TSMC: Semiconductors and Borders of Light

“Globalization is almost dead,” Morris Chang said on Tuesday. “Free trade is almost dead.” 

If anyone should know, it’s Chang. The founder and two-time CEO of Taiwan Semiconductor Manufacturing Company, better known as TSMC, built one of the world’s largest businesses on geographical arbitrage. Much more went into creating the leading semiconductor foundry, but it could not have reached its $417 billion market cap without access to Taipei’s low-cost labor.

While Chang’s pronouncement sounded downbeat, it arrived in auspicious circumstances – at least from the vantage of the American government and private sector. With Joe Biden and Tim Cook in attendance, Chang announced that TSMC plans to invest $40 billion in Arizona to build two new factories. Not only was this triple the amount TSMC had initially pledged as part of a 2020 initiative, it is one of the largest direct investments in American history, comfortably eclipsing Samsung’s $17 billion plan for a Texan chip facility.

More significant than the scale of TSMC’s investment are its implications. Though less buzzy than many other technologies, semiconductors are fundamental to our security, industry, and quality of life. Tiny chips, with transistors just a few nanometers wide, power the computers we use to work, the phones we need to communicate, the cars and trucks transporting people and goods, and the weapons that countries like Ukraine deploy to repel invasion. They are, without exaggeration, vital to modern existence. Even delays in the semiconductor supply chain can choke industries and rile consumers – if companies like TSMC ever shut down, catastrophe would ensue. “GDP would go down 10% overnight,” NZS investor Jon Bathgate remarked. “We’d put ourselves into a global depression.” These wafers of silicon embroidered with circuits act as a border of light, separating us from darkness.  

Because of their importance to all aspects of modern living, semiconductors have become geopolitically strategic, especially for China, the U.S., and Taiwan. China has invested tens of billions of dollars in building a domestic supply chain, the U.S. has issued sanctions to thwart China’s technological progress, and Taiwan relies on the presence of TSMC to provide a “silicon shield,” deterring invasion. 

All of which makes Tuesday’s announcement more notable. Not only will this bring cutting-edge manufacturing to the U.S., it will decentralize it away from Taiwan. The development speed, scale, and ultimate success of the Arizona venture will inform just how meaningful this move is. If TSMC’s $40 billion investment represents the totality of its interest in foreign manufacturing, the new factories will create jobs, increase capacity, and provide a much-needed, albeit minor, hedge against turmoil in Taiwan. But if it is the start of a broader trend – if Morris Chang is right about the collapse of globalization – it will have profound effects politically, commercially, and technologically.

Semiconductors are impossible

On a plane the other day, I read Chip War by Chris Miller. It’s an exceptional account of semiconductor history and its global importance and was my favorite non-fiction book of the year. It also happens to be tremendous fun, mainly for the way in which it contextualizes how absolutely impossible it seems to make a semiconductor. It relies on a stack of complexities all piled on top of each other: mechanical, physical, chemical, logistical, and commercial. “It’s like standing a needle on top of a needle on top of a needle,” Bathgate said. 

There are so many examples of this that I want to share, but first, let’s quickly talk about what a semiconductor is. At a high level, a semiconductor is a material that conducts electricity conditionally. It’s not like glass, an insulator, a poor conductor – nor is it like copper, a good conductor. It’s somewhere in between, and critically, modifiable. Silicon is a famous semiconductor. 

Now, this may be the definition of a semiconductor, but it’s not really what most people mean when they use that term. Rather, “semiconductor” is often shorthand for “semiconductor chips.” Chips are usually made from smooth, dark slices of silicon. Various components are positioned on the silicon surface to modify electrical currents. Transistors are especially important components. These work like a switch, directing currents one way or another. Each of these transistor “gates” translate information into binary code, “convert[ing] real world sensations like images, sound, and radio waves into millions and millions of 1s and 0s,” Miller writes in Chip War. This works only because of the fundamental properties of semiconductor materials.

A single chip may house tens of billions of transistors, each with a gate just a few nanometers wide. The smallest transistor gate is just 0.3 nanometers wide, though it’s nowhere near commercial production. Apple’s A16 chip, used in the latest iPhones, relies on a processor with transistor gates of 5 nanometers (with some modifications). To put this into perspective, a human hair is up to 100,000 nanometers wide, while a single red blood cell measures 7,000 nanometers across. We are talking about minuscule, microscopic dimensions. 

