The much-lauded CHIPS Act is not going well. And TSMC, which is supposedly to be one of the Act’s biggest beneficiaries, is starting to hedge against a failing industrial program.
Since its passage 16 months ago, the Act’s implementation led by the Commerce Department has done very little to restore US leadership in semiconductors. Specifically, on subsidizing and supporting advanced chips manufacturing, the program has only made two relatively inconsequential grants so far – $35 million to BAE Systems, $162 million to Microchip Technology – both for making mature (meaning old, not advanced) chips that are commonly used in the defense industry.
There is, of course, nothing wrong with supporting defense contractors and companies that supply the defense industry; onshoring the production of chips that our military relies on is important for strengthening national security. However, if the reason why we needed the CHIPS Act was to onshore advanced chips manufacturing by leading foundries – TSMC, Intel, Samsung – all of which have been lured to building more fabs in the US under the assumption of receiving sizable government subsidies, this industrial program is simply not delivering.
In TSMC’s Q4 2023 earnings call earlier this week, Mark Liu, the company’s outgoing chairman and point person in driving its expansion into America, gave us some clues to how TSMC really feels about the situation.
Mark’s Last Words
In his prepared remarks for probably his last earnings call, Liu shared this progress update for investors on TSMC’s Arizona fabs:
“In Arizona, we are in close and constant communication with the U.S. government on incentive and a tax credit support and making strong progress in facility supply chain infrastructure, utility supply and equipment installation for our first fab. We continue to work closely and develop strong relationships with our local union and trade partners in Arizona, including recently signed an agreement with Arizona Building and Construction Trades Council on a new framework for cooperation….
We are well on track for volume production of N4 or 4-nanometer process technology in the first half of '25 and are confident that once we begin operations, we will be able to deliver the same level of manufacturing quality and reliability in Arizona as from our fabs in Taiwan.”
While the delay in operationalizing TSMC Arizona’s first fab to 2025 has been widely reported, this is the first time the company shared a more precise timeline for volume production – first half of 2025. It also looks like TSMC’s tense relationship with local unions has come to a steady state, where at least work will move forward and not be stalled by possible strikes or more public relations complaints and campaigns. Furthermore, the CHIPS Act subsidy is still being haggled over with the US government. TSMC is reportedly seeking $15 billion in tax credits and subsidies (out of a pool of $39 billion). But negotiations have been difficult because the Commerce Department is attaching a lot of strings to the money, including compelling the sharing of confidential information about operational details and customers, which is just a big no-no for the foundry business model that TSMC pioneered. (Samsung has raised similar objections to this condition.)
But this is not the hedge; these issues have been widely anticipated (and still unresolved). The hedge against the CHIPS Act came when Liu was asked about the second fab that TSMC plans to build in Arizona. Here is what he said when an analyst asked about it:
“The second fab shell is under construction. But what technology in that shell is still under discussion, and I think that also has to do with how much incentives that fab – the U.S. government can provide…there will be a gap, at least current planning is '27 or '28…that will be the timeframe...
All the fabs in overseas, what's actually being loaded, what technology is being set up, really, it's a decision of customers' demand in that area…so nothing is definitive.”
The 2027/2028 timeline is a further delay from the previously shared 2026 deadline. Reading between the lines of Liu’s diplomatic answer, if the CHIPS Act subsidies do not come through as promised, TSMC doesn’t plan to put its best technology and capability in the second fab. And it is in no hurry to complete it until all the pieces are figured out to make sure this second fab will be a profitable one. If that does not happen until 2028, that is also about the time the CHIPS Act program expires (it is a five-year program).
What a shame, and fail, that would be.
Dream Unfulfilled
Whether it is the first half of 2025 or some time between 2027 to 2028, Mark Liu will not see the fruit of either of the Arizona fabs; he is scheduled to officially retire this summer when TSMC’s shareholders meeting will take place to elect the company’s next chair. One might think that he would want to witness this project – this dream – somewhat fulfilled, if the timeline is to stick around for another 18 months or so. Alas, even with the high-profile honor of introducing President Biden at the fab’s tool-in ceremony a year ago and guaranteed accolades when the first fab finally reaches volume production, it was not enough for Liu to stay. He likely did not imagine dealing with Commerce Department bureaucrats as how he would close out his illustrious 30-year career at TSMC.
