Yves here. Marshall gives a helpful, high level overview of the stat of competition in the chip biz and why Taiwan is a key player, as well as how Taiwan’s prominent position feeds into US/China economic and geopolitical jousting.
Marshall Auerback is a market analyst and commentator. Produced by Economy for All, a project of the Independent Media Institute
The media likes to dabble in war game fantasies between the 21st-century great powers China and the U.S., but it’s a distraction from the hybrid economic warfare that is underway—from Trump’s tariff hikes to the shores of the advanced economy.
Here in a nutshell is the problem facing the United States. The country that used to be a world leader in all forms of high tech, especially semiconductor chips, now spends its time re-designing chocolate chips. By contrast, Taiwan, officially a “rogue province” of China, but in reality operating as an independent nation of 23 million people, ranked 22nd as a world economy (right behind Switzerland), is now a leading global player in the production of semiconductor chips. As such, it has emerged as the key supply link to a multiplicity of American and Chinese high-tech companies at a time when the Trump administration is working hard to cut China’s access to Taiwan’s semiconductors.
For all of China’s significant technological advancements, the country still lags in the production of semiconductor chips.
Memory chips are principally made by two South Korean companies, Samsung and SK Hynix, and one U.S. company, Micron. Intel, another U.S. company, also makes some memory chips for its own use. Memory chips are a big issue for China. Beijing has deployed considerable fiscal resources into producing them and last year set a goal of producing 5 percent of the world’s total production by the end of 2020.
That’s ambitious. It’s one thing to produce memory chips, another to get a usable “yield,” i.e., the percentage of output that actually works. It is a singularly challenging industry in which to attain industrial self-sufficiency.
Taiwan Semiconductor Manufacturing Company (TSMC) produces customized semiconductor chips “for use in various types of electronics, such as digital cameras, smartphones, and the new technologically sophisticated ‘smart’ cars.” It is a “fabless chip maker,” meaning it “designs and sells the hardware and semiconductor chips but does not manufacture the silicon wafers, or chips, used in its products; instead, it outsources the fabrication to a manufacturing plant or foundry.” TSMC also produces chips for the military, and for 5G base stations. China’s leading telecom equipment manufacturer, Huawei, had been a large customer for TSMC. But in May, the Trump administration mandated that all semiconductor chip manufacturers using U.S. “chipmaking equipment, intellectual property or design software will have to apply for a license before shipping chips to Huawei,” as Nikkei Asian Review reported, thus forcing TSMC to stop taking fresh orders from Huawei, as TSMC’s manufacturing process uses equipment from U.S. companies such as Lam Research and Applied Materials.
The wisdom of so many companies relying on manufacturing facilities located in Taiwan is debatable. Intel and Micron locate fabrication plants (“fabs”) around the world, in part to diversify risk (earthquake, weather, politics) and to access skilled labor pools. Intel has long had production facilities in Ireland, Israel, and China itself; it has also purchased Israeli companies for their research and development. But it also has retained significant production facilities still in the United States. Similarly, Micron has fabs both abroad—in Singapore, Japan, and Scotland—and in the U.S., in Boise, Idaho; Utah; and Manassas, Virginia (right near the CIA and Pentagon).
TSMC is important because it is pretty much the only place to get processor chips fabricated, unless you’re Intel. In that regard, Intel’s second-quarter earnings announcement on July 23 that its planned launch of the company’s next generation of chips will be delayed by six months is most concerning. News of the production delay (which now pushes the production of the company’s latest central processing unit—aka the “brain” of the computer—out to early 2023) generated considerable market anxiety, as evidenced by the 17 percent fall in the share price in the wake of the disclosure. From a long-term perspective, however, the more alarming aspect is Intel’s decision to consider outsourcing its manufacturing capacity, a sharp break from the company’s historic practice.
