Chip War: The Fight for the World's Most Critical Technology

by Chris Miller

One tiny speck of Chinese territory escaped the horrors of the Cultural Revolution. Thanks to a quirk of colonialism, Hong Kong was still governed temporarily by the British. As most Chinese were meticulously memorizing the quotations of their crazed chairman, Hong Kong workers were diligently assembling silicon components at Fairchild’s plant overlooking Kowloon Bay. A couple hundred miles away in Taiwan, multiple U.S. chip firms had facilities employing thousands of workers in jobs that were low-paying by California’s standards but far better than peasant farming. Just as Mao was sending China’s small set of skilled workers to the countryside for socialist reeducation, the chip industry in Taiwan, South Korea, and across Southeast Asia was pulling peasants from the countryside and giving them good jobs at manufacturing plants.

During the decade in which China had descended into revolutionary chaos, Intel had invented microprocessors, while Japan had grabbed a large share of the global DRAM market. China accomplished nothing beyond harassing its smartest citizens.

The geography of chip fabrication shifted drastically over the 1990s and 2000s. U.S. fabs made 37 percent of the world’s chips in 1990, but this number fell to 19 percent by 2000 and 13 percent by 2010. Japan’s market share in chip fabrication collapsed, too. South Korea, Singapore, and Taiwan each poured funds into their chip industries and rapidly increased output. For example, Singapore’s government funded fabrication facilities and chip design centers in partnership with companies like Texas Instruments, Hewlett-Packard, and Hitachi, building a vibrant semiconductor sector in the city-state.

South Korea’s semiconductor industry did even better. After dethroning Japan’s DRAM producers and becoming the world’s leading memory chipmaker in 1992, Samsung grew rapidly through the rest of that decade. It fended off competition in the DRAM market from Taiwan and Singapore, benefitting from formal government support and from unofficial government pressure on South Korea’s banks to provide credit. This financing mattered because Samsung’s main product, DRAM memory chips, required brute financial force to reach each successive technology node—spending that had to be sustained even during industry downturns.

Chang’s strategy was simple: do as TSMC had done. In Taiwan, TSMC had hired the best engineers it could find, ideally with experience at American or other advanced chip firms. TSMC bought the best tools it could afford. It focused relentlessly on training its employees in the industry’s best practices. And it took advantage of all the tax and subsidy benefits that Taiwan’s government was willing to provide.

SMIC’s local engineers learned quickly, and were soon perceived to be so capable they began receiving job offers from foreign chipmakers. The company’s success in domesticating technology was only possible thanks to this foreign-trained workforce.

Now TSMC had competition from multiple foundries in different countries in East Asia. Singapore’s Chartered Semiconductor, Taiwan’s UMC and Vanguard Semiconductor, and South Korea’s Samsung—which entered the foundry business in 2005—were also competing with TSMC to produce chips designed elsewhere. Most of these companies were subsidized by their governments, but this made chip production cheaper, benefitting the mostly American fabless semiconductor designers they served. Fabless firms, meanwhile, were in the early stages of launching a revolutionary new product chock-full of complex chips: the smartphone. Offshoring had reduced manufacturing costs and spurred more competition. Consumers benefitted from low prices and from previously unthinkable devices. Wasn’t this exactly how globalization was designed to work?

By 1992, Intel was again the world’s biggest chipmaker, on the strength of Grove’s decision to focus Intel’s efforts on microprocessors for PCs. It was flush with cash and as committed as ever to Moore’s Law.

Whereas Japanese competitors tried to build everything in-house, ASML could buy the best components on the market. As it began to focus on developing EUV tools, its ability to integrate components from different sources became its greatest strength. ASML’s second strength, unexpectedly, was its location in the Netherlands. In the 1980s and 1990s, the company was seen as neutral in the trade disputes between Japan and the United States. U.S. firms treated it like a trustworthy alternative to Nikon and Canon. For example, when Micron, the American DRAM startup, wanted to buy lithography tools, it turned to ASML rather than relying on one of the two main Japanese suppliers, each of which had deep ties with Micron’s DRAM competitors in Japan.

The scientific networks that produced EUV spanned the world, bringing together scientists from countries as diverse as America, Japan, Slovenia, and Greece. However, the manufacturing of EUV wasn’t globalized, it was monopolized. A single supply chain managed by a single company would control the future of lithography.

By 2006, Intel already supplied the processors for most PCs, having spent the previous decade successfully fending off AMD, the only other major company producing chips on the x86 instruction set architecture—a foundational set of rules that govern how chips compute—that was the industry standard for PCs.