As transistors have shrunk, the number of them that a single chip can accommodate has risen exponentially. “Moore’s Law” has become so ubiquitous in the tech world that it often seems to be used as a shortcut for a kind of basic “tech go fast” message. 

What it really refers to, though, is the increasing number of transistors that can fit on a single chip. In 1965, Intel co-founder Gordon Moore made a specific prediction: that chips would be able to accommodate double the number of transistors every two years for the next decade. Though Moore was right about the direction of travel, his timing was off. For the past six decades, Moore’s law has held (though some believe it is reaching its end). Such efficiency gains have relied on remarkable technological advancements, introducing new techniques, machinery, and chip structures.

Though chips have shrunk, their growing ubiquity has massively expanded the size of the semiconductor market. In 1991, semiconductor sales reached $55 billion. Three decades later, it is ten-fold larger, surpassing $555 billion. In 2022, the market size is expected to reach nearly $614 billion. In the near future, chips will likely become a trillion-dollar industry, with McKinsey expecting it to reach that landmark by 2030. 

Congratulations, Glorious Emperor

As you might expect, manufacturing chips is incredibly complex. You can’t just spin up a factory and start stamping out new semiconductors like Hot Pockets. There’s a delicate, imbricated ecosystem of players, each with bespoke roles. To explain this better, let’s embark on a thought experiment to outline the different stakeholders. 

Congratulations! You’ve just been made Glorious Emperor of the Imaginary Republic of Quartz. As a modern, tech-forward ruler, you know that it’s essential your country has a steady supply of semiconductors. It’s necessary for your cars, machinery, toasters, and, of course, missile systems. The country’s major tech giants need cutting-edge chips for their smartphones, computers, and data centers. 

Previous administrations have allowed the Republic’s public and private sectors to purchase chips from foreign players like TSMC and Samsung, but you want to do things differently. To ensure your great nation has control of its technological progress, you’ve set a goal to reach 100% on-shore semiconductor manufacturing by 2050. You don’t want to be reliant on the U.S., China, Taiwan, or anyone else. Here’s a (very) simplified guide to make it happen:

Step 1: Get raw materials

If you want to make semiconductors, you’ll need some silicon. Luckily for you, the purest silicon comes from quartz, one thing that the Imaginary Republic of Quartz has in abundance. You’ll also need to amass a good amount of common and precious metals (think: aluminum, copper, and so on) and elements like phosphorus and boron. Thankfully, your nation is rich in natural resources – cross this one off the list. 

Step 2: Develop specialized design software

Raw materials aren’t much use if you don’t know how to turn them into chips. To do that, you need specialized software that helps you design the billions of components a semiconductor requires. 

If you wanted to take a shortcut, you’d use a solution like Cadence or Synopsys. In very simple terms, these are kind of like Figma or Adobe for chips, except orders of magnitude more complex. They boast specialized features to support technologies like machine learning, AI, augmented reality, virtual reality, robotics, and more. Both companies have been developing their software solutions since the 1980s.

They also both happen to be based in California, meaning your supply chain will always depend on the goodwill of the United States.

If you want to adhere to your goal, you’ll have to develop a viable alternative, either galvanizing domestic companies to acquire smaller foreign players or aggressively funding commercial attempts to compete. It’s likely to be a long, expensive road ahead.

Step 3: Engineer the world’s most complex machinery

As if the software problem wasn’t tricky enough, you also have to find a way to get the equipment you need to make semiconductors. 

Here’s the problem: if you want to make leading-edge chips, there’s currently only one company to buy from. That’s Advanced Semiconductor Materials Lithography, better known as ASML. The Dutch company is the sole provider of extreme ultraviolet (EUV) lithography equipment.

That’s a lot of acronyms in a short space, so let’s break this down a little further, especially since it is a prime example of why this industry is so fiendishly hard. 

For several decades, semiconductor manufacturers have used light as a tool. By coating chips with photoresist materials, fabricators can manipulate visible light to carve shapes onto components. As transistors shrunk, this method stopped working. Why? Because visible light became too blunt an instrument. (Allow that to sink in.) Original-flavor light has a wavelength as wide as 700 nanometers, while components were becoming much smaller.

Scientists began experimenting with ultraviolet light and, eventually, parts of the extreme ultraviolet spectrum, which had a wavelength of just 13.5 nanometers. After decades of development and billions of dollars invested, only ASML succeeded in creating a commercially viable EUV lithography machine. Each machine relies on hundreds of thousands of components working together to complete the most insane-sounding tasks, like, using a laser to hit a speck of tin 50,000 times a second to create sufficient EUV light. It’s not surprising, then, that a single machine costs $100 million. One industry insider called it “the most complicated machine humans have built.”

