"That Competition Is Tough"

On the morning of March 25, 1980, Richard W. Anderson stood at the front of a hotel conference room in Washington and prepared to ruin the day for several hundred American chip executives. Anderson ran Hewlett-Packard’s Data Systems Division. He was an engineer, not a polemicist, and he had brought with him only a stack of overhead transparencies. The audience belonged to the Electronic Industries Association, the trade group that spoke for American electronics. The men in the room had spent the late 1970s telling each other a story in which the United States was the planet’s leading producer of memory chips, and the Japanese were lurching along somewhere behind, a generation late and a step slow.

Anderson started with a simple problem. Beginning in 1977, HP had needed an enormous quantity of 16-kilobit dynamic random-access memory chips, the workhorse memory of the era, to put inside its minicomputers. Domestic suppliers could not produce enough. HP had quietly added three Japanese vendors to its approved list: NEC, Hitachi, and Fujitsu. HP then did what HP did to every supplier. It tested incoming parts and tracked field failures with the obsessive bookkeeping the company had always brought to instruments. The numbers had now built up over three years.

Anderson clicked to his next transparency. At the receiving dock, the worst Japanese vendor’s incoming-defect rate was lower than the best American vendor’s. In the field, after a thousand hours of operation, the Japanese parts failed at a rate roughly an order of magnitude below the American parts. Some Japanese batches showed effectively zero failures. The best US chips, on the same bench, were running six times worse than the worst Japanese chips. The Japanese chips cost about the same.

There was a long pause in the room. American engineers had assumed that Japan’s edge, if it existed at all, was a few cents on labor and a vague advantage in factory discipline. Anderson’s data said something different. On the metric that mattered to a customer putting a chip inside a million-dollar computer, the Japanese had already won.

The talk circulated in industry newsletters within weeks and entered legend as the “Anderson Bombshell.” It mattered because HP was nobody’s pawn. The company had no incentive to flatter Japan; it was an American firm that wanted to buy from American suppliers. Anderson was not asking for tariffs or sympathy. He was reporting a result. And the result, once stated in front of a room full of customers, could not be unstated. Within months, Detroit and Boston were asking the same questions about their suppliers as HP had asked about its own.

To understand how the United States had been overtaken in three years, one had to look back to a Tokyo conference room in late 1975, where bureaucrats from the Ministry of International Trade and Industry, usually called MITI, sat down with engineers from Japan’s five biggest electronics companies, ordinarily ferocious rivals: NEC, Toshiba, Hitachi, Fujitsu, and Mitsubishi Electric. MITI’s director-general for machinery and information industries had concluded that Japan’s domestic computer industry could not survive a coming generation of IBM mainframes without world-class memory chips, and that none of the five companies had the capital or the courage to develop those chips alone. He proposed a four-year cooperative program: a Very Large Scale Integration laboratory, jointly staffed and jointly funded, in which the rivals would share fundamental process technology, return to their own labs to commercialize it, and then go back to fighting in the market.

The VLSI Project ran from 1976 to 1980. Its budget came in around 72 billion yen, of which the government contributed roughly 30 billion as conditional subsidies. The program rented a research building near Kawasaki, seconded about a hundred of the five firms’ best engineers, and pointed them at the unsolved problems of the next two memory generations: the 64K and 256K DRAM. The lab did not design products. It built shared knowledge of electron-beam lithography, silicon crystal growth, photoresists, and chemical-vapor deposition. The participating engineers, by all later accounts, hated each other for the first year and worked side by side by the third. When the program closed in March 1980, MITI counted around a thousand patents, though the more important output was a generation of engineers who now shared a common technical vocabulary across five companies that, on paper, were trying to destroy one another.

While the lab worked on the underlying physics, the five firms poured capital into clean-room construction. Between 1976 and 1980, Japanese chip-equipment imports surged and then collapsed because Japan started building its own. MITI brokered orders for Nikon and Canon, which turned themselves from camera companies into precision-stepper companies. By 1980, Nikon shipped its first commercial wafer stepper, the NSR-1010G; the first customers were NEC and Toshiba. Lithographic alignment, the most expensive single tool in a fab, had begun to migrate to Japanese suppliers within four years.

When the global market for the 64K DRAM opened up around 1980, the Japanese were not playing catch-up. They had factories running, with steppers they understood intimately because they had helped design them, staffed by engineers a national program had let stand on five companies’ shoulders at once. They could ramp volume at a speed Americans, who had pioneered every previous DRAM generation, could not match.