The x86 architecture dominated PCs not because it was the best, but because IBM’s first personal computer happened to use it.

Today, nearly every major data center uses x86 chips from either Intel or AMD. The cloud can’t function without their processors.

However, Arm’s simplified, energy-efficient architecture quickly became popular in small, portable devices that had to economize on battery use. Nintendo chose Arm-based chips for its handheld video games, for example, a small market that Intel never paid much attention to. Intel’s computer processor oligopoly was too profitable to justify thinking about niche markets. Intel didn’t realize until too late that it ought to compete in another seemingly niche market for a portable computing device: the mobile phone.

Intel’s dilemma could have been easily diagnosed by the Harvard professor who’d advised Andy Grove. Everyone at Intel knew Clayton Christensen and his concept of “the innovator’s dilemma.” However, the company’s PC processor business looked likely to print money for a very long time. Unlike in the 1980s, when Grove reoriented Intel away from DRAM at a time when the company was bleeding money, in the 1990s and 2000s, Intel was one of America’s most profitable firms. The problem wasn’t that no one realized Intel ought to consider new products, but that the status quo was simply too profitable.

Shortly after the deal to put Intel’s chips in Mac computers, Jobs came back to Otellini with a new pitch. Would Intel build a chip for Apple’s newest product, a computerized phone? All cell phones used chips to run their operating systems and manage communication with cell phone networks, but Apple wanted its phone to function like a computer. It would need a powerful computer-style processor as a result. “They wanted to pay a certain price,” Otellini told journalist Alexis Madrigal after the fact, “and not a nickel more…. I couldn’t see it. It wasn’t one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100× what anyone thought.” Intel turned down the iPhone contract. Apple looked elsewhere for its phone chips. Jobs turned to Arm’s architecture, which unlike x86 was optimized for mobile devices that had to economize on power consumption. The early iPhone processors were produced by Samsung, which had followed TSMC into the foundry business. Otellini’s prediction that the iPhone would be a niche product proved horribly wrong. By the time he realized his mistake, however, it was too late. Intel would later scramble to win a share of the smartphone business. Despite eventually pouring billions of dollars into products for smartphones, Intel never had much to show for it. Apple dug a deep moat around its immensely profitable castle before Otellini and Intel realized what was happening.

So Intel never found a way to win a foothold in mobile devices, which today consume nearly a third of chips sold. It still hasn’t.

The company’s leadership consistently prioritized the production of chips with the highest profit margin. This was a rational strategy—no one wants products with low profit margins—but it made it impossible to try anything new. A fixation on hitting short-term margin targets began to replace long-term technology leadership. The shift in power from engineers to managers accelerated this process. Otellini, Intel’s CEO from 2005 to 2013, admitted he turned down the contract to build iPhone chips because he worried about the financial implications. A fixation on profit margins seeped deep into the firm—its hiring decisions, its product road maps, and its R&D processes. The company’s leaders were simply more focused on engineering the company’s balance sheet than its transistors. “It had the technology, it had the people,” one former finance executive at Intel reminisced. “It just didn’t want to take the margin hit.”

In the early 2010s, Intel retained the world’s most advanced semiconductor process technology, introducing smaller transistors before rivals, with the same regular cadence it had been known for since the days of Gordon Moore. However, the gap between Intel and rivals like TSMC and Samsung had begun to shrink.

“Run faster” was an elegant strategy with only a single problem: by some key metrics, the U.S. wasn’t running faster, it was losing ground. Hardly anyone in government bothered to do the analysis, but Andy Grove’s gloomy predictions about the offshoring of expertise were partially coming true. In 2007, the Defense Department commissioned a study from former Pentagon official Richard Van Atta and several colleagues to assess the impact of semiconductor industry “globalization” on the military’s supply chains. Van Atta had worked on defense microelectronics for several decades and had lived through the rise and fall of Japan’s chip industry. He wasn’t prone to overreaction and understood how a multinational supply chain made the industry more efficient. In peacetime, this system worked smoothly. However, the Pentagon had to think about worst-case scenarios. Van Atta reported that the Defense Department’s access to cutting-edge chips would soon depend on foreign countries because so much advanced fabrication was moving abroad. Amid the hubris of America’s unipolar moment, hardly anyone was willing to listen. Most people in Washington simply concluded the U.S. was “running faster” without even glancing at the evidence. However, the history of the semiconductor industry didn’t suggest that U.S. leadership was guaranteed. America hadn’t outrun the Japanese in the 1980s, though it did in the 1990s. GCA hadn’t outrun Nikon or ASML in lithography. Micron was the only DRAM producer ...