The thing is, of all the incredible difficulties and complexities that go into setting up a new advanced chips fab in America, receiving the CHIP Act subsidies was supposed to be the easy part. During the same tool-in ceremony, TSMC’s founder, Morris Chang, gave a short but sobering speech, where he talked about the death of globalization and recounted all the challenges he encountered when trying to set up TSMC’s first US fab in Camas, Washington in 1995. Chang was more hopeful of Arizona’s success rate in no small part because TSMC could count on support from US governments of all levels – federal, state, local – which he didn’t have 27 years ago.
That hope now looks foolishly misplaced. Meanwhile, Taiwan passed its own version of the CHIPS Act in early 2023 that is just as generous as the American one, if not more. More recently, South Korea announced a whopping $471 billion investment plan for the next 25 years to boost Samsung and SK Hynix to make South Korea the center of advanced chip making for the world. TSMC revealed in its earnings call that it is also having productive discussions with the Japanese government to receive support to build more advanced fabs there.
So it’s not like the few billions we’ve allocated is the only game in town; not even close! TSMC’s hedge is the first noticeable sign from a major industry player to signal that the CHIPS Act may not be what it is all cracked up to be. It won’t be the last, if America keeps dragging its feet.
台积电对美国《芯片法案》可能失败的规避
备受美国政界赞赏的《芯片法案》(CHIPS Act) 进展并不顺利。作为该法案最大可能的受益者之一，台积电，看似已经开始对法案实施失败的可能进行规避。
自从16个月前该法案通过以来，由美国商务部领头的落实工作几乎没有对恢复美国在半导体制造领域的领先地位做出什么贡献。在补贴和支持先进芯片制造厂方面，该计划到目前为止只做出了两项相对无关紧要的补贴拨款 —— 给BAE公司的三千五百万美元，给微芯科技的一亿六千两百万美元 —— 这两家公司的项目都是用来生产成熟的（也就是旧的，非先进的）芯片，这些芯片在国防和军用产业中普遍使用。
支持国防承包商和供应国防工业的公司当然没有什么问题；把美国军队所依赖的芯片生产本土化是加强国家安全的重要一步。然而，如果通过《芯片法案》的初衷是要把先进的芯片制造晶圆厂公司 —— 台积电、英特尔、三星 —— 更好的在美国本土化，何况这几家公司也都在默认会获得大量政府补贴的前提下被诱惑，开始在美国投资建厂，那这款工业计划目前根本没有达到其目标。
在本周台积电宣布2023年第四季度的财报电话会议中，即将卸任的公司董事长、也是推动其赴美扩张的关键人物，刘德音，给了大家一些关于台积电对这种状况的真实感受。
刘德音的退役词
在他参加的最后一次财报发言中，刘德音向股东们和投资人们分享了台积电亚利桑那州工厂的进度：
“在亚利桑那，我们与美国政府保持着密切且持续的沟通，就激励补贴和税收抵免支持方面取得了显著进展，并且在第一家工厂的设施供应链基础设施、公用事业供应和设备安装方面也进展顺利。我们继续与亚利桑那州的当地工会和贸易伙伴密切合作，建立更好的合作关系，包括最近与亚利桑那州建筑和建筑工会签订的一份新合作框架协议......