Intel has been one of the few leading American high-tech companies that has hitherto largely resisted the panacea of offshoring its production. As the Indian Express has written: “The Santa Clara, California-based company has been the largest chipmaker for most of the past 30 years by combining the best designs with cutting-edge factories, several of which are still based in the U.S.” Much of this is a product of the corporate culture established by former CEO Andy Grove, who had warned that Silicon Valley risked “squandering its competitive edge in innovation by failing to propel strong job growth in the United States,” according to a New York Times op-ed by Teresa Tritch written shortly after Grove’s death. Tritch explains Grove’s belief that the lower cost to companies that outsourced to Asia actually “masked the high price of offshoring as measured by lost jobs and lost expertise. Silicon Valley misjudged the severity of those losses, he wrote, because of a ‘misplaced faith in the power of start-ups to create U.S. jobs.’” She continues:
Mr. Grove contrasted the start-up phase of a business, when uses for new technologies are identified, with the scale-up phase, when technology goes from prototype to mass production. Both are important. But only scale-up is an engine for job growth—and scale-up, in general, no longer occurs in the United States. ‘Without scaling,’ he wrote, ‘we don’t just lose jobs—we lose our hold on new technologies’ and ‘ultimately damage our capacity to innovate.’
The expectation is that Taiwan’s TSMC would be the likely beneficiary if Intel were to embrace the offshoring option, which, according to Grove, would ultimately undermine its capacity to innovate. It is consistent with recent historic trends, but Intel’s decision comes at a time when American policymakers are finally beginning to appreciate the adverse economic and strategic consequences of such moves. Were Intel to follow through on its outsourcing threat, it too would further exacerbate America’s strategic reliance on Taiwan for customized semiconductor manufacturing (as well as eviscerating what is left of America’s industrial base).
Additional outsourcing will not only undermine the impact of recent legislative attempts to rebuild the country’s semiconductor manufacturing capacity, but also likely enhances the prospect of intellectual property theft, via sabotage or “hidden backdoors” that could facilitate external surveillance. This would also constitute a national security risk, particularly as production gets more automated.
A key thing to do in dealing with China (or Taiwan) is something like the economic patriotism bill that Joe Biden has proposed: namely, one that would reward companies for insourcing. Research and development tax credits on their own are unlikely to induce the requisite shift (as these can easily be matched by the recipient investment country’s government), as I’ve written before: “The state can and must drive this redomiciling process in other ways: via local content requirements (LCRs), tariffs, quotas and/or government procurement local sourcing requirements.” The Biden proposal doesn’t go that far, but it represents a good step in the right direction and definitely is preferable to a military response to solve what is essentially a strategic industrial vulnerability.
By contrast, economic competition that degenerates into out-and-out war would be a disaster for all sides. As David Arase, resident professor of international politics at the Hopkins-Nanjing Center of the Johns Hopkins University School of Advanced International Studies, contended in a July 20 Asia Times article, “Even an unsuccessful invasion of Taiwan would cause a supply chain disruption.” By the same token, actively upgrading diplomatic relations with Taiwan to something akin to the old mutual defense treaty that existed prior to Washington’s recognition of Beijing in 1979 as the one sovereign government representing China would almost certainly provoke a more aggressive response from Beijing. In that regard, Secretary of State Mike Pompeo’s hawkish speech, seemingly exhorting China’s population to rally against the Communist Party leadership, on July 23 was profoundly misguided.
U.S. goals should be far more modest: not to underwrite the freedom aspirations of another country (even a vibrant multiparty democracy such as Taiwan) but, rather, to fix a key vulnerability in the global supply chain that currently renders the U.S. so reliant on Taiwan. Even TSMC has implicitly acknowledged its own geographical shortcomings, as it announced plans in May to build a new $12 billion chip manufacturing facility in Arizona. Consider this a form of political risk insurance.
A full-scale defense of Taiwan would cost thousands of lives, and potentially entrench the U.S. military in a long-term quagmire; it would also represent a logistical nightmare in terms of supplying such a force over so many thousands of miles (while the Chinese army only needs to cross a mere 100 miles to reach Taiwan.) This is to say nothing of the risks posed to numerous substantial American multinationals already operating in China.