Morris Chang had drawn a similar conclusion several decades earlier, which is why he thought TSMC’s business model was superior. A foundry like TSMC could fabricate chips for many chip designers, wringing out efficiencies from its massive production volumes that other companies would find difficult to replicate.

Today, building an advanced logic fab costs $20 billion, an enormous capital investment that few firms can afford.

Given the benefits of scale, the number of firms fabricating advanced logic chips has shrunk relentlessly.

So long as Sanders was CEO, AMD, the company he founded, stayed in the business of manufacturing logic chips, like processors for PCs. Old-school Silicon Valley CEOs kept insisting that separating the fabrication of semiconductors from their design caused inefficiencies. But it was culture, not business reasoning, that kept chip design and chip fabrication integrated for so long.

Huang gave away CUDA for free, but the software only works with Nvidia’s chips. By making the chips useful beyond the graphics industry, Nvidia discovered a vast new market for parallel processing, from computational chemistry to weather forecasting. At the time, Huang could only dimly perceive the potential growth in what would become the biggest use case for parallel processing: artificial intelligence. Today Nvidia’s chips, largely manufactured by TSMC, are found in most advanced data centers.

For each generation of cell phone technology after 2G, Qualcomm contributed key ideas about how to transmit more data via the radio spectrum and sold specialized chips with the computing power capable of deciphering this cacophony of signals. The company’s patents are so fundamental it’s impossible to make a cell phone without them.

Qualcomm has made hundreds of billions of dollars selling chips and licensing intellectual property. But it hasn’t fabricated any chips: they’re all designed in-house but fabricated by companies like Samsung or TSMC.

Thanks to TSMC, Samsung, and other companies willing to produce their chips, Qualcomm’s engineers could focus on their core strengths in managing spectrum and in semiconductor design.

The biggest change, however, wasn’t simply new types of chips. By making possible mobile phones, advanced graphics, and parallel processing, fabless firms enabled entirely new types of computing.

The new class of CEOs who took over America’s semiconductor firms in the 2000s and 2010s tended to speak the language of MBAs as well as PhDs, chatting casually about capex and margins with Wall Street analysts on quarterly earnings calls.

Around the early 2010s, it became unfeasible to pack transistors more densely by shrinking them two dimensionally. One challenge was that, as transistors were shrunk according to Moore’s Law, the narrow length of the conductor channel occasionally caused power to “leak” through the circuit even when the switch was off. On top of this, the layer of silicon dioxide atop each transistor became so thin that quantum effects like “tunneling”—jumping through barriers that classical physics said should be insurmountable—began seriously impacting transistor performance. By the mid-2000s, the layer of silicon dioxide on top of each transistor was only a couple of atoms thick, too small to keep a lid on all the electrons sitting in the silicon.

Complications arose, though, because part of Samsung’s operation involved building chips that it designed in-house. Whereas a company like TSMC builds chips for dozens of customers and focuses relentlessly on keeping them happy, Samsung had its own line of smartphones and other consumer electronics, so it was competing with many of its customers. Those firms worried that ideas shared with Samsung’s chip foundry might end up in other Samsung products. TSMC and GlobalFoundries had no such conflicts of interest.

Morris Chang wasn’t about to give up dominance of the foundry business, though. He’d lived through every industry cycle since his old colleague Jack Kilby invented the integrated circuit. He was sure this downturn would eventually end, too. Companies that were overextended would be pushed out of business, leaving those that invested during the downturn positioned to grab market share. Moreover, Chang realized as early as anyone how smartphones would transform computing—and therefore how they would change the chip industry, too. The media focused on young tech tycoons like Facebook’s Mark Zuckerberg, but seventy-seven-year-old Chang had a perspective that few could match. Mobile devices would be a “game-changer” for the chip industry, he told Forbes, perceiving them as heralding shifts as significant as the PC had brought. He was committed to winning the lion’s share of this business, whatever the cost.

That’s the power of the Grand Alliance.” The financial implications of this were profound. “The combined R&D spending of TSMC and its ten biggest customers,” he bragged “exceeds that of Samsung and Intel together.”

