"The Cold War Is Over and You Have Won"

On the evening of January 8, 1992, in a banquet room of the prime minister’s residence in Tokyo, the president of the United States slumped sideways in his chair and vomited into the lap of his host. George H. W. Bush had spent the morning playing doubles tennis with the emperor and his son. He had told the press he felt fine. Between courses at the state dinner, his face went grey. Kiichi Miyazawa, the Japanese prime minister, caught him as he fell. A Secret Service agent dropped to one knee. Bush, recovering in the next minute, would later joke to his physician that they should roll him under the table until the dinner was over. Japanese networks looped the video for days. American networks looped it for weeks.

Bush had not flown to Tokyo to be sick. He had flown to plead. The trip, originally pitched as a foreign-policy victory lap by a president who had presided over the disintegration of the Soviet Union, had been hastily reframed in the autumn of 1991 as a trade mission. The reframing was forced on the White House by polling that showed Americans, in the middle of a recession, did not care that he had won the Cold War. They cared that the foreman at the Cleveland plant had been laid off and that the Japanese imports in the parking lot still outsold the Chryslers. The traveling delegation Bush brought with him from Washington reflected the new framing. Twenty-one American executives flew on Air Force One, including Robert Stempel of General Motors, Harold Poling of Ford, and Lee Iacocca of Chrysler, who would spend most of the trip telling reporters that whatever the president pried out of Tokyo would not be enough.

What Bush pried out, in the end, was a set of “voluntary” Japanese pledges to buy more American auto parts and a vague promise to import more American cars by 1994. Within forty-eight hours of his return, Miyazawa was on Japanese television clarifying that the targets had been goals, not promises. Iacocca, in a January 10 speech at the Detroit Economic Club, eviscerated both the deal and the man who had brought it home. The cameras zoomed in on Iacocca slamming a fist on the lectern. Bush watched the footage from the Oval Office. By February, in the snow of New Hampshire, his approval had cratered, and the Democratic field was running against Japan as much as against him.

The candidate who said it cleanest was a former senator from Massachusetts who had been left for dead by political handicappers a year earlier. Paul Tsongas was sixty-one, a cancer survivor, an unglamorous man who carried his own suit bag through airports and gave economic speeches that could double as accounting lectures. His campaign book, A Call to Economic Arms, ran to eighty-six pages of policy. In the middle of those eighty-six pages he had buried a line that, transcribed by Maureen Dowd in The New York Times on February 17, 1992, became the political slogan of the year. “The cold war is over,” Tsongas told audiences in Manchester and Concord. “Japan won.” He won the New Hampshire primary nine days after the column appeared.

The line landed because it scanned. It packaged a decade of anxiety into seven words: that America had spent forty years and trillions of dollars hardening itself against a military adversary while a quieter adversary, hiding inside the alliance, had eaten the industries the trillions were supposed to defend. Steel. Cars. Consumer electronics. Above all the chip industry that made all of those things modern, the industry that the Pentagon had decided in the late 1970s would be the platform on which it would outfight the Soviets. By the winter of 1992, with the Nikkei still falling and the Asahi running daily features on Japan’s industrial supremacy, no one in Washington thought Tsongas was being hyperbolic.

Three years later the line would seem ridiculous. The reversal happened in a window so narrow that the people who lived through it took a decade to believe what they had seen. To explain it requires going back to a December morning in Tokyo two years before Bush vomited on Miyazawa.

On December 17, 1989, a sixty-six-year-old career central banker named Yasushi Mieno took office as the twenty-sixth governor of the Bank of Japan. Twelve days later, the Nikkei 225 closed at 38,915.87, a number that would haunt Japanese savings statements for the next thirty-four years. Within a week of taking office, Mieno raised the discount rate from 3.75 percent to 4.25 percent. He raised it four more times over the next eight months, to 6 percent by August 1990. He told colleagues that the asset bubble was a moral hazard the Bank of Japan had created and that the Bank of Japan would have to kill it. He killed it. The Nikkei lost more than a third of its value in 1990 alone. Tokyo commercial real estate, which had famously been worth more than all of California, would lose roughly 87 percent of its peak value over the following fourteen years. Total wealth destruction across Japanese stocks and real estate would eventually be estimated at fifteen hundred trillion yen, three times the country’s annual GDP.