As Glorious Emperor, buying from ASML sounds tempting. Sure, it’s expensive, but they have a literal 100% monopoly on the equipment you need. Better yet, they’re in the Netherlands, far away from rival superpowers. Right?

If your goal is to create a genuinely self-sovereign supply chain, relying on ASML would be a mistake. The Netherlands is ultimately a U.S. ally, and the American government has few qualms about exerting its power in this sector. Even more importantly, ASML cannot produce its machines independently. It relies on a web of multi-national suppliers across Japan, Europe, and America. If any of these counterparties – or their governments – want to turn the screw, you could find yourself in trouble. To quote Prince Harry from episode one of Netflix’s cringey Harry & Meghan, “It’s not about who you trust. It’s who they trust.”

You might have to go solo. Time to figure out how to make the most complicated machine in human history. 

Step 4: Find and hire the industry’s foremost experts

After the previous step, this one feels a little more tractable. To build and operate the equipment you plan on using, you’ll need to recruit industry experts. Taking an EUV lithography machine for a spin isn’t simple, nor are the many other tasks semiconductor manufacturing requires.

To get up to speed, you’ll probably want to implement a comprehensive strategy of targeted recruitment and training. Do your best to tempt existing experts to relocate to the Republic and incentivize local labor to learn requisite techniques. Expect this to take a while.

Step 5: Set up an operationally-efficient foundry

Glorious Emperor, you must be tired. And yet, you have not made a single chip. Steps 1-4 may have felt arduous, but the real work has yet to begin. 

It is now time to actually manufacture semiconductors. To do so, you must set up a factory – often referred to as a foundry. During the last regime, the Republic relied on TSMC, Samsung, and Intel. The Taiwanese foundry is the only one capable of leading-edge production; Samsung’s manufacturing is dominated by girthy 28 nanometer chips. If you want to churn out modern semis and avoid the geopolitical tangle of Taiwan, you’ll have to spin up your own facility. 

That will cost around $20 billion. Once it's up and running, you’ll need to work relentlessly to improve the factory's efficiency. Because of the high costs associated with fabrication, running a foundry at low capacity simply isn’t economically viable. You need to optimize, scale and hit at least 95% utilization.

Step 6: Establish a supply chain that assembles components

There is some mercy, Emperor. Our final step is comparatively easy. If you truly want to manage the process from end to end, you need a facility that assembles components. A Foxconn-style factory capable of churning out phones, TVs, and other electronics. 

In case you failed to account for it earlier, you will also need skilled workers for this process – as well as plenty of unskilled ones.

We have now reached the end of our thought experiment. Do you see how absurdly impossible this industry is? Virtually every step requires extraordinary amounts of money, expertise, and coordination; as a result, many are susceptible to foreign power manipulation. Rather than a smooth-flowing supply chain, the semiconductor fabrication process is a string of chokepoints.

This is largely why China has struggled to nationalize parts of the chip industry. In Chip War, Miller calls the country “staggeringly dependent on foreign technology,” with just 1% of the semiconductor software market, 2% of “core” IP, and 1% of tools spending. Across the entire semiconductor supply chain, Miller suggests China’s stake is 6%, compared to the U.S.’s 39%. 

To date, China has invested tens of billions to reduce this dependency. While that’s helped buoy Chinese foundries like Semiconductor Manufacturing International Corporation (SMIC), cutting-edge fabrication is still far off. Earlier this year, news leaked that SMIC had manufactured a 7-nanometer chip. Given the foundry had previously topped out at 14 nanometers, this looked like a considerable leap forward. New York Senator Chuck Schumer fretted about SMIC “threatening U.S. competitiveness.” 

“It was a nice party trick,” NZS’s Brinton Johns said of the smaller chip. “But they can’t produce them at scale.” Will Hunt, an analyst at ​​the Georgetown Center for Security and Emerging Technology, expressed a similar opinion, “It’s one thing to make a single chip, but it’s another thing to have yield and volume.”

The country has had slightly better luck with Yangtze Memory Technologies Corp (YMTC), a manufacturer of NAND chips. These memory chips, often called “flash,” can remember data for longer. YMTC is not among the top-five players in the NAND category, but has typically been seen as comparatively competitive. (Recent sanctions may have changed that.) YMTC received more than $24 billion in government financing.