TSMC’s position at the center of the semiconductor universe required it to have capacity to produce chips for all its biggest customers. Doing so wouldn’t be cheap. Amid the financial crisis, Chang’s handpicked successor, Rick Tsai, had done what nearly every CEO did—lay off employees and cut costs. Chang wanted to do the opposite. Getting the company’s 40nm chipmaking back on track required investing in personnel and technology. Trying to win more smartphone business—especially that of Apple’s iPhone, which launched in 2007 and which initially bought its key chips from TSMC’s archrival, Samsung—required massive investment in chipmaking capacity. Chang saw Tsai’s cost cutting as defeatist. “There was very, very little investment,” Chang told journalists afterward. “I had always thought that the company was capable of more…. It didn’t happen. There was stagnation.” So Chang fired his successor and retook direct control of TSMC. The company’s stock price fell that day, as investors worried he’d launch a risky spending program with uncertain returns. Chang thought the real risk was accepting the status quo. He wasn’t about to let a financial crisis threaten TSMC in the race for industry leadership. He had a half-century-long track record at chipmaking, a reputation he’d honed since the mid-1950s. So at the depths of the crisis Chang rehired the workers the former CEO had laid off and doubled down on investment in new capacity and R&D. He announced several multibillion-dollar ...

The greatest beneficiary of the rise of foundries like TSMC was a company that most people don’t even realize designs chips: Apple. The company Steve Jobs built has always specialized in hardware, however, so it’s no surprise that Apple’s desire to perfect its devices includes controlling the silicon inside. Since his earliest days at Apple, Steve Jobs had thought deeply about the relationship between software and hardware. In 1980, when his hair nearly reached his shoulders and his mustache covered his upper lip, Jobs gave a lecture that asked, “What is software?” “The only thing I can think of,” he answered, “is software is something that is changing too rapidly, or you don’t exactly know what you want yet, or you didn’t have time to get it into hardware.”

As Jobs introduced new versions of the iPhone, he began etching his vision for the smartphone into Apple’s own silicon chips. A year after launching the iPhone, Apple bought a small Silicon Valley chip design firm called PA Semi that had expertise in energy-efficient processing. Soon Apple began hiring some of the industry’s best chip designers. Two years later, the company announced it had designed its own application processor, the A4, which it used in the new iPad and the iPhone 4.

Now Apple not only designs the main processors for most of its devices but also ancillary chips that run accessories like AirPods. This investment in specialized silicon explains why Apple’s products work so smoothly.

Like Qualcomm and the other chip firms that powered the mobile revolution, even though Apple designs ever more silicon, it doesn’t build any of these chips. Apple is well known for outsourcing assembly of its phones, tablets, and other devices to several hundred thousand assembly line workers in China, who are responsible for screwing and gluing tiny pieces together. China’s ecosystem of assembly facilities is the world’s best place to build electronic devices. Taiwanese companies, like Foxconn and Wistron, that run these facilities for Apple in China are uniquely capable of churning out phones, PCs, and other electronics. Though the electronics assembly facilities in Chinese cities like Dongguan and Zhengzhou are the world’s most efficient, however, they aren’t irreplaceable. The world still has several hundred million subsistence farmers who’d happily fasten components into an iPhone for a dollar an hour. Foxconn assembles most of its Apple products in China, but it builds some in Vietnam and India, too.

As semiconductor fabrication capacity migrated to Taiwan and South Korea, so too did the ability to produce many of these chips. Application processors, the electronic brain inside each smartphone, are mostly produced in Taiwan and South Korea before being sent to China for final assembly inside a phone’s plastic case and glass screen. Apple’s iPhone processors are fabricated exclusively in Taiwan. Today, no company besides TSMC has the skill or the production capacity to build the chips Apple needs. So the text etched onto the back of each iPhone—“Designed by Apple in California. Assembled in China”—is highly misleading. The iPhone’s most irreplaceable components are indeed designed in California and assembled in China. But they can only be made in Taiwan.

Apple isn’t the only company in the semiconductor business with a bewilderingly complex supply chain. By the late-2010s, ASML, the Dutch lithography company, had spent nearly two decades trying to make extreme-ultraviolet lithography work. Doing so required scouring the world for the most advanced components, the purest metals, the most powerful lasers, and the most precise sensors. EUV was one of the biggest technological gambles of our time. In 2012, years before ASML had produced a functional EUV tool, Intel, Samsung, and TSMC had each invested directly in ASML to ensure the company had the funding needed to continue developing EUV tools that their future chipmaking capabilities would require. Intel alone invested $4 billion in ASML in 2012, one of the highest-stakes bets the company ever made, an investment that followed billions of dollars of previous grants and investments Intel had spent on EUV, dating back to the era of Andy Grove.

