Last month, Morris Chang, the founder and two-time CEO of TSMC (now retired and in his 90s), gave a rare interview, hosted by the Brookings Institution and CSIS, two of the most prestigious DC think tanks.
During this interview, in no uncertain terms, Morris called the United States’ current effort to bolster its domestic semiconductor industry “a very expensive exercise in futility”. The interview’s YouTube video currently has less than 5,000 views. It clearly did not go viral, but it should have, because if we, as a country, are serious about what semiconductors mean in America's future, we should listen more to Morris Chang.
So what did Morris say that our politicians and policymakers should have listened to?
Math (Still) Does Not Work
During this interview, Morris recounted the trial and tribulation of building TSMC’s first manufacturing facility in the US in 1997. The plant was in Oregon. During this plant’s 25 years of operation, TSMC tried everything it could to improve its performance and reduce its cost. Although performance did improve and the plant remains profitable until today, its cost also remains stubbornly high – 50% higher than a similar facility in Taiwan.
Simply put, the math did not work! In Morris’s eyes, due to America’s lack of manufacturing talent and higher cost, no US-based semiconductor manufacturing facility can compete with a Taiwan-based facility, due to the cold logic of unit economics.
The math never worked for TSMC in Oregon. Have things changed?
Last Friday, President Biden made headlines by asking Congress to “pass the damn bill” during a speech in Ohio. The bill he was referring to is the so-called Bipartisan Innovation Act (BIA), which, if passed, contains a $52 billion semiconductor subsidy that would, in part, fund a new Intel plant in Ohio.
Intentionally or not, Biden told the crowd that the average salary at this new plant would be $135,000 a year!
Guess how much is TSMC’s median salary, as of last year? $61,000! (181,000 New Taiwan Dollars to be more exact.)
On the measurement of annual salary alone, the Intel Ohio plant would cost more than twice as much as a TSMC plant in Taiwan. This is removing the less obvious experience gap between the two workforces and the more obvious difference in healthcare cost. (Taiwan is well-known for having high-quality affordable healthcare; America is not.)
25 years after TSMC built its Oregon plant, the math still does not work. Actually, it has gotten worse! Meanwhile, the $52 billion subsidy will only last for five years, not nearly enough time to make Intel's Ohio plant less costly and more globally competitive. It’s no wonder Morris thinks this is an “exercise in futility”.
What Should America Invest In?
Despite the pessimistic perspective, Morris was not completely down on America’s place in the semiconductor industry as a whole. He believes that America is superior in semiconductor design, just not manufacturing.
From Apple’s M1 chip and Google’s Tensor Processing Unit (TPU), to Tesla’s Full Self-Driving chip and every GPU Nvidia has ever designed, American tech companies have an unquestionable competitive advantage in chip design.
Taiwan does not. TSMC has no design capabilities. Other Taiwanese chip design firms are far behind. Neither does Mainland China. While both Huawei and Alibaba are known for designing powerful chips, they and other Chinese tech firms have been hobbled by US sanctions with no domestic solution in sight.
Instead of further investing in what we are good at – chip design – the BIA is pouring taxpayer money into something we have not done well for decades – chip manufacturing.
By not listening to the sobering voice of Morris Chang, America seems to be playing to our weakness, not our strength.
Exactly one year ago, I wrote a post dissecting an hour-long speech Morris delivered (in Chinese) in front of a room full of Taiwanese government and business leaders. (You can read our English transcription in “Morris Chang’s Last Speech” (premium content))
One of the key lessons I drew from that speech is: innovate to motivate.
When he started TSMC in the mid-1980s, Taiwan was poor, backward, and eager to advance. For Taiwan, semiconductor manufacturing was innovative and motivated many generations of Taiwanese people to this day! For America, manufacturing was already not a “cool place to work” in the 80s – talented young people were gravitating toward Wall Street and a nascent but promising thing called the Internet – a trend that has also continued to this day.
Despite having the word “Innovation” in the bill’s name, the BIA is funding the recreation of the “semiconductor manufacturing” wheels. It is a missed opportunity to truly innovate and inspire, in my view, for a bill that has such a rare high-level of bipartisan support.
Instead of saying “We choose to go to the Moon'', like JFK in 1962, the BIA is saying “We choose to do something Taiwan already does really well because we are scared of China”, even though going to the Moon and Mars, or tackling climate change, or developing quantum computing, or exploring biotech would have needed more semiconductors anyways. The custom-built semiconductors powering these innovations could all be designed by American companies.
So will this Bipartisan Innovation Act be a complete waste of money and time? Not completely, but we should temper our expectations.
Temper Our Expectations
It is a foregone conclusion that chips made in America will cost significantly more than in Taiwan, whether they are from Intel’s new Ohio plant or TSMC’s new plant in Arizona. But they will still have customers. Any plant that BIA will subsidize would onshore some semiconductor production capacity to supply highly-regulated or highly-sensitive industries, like defense, telecoms, and cloud computing for government use. These industries will be required to “Buy American”, no matter how pricey these American-made chips are.