Ultimately, China has gotten closest to trying to complete the thought experiment we laid out. That it is still far from rivaling the capacity of Taiwan and the U.S. illustrates the difficulty of the task. Though TSMC’s facilities are approximately 100 miles from mainland China, mimicking its capabilities is a world away.

Morris Chang’s playbook 

TSMC’s origin story violates contemporary rules of startup success. It wasn’t founded by a precocious college dropout but a 56-year-old emerging from professional disappointment. It wasn’t funded by venture capitalists but the Taiwanese government (and a series of local angels firmly “encouraged” to participate). And it wasn’t an asset-light internet product – a SaaS platform or social network – but an expensive, complex industrial project. 

Great histories have been written about Morris Chang and his creation. (I am especially partial to Acquired’s excellent episode on the company.) A summary: 

After decades of driving the semiconductor industry forward at Texas Instruments, Chang was overlooked for the CEO job. In 1987, after a brief, unremarkable spell at another company, Chang moved to Taiwan to serve as director of the country’s Industrial Technology Research Institute. The veteran had never lived in Taiwan before, having grown up in Hong Kong before spending most of his adult life in Texas.

Though Chang’s role may have sounded bureaucratic, he was essentially free to operate as he wished – with the government’s financial backing. 

It proved a wise move. Chang oversaw the ascendance of one of the world’s largest companies, valued at $417 billion at the time of writing. TSMC’s 2021 revenue eclipsed $57 billion; its fourth-quarter revenue may top $20 billion after a bumper November, which sees the foundry up 50% year-over-year. “The scale they’re growing at is basically unprecedented,” Brinton Johns remarked. The NZS founders noted that many commentators highlight Amazon’s AWS as a kind of business miracle: a division compounding approximately 30% with an annualized run rate of nearly $80 billion. Meanwhile, TSMC is achieving a similar feat at a slightly quicker clip.

How has Chang built a company that keeps compounding from a massive base – and is essential to the global economy? His playbook involves a mix of geographical arbitrage, a sharp business model, savvy R&D investing, and relentless execution. 

Geographical arbitrage

Morris Chang was not the first chipmaker to leverage a low-cost workforce. In the 1960s, companies like Fairchild Semiconductor and Texas Instruments started shifting labor to Asia. Taiwanese workers cost 19 cents an hour, compared to $2.50 for American workers. In 1969, led by Morris Chang, Texas Instruments opened a factory in Taiwan. By the time Chang started TSMC nearly two decades later, the country had established local talent with experience working in the industry, still available at lower prices. 

The price differential is not so stark today, but a significant gap remains. Semiconductor engineers in Taiwan average $35,200 salaries, compared to $118,300 for U.S. equivalents.

Picking an aligned business model

Chang’s best decision was to make TSMC a “pure-play” foundry. Unlike Samsung or Intel, for example, TSMC didn’t design chips or manufacture them for its own use. This made it a less complicated partner for customers. Electronics companies or chip design firms didn’t have to worry about TSMC seeing their work and using it to improve their own product line. They knew that Chang’s company was just a fabricator. 

By remaining neutral, TSMC became the preferred provider for the ecosystem’s players. Over time, this put the company in a powerful position, giving it the leverage to set industry standards and benefit from counterparties’ innovation to improve production. Jon Bathgate noted that this “open ecosystem” gave TSMC network effects. “Once you get ahead, there’s a compounding effect.”

Timely investing in R&D

Moore’s Law is not a rule of the universe. It’s not like gravity or the conservation of energy. It was simply an observation and prediction. Which means it may one day stop being true. To try and keep pace with Moore’s Law and outmaneuver competitors, semiconductor manufacturers spend as much as 15% of chip sales on R&D.

Morris Chang has been especially savvy regarding R&D spending, particularly concerning the mobile revolution. Despite his advanced age, Chang knew mobile phones would upend the world and the chip industry. If TSMC wanted to stay relevant and expand its power, it would need to be positioned to capitalize.

After initially handing over company leadership in 2006, Chang returned after his successor cut R&D spending following the 2008 financial crisis. Despite challenging economic conditions, Chang increased investment in new technologies through the 2010s to capture the mobile opportunity. 

One critical decision Chang made as part of this push was to lean into ASML’s EUV technology. In Chip War, Miller writes that no one “bet more heavily” on the technology than TSMC’s CEO. Though unproven at the time, one of Chang’s trusted lieutenants believed in its potential. Chang lavishly financed the testing and implementation of EUV, which gave TSMC the advantage in leading-edge fabrication.