The wealth destruction did not arrive in the chip industry as a single shock. It arrived as a slow rationing of capital. The keiretsu banks that had funded NEC, Hitachi, Toshiba, Fujitsu, and Mitsubishi through the 1980s memory wars stopped lending freely in 1991. The parent conglomerates, hemorrhaging money on real estate and bad equity bets, started asking their semiconductor divisions to justify their investment plans the way American executives had been doing for years. The patient money that had been the Japanese chip industry’s structural weapon since the Anderson bombshell in 1980 was no longer patient. By 1992 NEC’s chip group, which had cheerfully lost money on memory through three downturns to take share, was being told to make money this quarter.

The Korean producers chose that moment to do exactly what the Japanese had done a decade earlier. Samsung had developed the world’s first 64-megabit DRAM in 1992, beating NEC and Toshiba to a milestone the Japanese had assumed was theirs. By 1993 Samsung had passed every Japanese firm to become the largest DRAM maker on the planet. Hyundai Electronics and the smaller LG Semicon, both of them creatures of chaebol patient capital structurally identical to the Japanese keiretsu, ramped fabs at speeds Tokyo had once been famous for and Tokyo could no longer match. By 1995 the Korean share of world DRAM was approaching 40 percent and rising. The Japanese share was falling toward 40 percent and dropping. The cycle that the Japanese had run on the Americans was being run on the Japanese, with a strategy lifted almost without modification from MITI’s old playbook, and with cheaper engineers.

Some of those engineers were Japanese. The phenomenon went mostly unreported in the United States and obsessively documented in the Japanese trade press. Samsung, Hyundai, and the smaller Korean firms began hiring retired or near-retired Japanese process engineers as weekend consultants, paying them in cash to fly to Suwon or Icheon on Friday and fly home Sunday night. The traveling consultants brought with them, intentionally or not, the proprietary process knowledge that Japanese clean rooms had spent fifteen years compounding. Korean firms had access by the early 1990s to almost everything the VLSI Project had once kept inside Kawasaki. By the mid-1990s the asymmetry had reversed: it was Japanese fabs that were buying steppers from Korean-trained applications engineers and process recipes from Korean technology suppliers.

The reversal was not only Korean. Across the Pacific, the American memory company that had nearly died in 1985 and had refused to die, the small Boise outfit that had been the named complainant in the antidumping cases of the Reagan years, was running its third turnaround. Micron Technology, having survived the dumping war, the Sematech subsidies, the consolidation of the American memory industry into a handful of survivors, and a 1991 near-bankruptcy, finally posted a billion-dollar profit in fiscal 1995. Its trick was the trick the Japanese had taught the world and that nobody at MITI had noticed they were forgetting: relentless cost engineering, ruthless die-shrink discipline, and the willingness to keep capital expenditure flat through the up cycle so that the down cycle could be ridden out without pain. By 1996 Micron and the Korean producers together held more world DRAM share than all five of the original MITI companies combined.

While the memory war was being relost, Japan was losing a different war it had barely noticed it was fighting. In Santa Clara, the company that had walked away from the memory business in late 1985 was quietly building a monopoly in microprocessors. Intel released the 486 in 1989, the Pentium on March 22, 1993, the Pentium Pro in 1995, and a generation of “Intel Inside” stickers that turned the company’s logo into a mass-market consumer brand for the first time in any chipmaker’s history. Microsoft’s Windows 3.1, released in April 1992, and Windows 95, released in August 1995, locked the personal computer industry to the x86 instruction set in a way nothing on a Japanese designer’s whiteboard could touch.

Japanese semiconductor firms had not failed to notice the personal computer. They had simply built the wrong companies for it. The keiretsu logic that had made memory chips a winning bet for an integrated electronics conglomerate worked against logic chips and microprocessors. NEC, Toshiba, Hitachi, Fujitsu, and Mitsubishi each made microprocessors, but each of them made many other things, and inside each of them the microprocessor business competed for capital with consumer audio, with industrial motors, with mainframe computers, with white goods. None of them was prepared to bet a quarter of the company on the right architecture and a development tool chain that would lock the world’s software developers in. None of them tried to. Toshiba shipped capable processors that nobody outside Japan bought. Hitachi’s SuperH would find a niche in the Sega Saturn and then in automotive electronics, but never in PCs. NEC’s V-series x86 clones, after a complicated patent fight with Intel, faded as Intel pulled ahead on each new node.