It took Trumpf a decade to master these challenges and produce lasers with sufficient power and reliability. Each one required exactly 457,329 component parts.

Ultimately, Zeiss created mirrors that were the smoothest objects ever made, with impurities that were almost imperceptibly small. If the mirrors in an EUV system were scaled to the size of Germany, the company said, their biggest irregularities would be a tenth of a millimeter.

The company had no choice but to rely on a single source for the key components of an EUV system. To manage this, ASML drilled down into its suppliers’ suppliers to understand the risks. ASML rewarded certain suppliers with investment, like the $1 billion it paid Zeiss in 2016 to fund that company’s R&D process. It held all of them, however, to exacting standards. “If you don’t behave, we’re going to buy you,” ASML’s CEO Peter Wennink told one supplier. It wasn’t a joke: ASML ended up buying several suppliers, including Cymer, after concluding it could better manage them itself.

The result was a machine with hundreds of thousands of components that took tens of billions of dollars and several decades to develop. The miracle isn’t simply that EUV lithography works, but that it does so reliably enough to produce chips cost-effectively.

EUV machines cost over $100 million each, so every hour one is offline costs chipmakers thousands of dollars in lost production.

Producing advanced semiconductors, however, has relied on some of the most complex machinery ever made. ASML’s EUV lithography tool is the most expensive mass-produced machine tool in history, so complex it’s impossible to use without extensive training from ASML personnel, who remain on-site for the tool’s entire life span. Each EUV scanner has an ASML logo on its side. But ASML’s expertise, the company readily admits, was its ability to orchestrate a far-flung network of optics experts, software designers, laser companies, and many others whose capabilities were needed to make the dream of EUV a reality. It’s easy to lament the offshoring of manufacturing, as Andy Grove did during the final years of his life. That a Dutch company, ASML, had commercialized a technology pioneered in America’s National Labs and largely funded by Intel would undoubtedly have rankled America’s economic nationalists, had any been aware of the history of lithography or of EUV technology. Yet ASML’s EUV tools weren’t really Dutch, though they were largely assembled in the Netherlands. Crucial components came from Cymer in California and Zeiss and Trumpf in Germany. And even these German firms relied on critical pieces of U.S.-produced equipment. The point is that, rather than a single country being able to claim pride of ownership regarding these miraculous tools, they are the product of many countries. A tool with hundreds of thousands of parts has many fathers.

Having worked in Texas and California as well as in Taiwan, Chiang was always struck by the ambition and the work ethic that drove TSMC. The ambition stemmed from Morris Chang’s vision of world-beating technology, evident in his willingness to spend huge sums expanding TSMC’s R&D team from 120 people in 1997 to 7,000 in 2013. This hunger permeated the entire company. “People worked so much harder in Taiwan,” Chiang explained. Because manufacturing tools account for much of the cost of an advanced fab, keeping the equipment operating is crucial for profitability. In the U.S., Chiang said, if something broke at 1 a.m., the engineer would fix it the next morning. At TSMC, they’d fix it by 2 a.m. “They do not complain,” he explained, and “their spouse does not complain” either. With Chiang back in charge of R&D, TSMC charged forward toward EUV. He had no difficulty finding employees to work all night long. He requested that three EUV scanners for testing purposes be built in the middle of one of the company’s biggest facilities, Fab 12, and in the company’s partnership with ASML he spared no expense in testing and improving EUV tools.

By 2015, thanks to these acquisitions, GlobalFoundries was by far the biggest foundry in the United States and one of the largest in the world, but it was still a minnow compared to TSMC. GlobalFoundries competed with Taiwan’s UMC for status as the world’s second-largest foundry, with each company having about 10 percent of the foundry marketplace. However, TSMC had over 50 percent of the world’s foundry market. Samsung only had 5 percent of the foundry market in 2015, but it produced more wafers than anyone when its vast production of chips designed in-house (for example, memory chips and chips for smartphone processors) were included. Measured by thousands of wafers per month, the industry standard, TSMC had a capacity of 1.8 million while Samsung had 2.5 million. GlobalFoundries had only 700,000.

Building cutting-edge processors was too expensive for everyone except the world’s biggest chipmakers. Even the deep pockets of the Persian Gulf royals who owned GlobalFoundries weren’t deep enough. The number of companies capable of fabricating leading-edge logic chips fell from four to three.