Such boldness was rewarded. Though Apple initially relied on Samsung semiconductors, as TSMC advanced, it became the only fabricator of truly cutting-edge chips. Eventually, Apple turned to TSMC, making it the sole manufacturer of its iPhone processors. Apple is now the company’s largest customer. Even beyond Apple, TSMC has an estimated 84% of the leading-edge market. Without sustained R&D spending or ASML’s tooling, it could not have achieved such dominance. 

Relentless execution

“In 2016, it was unimaginable that TSMC could ever be ahead of Intel,” Brinton Johns remarked. Through the late-2010s, TSMC made up ground and eventually overtook the American behemoth largely thanks to its savvy R&D spending. But it also seems to have simply out-executed its rival. NZS’s investors noted there might be a cultural aspect to TSMC’s efficiency on this front. 

A passage in Chip War echoes this opinion. The same lieutenant that encouraged Chang to invest in EUV said, “People worked so much harder in Taiwan.” If a piece of equipment failed in the middle of the night, a worker would rush to fix it rather than wait until the following day. Employees would work through the night to push ahead on crucial projects. When maintaining uptime saves millions of dollars, and landing the next technological breakthrough results in billions, such fine margins make a difference. Notably, TSMC’s exceptional execution owes much to Chang’s second successor. Mark Liu took the reigns from Chang in 2018 and has been pivotal in maintaining standards. 

In 1987, Morris Chang’s career looked to be coming to a close. But rather than sailing off into retirement, TSMC’s founder spent the next thirty-five years creating a company that redefined an industry, galvanized technological progress, and became so essential it acts as a “chokepoint for the global economy,” according to Jon Bathgate. That importance gives TSMC extraordinary power but also puts it at the center of China and America’s battle for supremacy.

The silicon front 

As deterrence strategies go, the one advanced in “Broken Nest,” an essay written by military academics Dr. Jared McKinney and Dr. Peter Harris, is among the most direct: 

To start, the United States and Taiwan should lay plans for a targeted scorched-earth strategy that would render Taiwan not just unattractive if ever seized by force, but positively costly to maintain. This could be done most effectively by threatening to destroy facilities belonging to the Taiwan Semiconductor Manufacturing Company, the most important chipmaker in the world and China’s most important supplier.

For the threat to be credible, the authors suggest that the U.S. and Taiwanese governments state their position explicitly, draft a plan for the safe passage of skilled Taiwanese workers, and install an “automatic mechanism” engineered to detonate explosives at TSMC should an invasion occur.

Whether this is a promising military strategy or not, I don’t know. But it should be clear by now just what the destruction of TSMC would mean for the world. Even if we focus our gaze narrowly on this one company and its facilities (rather than the whole of Taiwan), it would be catastrophic. Global productivity would plummet. Rare expertise would be lost and difficult to recover. Supply chains would snarl and halt. So begin the years of darkness.

While “Broken Nest” may sound like some rogue plan, it is not entirely detached from current American strategy. Though the U.S. has not gone so far as to wire up vital facilities, it has executed a sustained economic assault against China, waged via the semiconductor industry. To be clear, China has not been a passive victim in this war. As summarized by the excellent Seminanalysis newsletter: “China intensified its covert economic warfare in the last decades through state-sponsored corporate espionage, state-sponsored hacking, dumping, and draconian restrictions for market access.” In part, Washington’s moves are in response to such actions and attempts to ensure American dominance.

In 2019, for example, Huawei was the second-largest purveyor of smartphones by volume. It was also TSMC’s second biggest customer, and a promising chip designer. To halt the Chinese company’s progress, the U.S. issued brutal sanctions blocking its access to American technology – and products that used American technology. Suddenly, Intel, TSMC, Cadence, Synops, ASML, and many other vital providers were off the menu. “We took a hundred billion dollar company to its knees,” Jon Bathgate said. 

In October this year, the U.S. made an even more aggressive move. As well outlined by The Center for Strategic and International Studies (CSIS), President Biden is putting China into a four-pronged sleeper hold: blocking access to leading-edge AI chips, the software needed to design such chips, the equipment required to make them, and the components necessary to try and engineer a domestic equipment maker. Though the focus is on AI, the sanctions are broad, including even mid-range chips (and associated technologies). It’s a comprehensive, brutal squeeze that should hamstring companies like SMIC and YMTC. “They’re toast,” Brinton Johns said of the NAND manufacturer.