The deepest cause was that the Japanese chip industry had been organized around a single bet, and the bet was now wrong. The bet was that memory was the strategic high ground of the digital era, that whoever made the cleanest, densest, cheapest memory chip would set the pace of every other product, and that an integrated electronics company with a captive consumer-electronics market could fund the manufacturing investment forever. The bet had been right in 1980 and it had been right in 1985. By 1992 the high ground had moved. Value was migrating from the commodity DRAM at the center of the motherboard to the proprietary microprocessor next to it, and from there to the ecosystem of compilers and operating systems and applications above it. Japanese firms had no software franchise to speak of. Their compiler teams were small. Their relationships with the American developer community were, by Silicon Valley standards, cold. The pivot they would have needed to make was not a pivot inside chip design. It was a pivot in what kind of company they were.

A handful of executives in Tokyo saw this clearly and could do nothing about it. Akio Morita of Sony, who had co-authored The Japan That Can Say No in 1989 with Shintaro Ishihara, in part to argue that Japan’s semiconductor leverage made it strategically indispensable to the United States, suffered a debilitating cerebral hemorrhage in November 1993 and never returned to active management. The book that had once read as a warning to America began, by 1994, to read as a period piece. The Japanese semiconductor leverage Morita had described was draining away in real time. By 1995, foreign firms held just under 30 percent of the Japanese semiconductor market, finally clearing the threshold the 1986 Trade Agreement’s side letter had set, and the agreement itself was scheduled to expire in 1996. Tokyo did not fight to renew it. Washington did not fight to extend it. When negotiators met in Vancouver in August 1996 to sign a successor arrangement, the new document removed the foreign-share targets entirely and replaced the bilateral monitoring with a multilateral statistics-sharing forum. The trade lawyers in attendance recognized the moment for what it was. The semiconductor war that had nearly torn the alliance apart was being quietly closed out, because the war was no longer being fought.

The numbers told the story without commentary. In 1988, at the peak, Japanese firms had held 50.3 percent of the world semiconductor market and U.S. firms had held 36.5 percent. By 1993 the U.S. had retaken the lead. By 1995 Japan was at roughly 41 percent and dropping. By 1998 it was at 26 percent. By the mid-2000s it would settle below 20 percent and keep falling toward 10 percent. The Korean share, almost zero in 1985, was approaching 25 percent by 1998. The American share stabilized in the high 40s and stayed there. The neat triangulation that the Reagan-era trade hawks had imagined, of Japan checked and Korea cultivated and America recovered, had unfolded in reality, except that the Japanese check had been more total and the Korean cultivation more successful than even the most aggressive Pentagon planner had predicted.

Inside Japan, the response was consolidation, the standard reflex of declining national champions. In 1999 the DRAM operations of NEC and Hitachi were spun out and merged into a new joint venture initially called NEC Hitachi Memory, renamed within months to Elpida Memory. The name was meant to evoke the Greek word for hope, with a “d” added for “dynamic.” Mitsubishi Electric joined the venture in 2003. By then the company was the only Japanese DRAM maker left. Elpida would survive the 2000s on government loans and scale economies that, by the standards of Samsung and Hynix, were no longer scale economies at all. In February 2012 it would file for bankruptcy with debts of 448 billion yen, the largest corporate failure in Japan since Japan Airlines two years earlier. A year later Micron, the Idaho company that had filed the original antidumping case in 1985, would buy Elpida out of bankruptcy for less than a fifth of its peak market value. The Boise potato-chip makers ended the decade owning the last Japanese DRAM fab.

Elpida was the visible end of one part of the story. The other part ended the same way. In April 2003, Hitachi and Mitsubishi Electric folded their non-memory chip operations into a joint venture called Renesas Technology, with Hitachi holding 55 percent and Mitsubishi 45 percent. NEC’s chip arm joined in 2010. The combined enterprise gathered the surviving non-memory operations of three of the original five MITI VLSI partners under a single roof, generated meaningful revenue and persistent losses, and would require a Japanese state-led bailout in 2013 to keep running. Two of the other MITI partners, Toshiba and Fujitsu, would by the late 2010s leave general-purpose semiconductor manufacturing entirely, surviving only in narrow specialties such as Toshiba’s NAND flash franchise, eventually spun out as Kioxia.