我们正按计划进行，预计在2025年上半年实现4纳米工艺技术的量产，并有信心一旦开始运营，我们将能够在亚利桑那提供与在台湾工厂同样水平的质量和可靠性。” （我个人从英文到中文的翻译，非官方翻译）
虽然台积电亚利桑那的第一所工厂的开张推迟到2025年一事已被广泛报道，但这是公司首次官方分享了更准确的量产时间表：2025年上半年。看起来，台积电与当地工会的紧张关系已经达到了一个稳定状态，至少施工将继续推进，而不会因可能的罢工或更多公关投诉和政治活动而停滞。此外，台积电还在与美国政府就《芯片法案》补贴进行讨价还价。据报道，台积电正在寻求150亿美元的税收抵免和补贴（整个法案总额390亿美元的补贴中的一部分）。但谈判进度一直很困难，因为商务部对补贴附加了许多条件，包括强制共享有关运营细节和客户的机密信息，这对于台积电的芯片制造代工模式来说是一大忌讳。（三星对这一条件也提出了类似的反对。）
但这并不是我所提到的规避；这些问题已早有预料（而且仍未解决）。当刘德音被问及台积电计划在亚利桑那州建造的第二家工厂时，他流露出来对《芯片法案》规避的想法。以下是一位分析师问及此事时，他的回答：
“第二家工厂的外壳正在施工中。但该工厂最终将采用什么技术仍在讨论中，我认为这也与美国政府能提供多少激励补贴有关……至少据当前的计划，将有一个空档期，即‘27年或‘28年……那将是大体的时间框架……
海外的所有工厂，实际上最终装载的科技、正在设置的技术，是取决于客户在那个地区的需求……所以目前没有确定任何具体方案。”（我个人从英文到中文的翻译，非官方翻译）
2027-2028年的时间线之前媒体透露的2026年截止日期又延迟了一大步。刘德音的回答虽然很委婉，但背后的意思很明确，如果《芯片法案》的补贴没有像预期到位，台积电不打算在第二家工厂中放置最好的技术和能力。并且在所有部件都确定好，确保这第二家工厂能盈利之前，台积电也不急于完成它。如果这种情况一直持续到2028年，那也正是《芯片法案》法案到期的时候（它是一个五年期限计划）。
如果真的如此，那对美国将是一个巨大的遗憾和失败。
无法实现的梦想
无论是2025年上半年还是2027年至2028年之间的某个时刻，刘德音都不会看到亚利桑那工厂的产出的果实；他计划在今年夏天正式退休，届时台积电的股东大会将选出公司的下一任董事长。旁观者可能会认为，再坚持一下，再待18个月左右，不就可以亲眼见证这一宏伟项目 —— 这个梦想 —— 的结果了吗？然而，即使在一年前的工厂开工仪式上高调与拜登总统同台，在庆祝第一家工厂最终实现量产时的荣耀和光环，这些都不足以刘德音多留下一天。在台积电30年的辉煌职业生涯即将结束时，他可能万万没有想到要与美国商务部的官僚讨价还价。
事实上，在美国搭建一家全新的先进芯片制造厂会涉及到的所有难题中，获得《芯片法案》补贴本应是最容易的一部分。在同一场工厂开工仪式中，台积电创始人张忠谋发表了一篇简短但发人深省的演讲，谈到了全球化的终结，并回顾了他在1995年试图在华盛顿州的卡马斯建立台积电第一家美国工厂时所遇到的所有挑战。张忠谋对亚利桑那工厂成功给予的希望很大程度上是因为台积电这次能依靠美国各级政府，从联邦到州级到地方的各种支持。这是他27年前没有的。
但现在看来，这一寄托可能错放了。与此同时，台湾在2023年初通过了自己类似于《芯片法案》的计划，其补贴的慷慨程度不亚于美国，甚至更多。最近，韩国也宣布了一项高达4710亿美元长达25年的投资计划，以提升三星和SK海力士的能力，使韩国成为世界先进芯片制造的中心。台积电在其财报会议中透露，它还在与日本政府进行富有成效的交涉，以获得资助在那里建立更先进的工厂。
所以，美国拨出的那几十亿美元并不是台积电的唯一选择，甚至都谈不上是最多的！台积电的规避是来自芯片行业巨头中的第一个信号，表明《芯片法案》的具体落实可能并非像宣传的那样。如果美国政府继续拖延，这将不会是最后一个信号。