A key conceptual problem that our policymakers and business leaders have today is an addiction to 19th-century concepts that are anomalous in the context of a 21st-century economy. The “comparative advantage” (“an economy’s ability to produce goods and services at a lower opportunity cost than that of trade partners”) of David Ricardo’s 1817 book has less relevance at a time when such advantage can be largely created as a byproduct of state policy. Countries such as Taiwan, South Korea and now China itself can dominate any number of targeted industries by subsidizing them aggressively, whether the industry is steel production, cars, or semiconductor chips. Because of increasing returns to scale, there is a winner-take-all pattern in which at any given time, a limited number of nations tends to dominate a huge global market share of the underlying product. We’ve seen this pattern manifest itself repeatedly in Asia since the 1970s through today, as Robert Wade illustrated in his work, Governing the Market. This strategy has also created huge employment opportunities in high-quality jobs for the countries as they scale up production. This was also a key insight of Andy Grove.
None of these countries had a natural “comparative advantage” in semiconductor production; they just followed the classic pattern of subsidizing their growth via substantial government support (creating them out of nothing over a matter of a few decades, in the case of South Korea and China) relentlessly driving down cost inputs to push other marginal and less efficient manufacturers out of the industry.
The incessant focus on market share usually comes at a cost of short-term profitability (a no-no for Wall Street, which focuses on quarterly earnings as intently as an audience waiting for the white smoke to emerge from a papal election). However, businesses usually recoup these costs later once they’ve established dominant market share. The semiconductor industry is one with a great high-value-added, high-tech manufacturing platform that has employed lots of people and has a huge, growing global market, and a significant multiplier effect on the domestic economy. It represents an area that should be prioritized by the U.S., not de-emphasized (as Intel’s proposed move threatens to do). The road back to manufacturing relevance is a long one, but the perpetuation of the current policy risks exacerbating longstanding pathologies in the U.S. economy, while simultaneously creating new national security vulnerabilities.
Taiwan is a vibrant multiparty democracy that constitutes a model of economic development. But those virtues could be threatened if we shortsightedly try to rush turning it into a U.S. protectorate to address problems that should be resolved much closer to home.
“U.S. goals should be far more modest: not to underwrite the freedom aspirations of another country (even a vibrant multiparty democracy such as Taiwan) but, rather, to fix a key vulnerability in the global supply chain that currently renders the U.S. so reliant on Taiwan”
You know, I think it would help all if we stopped pretending to want to be moral, and really said “it’s all about our interests”. But you know, the problem with that is that the way 1% behave is exactly that – “it’s all about our interests”. Or more generally, it’s the libertarian wet-dream “everyone cares for their interests, there’s no such thing as society”.
So we have to deal with the internal inconsistency (which, IMO, is really what being human is about), when we berate one group for not caring for others while wanting another group to care just for their interests regardless of moral.
(I’d like to point out that no-one, person, society, or country, has a physical _duty_ to care for others [except as enforced by laws etc]. All it has is a choice, and the consequences of the choice. No choice is costless).
TSMC is not a fabless chip maker, but a foundry: the firm actually manufactures chips under contract from fabless chip makers which do not have the super-expensive facility required to produce them.
TMSC is one of the very few corporations actually able to manufacture chips on a large scale, and currently a rare one able to do so with the most advanced 7nm process, and trialing out the even more advanced and costly 5nm technology.
This error made me doubt the credibility of the author.
Yes, this is such a fundamental misunderstanding of the business. The “fab” (a term somewhat interchangeable with foundry) in “fabless” is a semiconductor fabrication line.
TSMC does essentially zero design work, and only manufactures chips others develop. They’re a pure play foundry, which anyone even somewhat familiar with the industry would know. 2 minutes on Wikipedia and linking to it does not an analyst make, especially when you fail to even understand what was written there. He even linked to a list of semiconductor fabrication plants, but somehow managed to miss the numerous entries for TSMC…
Yep. They are the fab in fabless…
Whether by accident or design, Taiwan has made itself strategically essential in high tech supply chains, which massively raises the stakes if China tries to take it over (and make no mistake, this is a fundamental strategic objective for Beijing). And the crackdown on HK is a signal that China is no longer trying to woo Taiwan with promises a ‘one country, two (or three) systems’ approach.
