entire “book” into more digestible pieces was a good idea. Also, I think readers should be a bit worried about Chinese shutdowns and OSAT exposed companies right now. I will probably write about that soon.
Nice work, Doug. I forget if you have mentioned before, but Chips and Change is a decent book for anyone particularly interested in this.
thanks for sharing Doug. I'm doing the same exercise with dot com bubble, 2007-2009, 2013-2015 and 2017-2019 cycles; if you got anything would love to read it too; meanwhile happy to connect if u'd like although I just started this exercise