The maneuver also represents a change in tact, according to the CSIS:

The Biden administration is going beyond the traditional U.S. policy of seeking to keep the United States’ two semiconductor technology nodes ahead of China and is now trying to actively degrade China’s technological maturity below its current level.

Whether the U.S. is wise to be quite so aggressive remains to be seen. While it should certainly kneecap China’s progress, it could provoke a response. “It was overstepping,” Johns said, “the CCP has been backed into a corner.” While it may not have the advantage in semiconductors, China has 77% of battery production and better access to necessary raw materials. It also provides the U.S. with a hefty percentage of pharmaceutical ingredients. If it wishes, Beijing could try to pressure America through these avenues and others. 

Though an exciting development from one vantage, it’s hard not to see TSMC’s $40 billion investment in Arizona as another salvo at the silicon front. Another company in another industry might be able to finance a project of this magnitude without it meaning so much – but not in semiconductors. As the U.S. pushes China backward, it is tightening its own grip.

NZS’s Bathgate described the Arizona facility as a “rounding error” compared to TSMC’s Taiwan production, albeit before the recent announcement tripling the company’s investment. It has, presumably, ticked up a few deciles, and it may become much larger in time. Bathgate remarked that the initiative may snowball, with two factories turning into six or more.

What happens next?

As we’ve written previously, 2022 is the year of AI. Rapid evolutions are changing how we develop drugs, code, and build. Advances like ChatGPT may usher in an age of endless media in which algorithms generate the books, games, movies, and TV shows we enjoy. 

Progress in AI doesn’t come cheaply, primarily because of the amount of computing it requires. By some estimates, training GPT-3 – a newly outdated model – would cost $4.6 million. OpenAI’s upcoming GPT-4 model may cost orders of magnitude more. Critically, computing power is a crucial input when making progress in the field. OpenAI has outlined this, demonstrating the exponential growth in compute used to train models. 

If we believe AI is critical to human advancement, computing power is essential. In turn, this means that semiconductors are likely to become even more critical in the years to come. Time will tell what this effect has on the chip industry. 

Notably, AI relies on different chips. Usually, that means “graphics processing units,” known as GPUs. Nvidia is a leading designer of GPUs, though it does not manufacture them. In addition to more standardized GPUs, recent years have seen the emergence of customized AI chips designed to improve processing and reduce costs. Google invested in creating “tensor processing units,” abbreviated to TPUs, that better dovetail with its TensorFlow machine learning software. Amazon’s acquisition of Israeli chip-designer Annapurna Labs laid the groundwork for its release of “Inferentia,” a bespoke semiconductor designed for deep learning.

When you first hear about these innovations, there’s an inclination to wonder if they might disrupt incumbents. TSMC shot ahead of Intel thanks to its savvy handling of the mobile opportunity in the 2010s. Perhaps, the AI revolution could give rivals a similar chance?

Maybe. But dig a layer deeper, and it becomes clear that these innovations rely on TSMC just as much as the last generation of chips, if not more so. Nvidia may be based in California, but it depends on TSMC’s Taiwanese facilities for manufacturing. Though it’s less clear where Google and Amazon fabricate their creations, both are mooted to use TSMC to some degree. 

Even from a structural standpoint, it’s hard to see Morris Chang’s outfit being displaced. For one thing, the chip industry rewards scale. Only the biggest players can afford the equipment, operational expenses, and R&D spending necessary to compete. This is especially true when it comes to leading-edge semiconductors. Newcomers or laggards don’t have the money to produce game-changing AI chips. You need pockets as deep as TSMC. (Even that is often not enough, as China’s struggles with SMIC and YMTC have shown.)

Notably, TSMC is well-equipped to handle variety. If Google and Amazon’s forays indicate a broader trend, we might see more companies attempt to build custom semiconductors. Because it is a pure-play foundry, TSMC is a credibly neutral collaborator. As it has done in the past, it can use its surrounding ecosystem to improve processes in this nascent industry and perhaps introduce valuable forms of standardization. 

Science fiction author Arthur C. Clarke famously said that “any sufficiently advanced technology is indistinguishable from magic.”

Few creations so aptly encapsulate this as semiconductors. These are magical objects: tiny stamp-sized squares, enchanted with light. They are minuscule, gilded brains packed with memory and logic and latent life. They are the marvelous embers that impact how we move, learn, fight, and live. And the best of them are made by just one company on an island no greater than 90 miles across. If TSMC’s $40 billion investment in Arizona succeeds and is supplemented with further financing in the coming years, the ability to manufacture magic may flourish in America.