Where once, intervening to ‘save’ Taiwan would have been just a matter of politics and geopolitical game playing, it’s now become a crucial strategic matter for multiple other countries, including South Korea, Japan, and Europe. So even if a future US President was to step back from guaranteeing security for Taiwan, this would not be a green light for China to muscle in. In reality, it could lead to a very messy conflict with global implications.
Its worth pointing out that the US intervened directly in the 1970’s to stop Taiwan (when it was an autocracy) from building a nuclear warhead. No doubt they still have the blueprints.
You wonder if a second-term Trump would try to order Taiwan not to sell semiconductor chips to China under a law passed in the US. Don’t laugh. He is currently trying to tell the Germans that they are not allowed to buy Russian gas under a US law he had passed. At that point China would tell Taiwan that if they are not allowed to buy semiconductor chips, then nobody will be able to. And that they are no longer of any use to China.
I am not saying that China would invade Taiwan but a few loose missiles targeting semiconductor chips facilities in Taiwan would not help out the world economy. The smart thing to do would be to encourage companies like Intel to keep manufacturing in the US itself. But if they do not, well, aren’t the executives for those corporations due for personal tax audits by the IRS? Audits with extreme prejudice?
Most of Intel’s fabs are already in the U.S. (https://en.wikipedia.org/wiki/List_of_Intel_manufacturing_sites). The few that are abroad are mostly in what would be viewed as friendly countries–Ireland and Israel. The exception is a flash memory only fab in China.
However, at high level, there are two parts to making a CPU. One is to fabricate the die–this is the actual processor. Fabs make the die. The other part is installing the die into a larger package, which is the “CPU” unit sold to a customer. This larger supporting package tends to be assembled overseas. For example, the CPU I use in my main PC is marked “Malay” on the heat spreader, as it was assembled in Malaysia However, the actual die was almost certainly fabricated in Oregon.
I didn’t write my last comment as clearly as I intended. The die plus its packaging is a “CPU” sold to the end customer, and the assembly of the CPU (die + the rest) seems to mostly be done overseas. I suspect that is a cost measure.
In the computer industry, especially in the sectors that value performance, like gaming or supercomputing, Intel is now regarded as the giant that has stumbled—badly. AMD has taken the lead in developing faster CPUs, and that is with the aid of TSMC who manufacture their chips for them. The reason, and the competition, is about feature size. That is, how small transistors can be made on silicon: the smaller, the better as many more can be packed in the same space and use less power. Intel is stuck at an under-performing 10nm, while TSMC leapt ahead to a very successful 7nm process. This has allowed AMD, along with their better “chiplet” design, to make faster chips than Intel.
TSMC also manufacture other architectures, like ARM. Apple made waves recently for their announcement to drop Intel chips and use their own ARM designs—to be manufactured by TSMC. In high performance computing (HPC), which remains one the more profitable parts of the computer industry with its high margins for high end super computers, Intel has already been displaced from the top performers, by a Japanese ARM based design at No. 1, along with other non-Intel chips. AMD Epyc chips are now the preferred option for new supercomputers. This must be especially worrying for Intel because HPC is very profitable with most HPC sites (many state funded) prepared to pay high margins for performance. Worse for Intel, today’s HPC is usually tomorrow’s commodity computing.
I’m sure Intel is worried about losing short term profits, but the company is still quite able to bounce back. Anyways, I’m not sure that slamming Intel’s 10 nm is wholly fair–it actually has a higher transistor density than TSMC’s 7 nm. (Node size is not comparable between fabs at the moment.)
The lack of military knowledge taints this article. There is no way US can defend Taiwan let alone at the cost of mere thousands of lives. Warfare at that scale means at least tens of thousands lost per week, assuming no nuclear weapons.