To be clear, I support the BIA in spirit and believe that developing this redundant capacity and self-reliance is a worthy goal. We should just keep in mind that it is both more expensive and won’t make a dent in America’s global competitiveness.
The passage of the BIA is also not assured. The bill is inching towards the legislative phase, where the two chambers of the US Congress form what is called a “conference committee” to merge the two different versions previously passed by the House and the Senate. Congressional staffers have indicated that finishing this phase by the end of summer will be a Herculean effort.
The 5-year $52 billion semiconductor subsidy is designed to kick in at the start of the US Federal government’s fiscal year 2022, which means October. If the “conference committee” takes too long, Congress will likely pass the easiest parts – like this semiconductor subsidy – but delay some of the other more innovative elements – like funding wireless supply chain innovation, authorizing programs for space exploration, and increasing computer science education in elementary schools.
What we will end up with will be a watered-down version of the BIA that is already lacking in both innovation and inspiration. It will be a bill that definitely would not have attracted the next Morris Chang to come to America to study, work, and build semiconductors.
(本篇中文版文章是读者 Ben Yu 做的翻译，我做了一些修改而发表。非常感谢Ben的贡献，《互联》的读者们真伟大！)
上个月，台积电创始人兼两任 CEO 的张忠谋（现已退休，目前 91 岁）罕见地接受了一次采访，这次采访由布鲁金斯学会（Brookings Institution）和美国战略与国际问题研究中心（CSIS）主持，这是华盛顿最知名的两个智囊团。
在这次采访中，张忠谋毫不含糊地说到美国现在提升本土的半导体制造业的努力是在白白浪费，且徒劳无获。这套采访的YouTube视频目前点击量也就 5000 次，显然没有病毒式传播开，但它应该得到更多人的关注，因为如果想要了解半导体产业对于美国的未来究竟意味着什么，就应该多听听张忠谋说的话。
在这次采访中，张忠谋提到他在 1997 落地台积电的第一个美国工厂时的经历，工厂位于俄勒冈州，台积电在这 25 年间一直努力提高生产表现的同时降低成本，但虽然今天这个工厂能够盈利，成本依然比台湾同类型工厂高出 50%。
无论是否有意，拜登透露了一个信息，这家新工厂的平均年薪将是十三万五千美元，而台积电的去年平均工资是六万一千美元（即 181,000 新台币)。
在台积电俄勒冈州建厂 25 年后的今天，这种成本优势依旧存在，甚至越发明显，与此同时，这 520 亿美元的补贴只能维持五年，远远不够帮助英特尔在俄亥俄州的工厂降低成本，提高全球竞争力，这样来看张忠谋说的话是合乎情理的。
上世纪 80 年代中期，在张忠谋创办台积电时，台湾贫穷落后，渴望进步，而半导体制造在大众的概念中就是富有创新的，直到今天依然激励了许多代台湾人。对于同时期的美国来说，有才华的年轻人都被吸引去了华尔街从事金融行业，或者加入互联网行业，而不是传统的制造业，这种趋势直到今天依然存在。
尽管法案名称中有“创新”一词，但实际上 BIA 只是在制造半导体上“重复造车轮”，这项法案罕见地得到了两党统一的支持，但在我看来，此法案使美国错失了一次真正的创新机会。
不像肯尼迪在 1962 年说的“我们决定登月”，BIA 想说得仅仅是 “我们决定去做台湾已经做得很好的事情，因为我们害怕中国的崛起”，即使登上月球火星、应对气候变化、开发量子计算机、或者探索生物科技都需要用更多的半导体，而这些定制化的半导体都可以由美国公司设计。
那么 BIA 是否完全是在浪费金钱和时间呢？也不完全是，但大家应该降低期望值。
无论是英特尔在俄亥俄州的新工厂，还是台积电在亚利桑那州的新工厂，美国制造芯片的成本都远远高于台湾，这是无可避免的。但是这些工厂依然会有客户，只要 BIA 提供补贴，任何工厂都会向美国国内提供一定的半导体生产能力以供政府使用。由于国防、电信和云计算等行业的高度监管和敏感性，这些行业会被要求只能购买美国本国生产的芯片。
需要说明一点：我支持 BIA 法案所代表的欲望和精神，我也相信发展这种本土能力实现产业自给自足是有价值的，只是我们应该知道这样一个客观事实，在美国发展芯片制造业，它既更加昂贵，也无利于美国在全球提高竞争力。
BIA 能不能通过也还不确定，该法案正在步入一个特殊的立法阶段，即美国国会两院组成所谓的 "会议委员会" （conference committee），以合并众议院和参议院之前通过的两个不同版本。有国会工作人员透露，要在暑假结束之前完成这个阶段是一项艰巨的任务。
这项为期 5 年、价值 520 亿美元的半导体补贴旨在在美国联邦政府 2022 财年（即 10 月）开始时启动。 如果 “会议委员会” 过程花费的时间太长，国会可能会通过最简单的部分，比如半导体补贴，但会推迟一些其他更具创新性的部分——比如资助无线供应链创新、授权太空探索项目以及增加小学阶段的计算机科学教育。