The economists watched and updated their models. Kenneth Flamm’s Mismanaged Trade?, published by Brookings in 1996 as the original Trade Agreement expired, argued that the agreement and the tariffs had done less to restore American chipmaking than the trade hawks claimed and more to accelerate the rise of Korea than anyone in Washington had intended. The intervention had cost American computer-makers real money in higher memory prices through the late 1980s. It had probably saved Micron, which mattered. It had not visibly rescued the rest of the American memory industry, which mostly disappeared anyway. What the agreement had done, Flamm wrote, was buy time, and what the time had bought was not the resurrection of American DRAM but the construction of two adjacent industries, microprocessors and packaged software, that the Japanese could not match. The Reagan hawks had asked for one thing and gotten another. The institutional advantages that had powered Japan’s rise, patient capital, lifetime employment, vertical integration, had become by the early 1990s the institutional liabilities that made it impossible to spin out a fabless startup, to write a winning compiler, or to bet a company on a single architecture.

By the second half of the decade, the language American politicians used about Japan changed. Tsongas, who lost the 1992 Democratic nomination to Bill Clinton, returned to teaching and to the Concord Coalition before dying in January 1997 from complications of his cancer. The Clinton administration, after an early flurry of trade-hawk appointments, quietly de-escalated. The trade press, which had run six years of “Japan Inc.” cover stories from 1986 through 1992, ran almost none after 1995. By 1999, when The Economist surveyed Japan, the cover line was about deflation, not domination.

The American chip industry that emerged from the trade war was leaner, more concentrated, less labor-intensive, and more profitable than the one that had preceded it. The 60,000 workers laid off in 1985 and 1986 had not, by and large, come back. The companies they had been laid off from, in many cases, had not come back either. Mostek was gone. National Semiconductor would eventually be absorbed by Texas Instruments. The recovery, on close inspection, was not a recovery of the old industry. It was a different industry that happened to occupy the same office parks. Robert Noyce had died of a heart attack in his Austin swimming pool in June 1990, two and a half years into his run as the head of Sematech, with the rebound he had helped engineer not yet visible. Andy Grove, who had pulled Intel out of memory in 1985 and built it into the dominant microprocessor company, would publish Only the Paranoid Survive in 1996. Asked in interviews through the late 1990s about Japan, Grove tended to look almost embarrassed. The country he had once likened to the Soviet army no longer looked like an army at all.

The line Tsongas had used in February 1992, the line that had carried him to a New Hampshire victory and that became the slogan of a moment, did not so much fade as invert. By 1995 a senior MITI official, asked at a quiet conference in Singapore by an American counterpart what message he should carry back to Washington about the trade war, paused for a long time and then offered a wry alternative formulation. The Cold War was over, the Japanese official said, and the Americans had won. Both of them. He meant the one they had been fighting against the Soviets, and the one they had not entirely realized they had also been fighting against Tokyo. The American counterpart did not know whether to laugh or apologize, and so did neither.

The Japanese chip industry did not disappear. It settled, painfully, into specialized niches in which it remained world-class: chemicals and resists, image sensors, NAND flash, packaging materials, certain test and deposition tools. Sony’s image sensor would dominate every flagship smartphone camera by the 2010s. Shin-Etsu and SUMCO would supply most of the world’s silicon wafers. Tokyo Electron, Screen, and Advantest would survive as essential members of the global equipment ecosystem. None of these niches generated the national prestige that 1980s memory dominance had. Japan in 2000 made about a fifth of the world’s chips. By 2020 it would make a tenth.

The first chip war ended quietly because nobody in either capital wanted to keep fighting it. The Japanese had lost and most of the Americans who had won had moved on to other arguments. By 1995 the question that had obsessed Washington for a decade, whether the United States could remain a chip-making power, had been answered, ambiguously and partially, in the affirmative. The next question, the one that would consume the next thirty years, was already forming a continent away, in a Taiwan industrial park where a former Texas Instruments executive named Morris Chang had spent the late 1980s building a company that nobody in Tokyo or Washington had yet learned to take seriously. The answer to America’s first chip war turned out to be the precondition for its next one. The world that emerged from the rubble of the Japanese semiconductor industry was a world in which design and manufacturing had been split apart, and the manufacturing had not come home to America. It had gone somewhere else.