The fact is if china wants, it can invade Taiwan and US can do nothing except perhaps watching its fleet go down the pacific, then…
It’s not that simple. If China invaded, it would provoke a major response from many important countries, as noted in the article. It would cause major economic pain in China at a time when they cannot afford it. The strategic advantages in gaining control of the TSMC infrastructure wold be somewhat short-lived. Much of the expertise would immediately flee the island. The US and Europe would gladly welcome them in order to help build an alternative manufacturing chain. It would take time and serious money but the strategic necessity would quickly push that aside. It would not be pretty for anyone and could quickly escalate out of control. But it would not be a walkover for China.
I think this is the bottom line. China certainly can seize Taiwan (though with more casualties than they would prefer, I believe, due to lack of landing craft). However, China would not successfully retain the base of human knowledge that would make it attractive to conquer in the first place, making the venture pointless.
Irrespective of any war talk… Just for the record – China’s semiconductor advancement has been happening because thousands of Taiwanese moved to mainland China. Economics trumps politics (no pun intended).
COVID has shown that “immediately flee” is harder than it seems. Events have this nasty way of going faster than reaction time, especially when the reaction consists of letting go extended family, properties and retirement entitlements. Look at how few people emigrate from HK, despite the proverbial writing on the wall !
Even when the decision to leave is made up, execution is difficult : due to its insular location, Taiwan and its allies would need to retake control of the skies and the sea, which is practically equivalent to winning the war, thus obviating the need to flee in the first place !
The best I can envision is special-force-like exfiltration of carefully selected top engineers/Scientists, and it is not guaranteed to work.
Actually there is a huge brain drain from Taiwan to mainland China in the semiconductor industry. Literally thousands have left. There is only one way China invades – and that is if they declare independence (and they would only do that if the US urges them). They won’t do it for TSMC. Behind all the political optics – there are huge exchanges between the two sides. The overwhelming majority of Taiwanese off the island – don’t live in the US or Europe. They live across the Strait on mainland China.
Americans always fantasize about the nuclear option, to terminate the game promptly with overwhelming force that allows no reply. That strategy has exhausted itself in part because, like most Americans and especially right-wing gamers, it ignores the existence of the meta-game and the existence of other players who have interests that specifically exclude playing bullshit games with other people. Even some of the most ill-adjusted people have qualms about the real-life consequences of ending the game and will switch to another game.
US intervention to defend Taiwan would exclusively be Air Force and Navy, and mostly far from China : think oil tankers in the Indian Ocean. The USA learnt its lessons to not wage war on the Asian landmass. Furthermore, Japan and Korea would understand that a successful attempt on Taiwan means that they are next, and only protecting their own territories and supplies will make them de facto Taiwan allies. Last but not least, a resolute Taiwan is very difficult to tame by itself, they have missiles that can reach critical infrastructure as dams, they train into guerrilla warfare, etc… : landing an expeditionary force is not the end of the story if it becomes cut from supply lines and reinforcement…
In a long conventional fight WW2 style, as the most probable of scenarios, I don’t see a victorious China ; but I fear a loosing China would trigger nuclear confrontation : China has a no-first-strike policy, but the list of what anything is meant to be a first strike is very long, and explicitly includes anything that would prevent it to retake Taiwan.
Best outcome for the US is actually that China gains Taiwan To save face, but looses everything else in the process of a long fight : access to world consumers and world commodities, and overseas assets.
Sometimes I feel like it is actually the US plan, and that the CCP would go along with it, as it consolidates its power while providing an excuse to not become a prosperous society. North Korea on steroids, a scary thought !
This may be a stupid question but – Why this race to get smaller and smaller and faster and faster – I mean, for me – I have had enough computational processor speed for decades – -my reason to upgrade my devices was more driven by ever more bloated and patched software applications – We sent people to land on the moon with an extremely limited computational software footprint. Most people, including myself – are paying for far to much capacity in their home environments than is needed – seems that some providers want to bloat their stuff with tons more data than is needed. I remember when high def TV was being rolled out and the FCC doubled the amount of broadcast space given to the consolidating media corps.
The whole high def was gamed to the point that that extra broadcast space was a free ride as rules were diluted and tech enabled 4K non-interlaced broadcast to occur over existing wires – I am not sure if even today we have full 4K non-interlaced broadcast as we should be getting – Probably a whole article is needed – still
My point is – how much do we really need when all the graft is pulled back away from it
It depends on what you mean by “we.”
As you suggest in your comment, for most normal users, current computational power is more than sufficient. However, there are various workloads that continue to see benefits from increased power, ranging from academic and research (e.g., supercomputing simulations), business (e.g., datacenter servers), professional (e.g., video editing, Computer-Aided Design, 3D modelling and rendering), and entertainment (e.g., gaming).
Many of the markets relating to such areas will continue to reward additional power for the foreseeable future because the business economics make sense–the increases in productivity readily justify the modest (for the relevant purchaser’s) costs. The only exception in what I listed above is gaming, but people are known to spend “extravagantly” on their favored hobbies and better hardware does permit a more realistic and fluid gaming experience. (Granted, in recent years, gaming has become much more dependent on graphics processors than on CPUs.)
Let me answer your very reasonable question with a counter-question: why aren’t there dozens of different types of consumer computers for every conceivable use, from fully-programmable by the end user to fully-secure and configured by a technician as part of a white-glove service that you could pay for from a local provider you trust? I’m an old millennial, so I vaguely remember Radio Shack and ColecoVision/Atari and soldering benches in suburban garages with the other repair equipment. I had friends with stuff like programmable ‘home automation’ using analog switches on the home wiring to set timers. As the chips got smaller and more prolific, why did all of the options also go away, the repair shops, the component stores, etc?
The answer to your question is the same as my question: lack of competition. At every single stage of the computer hardware manufacturing life cycle, for the past ~40 years, small organizations have been replaced by large ones, because the large ones gained increased ability/speed by making more and more of these smaller components that they could then load up into more and more devices. As the large got larger, they bought up software and hardware patents to absorb or kill off competitive technologies. This combined with private equity looting to remove all the smaller, “good enough” or niche options that may have been a bit larger or clunkier or, the current scam, lacking a cloud interface to move data around seemlessly, but were good enough for many people.
Another aspect – also competition-related – is the lack of widely available operating systems aside from osx/Mac, Windows, and x86/Linux. There are stripped down, open-source flavors of Linux that provide a lot of user control, but one has to have the requisite technical skills to manage that.
What is an OS, then, and why would anyone want more of them? This demonic fervor to jam markets into places they don’t belong and of which one displays no understanding of the dynamics of said market (see Bob Cringely, for actual dynamics and numbers) sounds really familiar, I hate to say.
So, the manufacture of semiconductors is just another reason for going local. Very local. The technology is widely available. At first you say oh god no it’s armageddon. But that’s such unmitigated nonsense it doesn’t require a response except to say it’s only Armageddon if vast amounts of money have been applied to a single source of semiconductors so that a vast profit can be sucked out of them – as opposed to many many localized manufactures of semiconductors. And gates against profiteering on a global scale blocking their usual maneuverings at hundreds of local points – that’s a good thing. And as in heaven (the internet) so on earth – us idiot chumps trying to communicate with out passive-aggressive computers – that’s also a good thing..
The thing about the ‘Free Market and 5G’ that has always been a sore spot is – with Free Market thinking– is that it somehow roots out bad actors and rewards good – that has not been the case since the dawn of the market
The free market as construed (perverted) today is an aide to the worst in us – frauds, deception, twisted business practices, the enabling through ‘citizens united’ giving voice to corporations (I think T Jefferson said corps do not posses a conscience) and bail out upon bail out of the to big to jail.
5G appears to enable complete cover ups of mis-deeds under the auspices of security in case of a breach or theft of device.
It seems a waste of computational power on High speed trading (a skimming operation in my view)