The American DRAM industry had, in fact, dominated every previous generation. The 1K DRAM was Intel’s invention; the 4K had been a Mostek triumph. By the late 1970s, Mostek, a Texas company spun out of Texas Instruments, owned roughly 85 percent of the world DRAM market. Intel, TI, Motorola, National Semiconductor, AMD, Mostek and a half dozen smaller firms had treated memory as a domestic specialty. The market was assumed to be theirs.

Then, in 1981, the Japanese share of the 64K DRAM market crossed 70 percent. The Americans had been overtaken in a single product generation. By 1985, Japanese firms were producing more than 60 percent of the world’s 64K DRAMs and over 90 percent of the more advanced 256K parts. The five MITI companies and a few smaller players had pushed the Americans almost entirely off the leading-edge memory map.

The reasons compounded. Quality, as Anderson had documented, was the most visible. Underneath sat a manufacturing story, and underneath that, a capital story.

Japanese DRAM yields, the percentage of usable chips per silicon wafer, were running 70 to 80 percent at a moment when American yields were still 50 to 60 percent. The gap was the difference between making money on a chip and losing money on it. Yield came from process discipline: cleaner clean rooms, tighter monitoring, more conservative scaling, and an unending series of small incremental improvements pushed from the factory floor. NEC and Hitachi factory tours in the early 1980s showed Americans something they had not expected. Operators, not engineers, were filling out statistical process-control charts on the wall. Defects had been treated as engineering problems in California; in Hyogo and Kawasaki, they were treated as everyone’s problem, and the stamp of W. Edwards Deming, the American statistician Japan had embraced and America had ignored, was visible on every line. By the time American memory plants tried to copy the practices, the Japanese had a five-year head start in compounding them.

The capital story was less visible but more decisive. Building a state-of-the-art memory fab in 1981 cost roughly $200 million, a number that would double every couple of years. Funding such a plant in the United States meant going to public equity markets and banks demanding double-digit interest rates from a Federal Reserve fighting inflation. American chip companies faced an effective cost of capital north of 20 percent in the early 1980s. Japanese firms, sitting inside keiretsu groupings, borrowed from in-house banks at single-digit rates and faced no quarterly earnings pressure to speak of. Hitachi’s chip business was a line in a conglomerate that also sold turbines and trains; if memory lost money for two years to capture share, the parent company shrugged. NEC operated under the same logic. The Japan Development Bank’s actual concessionary lending to the industry was modest, well under one percent of plant investment in any given year. The structural advantage was not direct subsidy. It was patient money.

What patient money allowed was the strategy that began to unsettle the American executives more than any other. In the spring of 1981, copies of an internal Hitachi sales memo began circulating in California. The memo instructed the company’s American sales force to “win with the 10% rule”: quote ten percent below any competing American 64K DRAM bid, no matter what the price was. If the American came back at a lower number, quote ten percent below that. The memo embarrassed Hitachi when the Justice Department got hold of it, and it would later anchor a long succession of dumping complaints. But to a chief executive in Sunnyvale, the memo was less a legal exhibit than a forecast. It said, in plain language, that the Japanese were prepared to lose money on every 64K chip until the Americans went away.

By 1984, they nearly had. The 64K DRAM that sold for around $3.50 in early 1984 was selling for about 35 cents by mid-1985. A ten-times collapse in eighteen months. No manufacturing improvement explained it; there was only a war of attrition. Mostek, the former 85-percent leader, had been bought by United Technologies in 1979 for $345 million and would be closed and sold off for about $71 million in October 1985, on the way to a French conglomerate. Intel, which had invented the DRAM, watched its memory market share fall from over 80 percent in the mid-1970s to barely one percent by 1984. National Semiconductor’s memory line was a wound that wouldn’t close. Charlie Sporck, National’s chief executive, who in 1977 had cofounded the Semiconductor Industry Association with Bob Noyce, Wilf Corrigan of Fairchild, Jerry Sanders of AMD, and John Welty of Motorola, watched his company hemorrhage and, in interviews afterward, came back to a single phrase about the Japanese: competition is tough. He used the words almost gently. He had watched American semiconductor firms beat each other for fifteen years in market battles he understood. This was something else.

Andy Grove, by then president of Intel, had been receiving troubling reports from Japan since 1981. Engineers returning from visits described an entire Hitachi office tower devoted to memory, with a separate floor of designers working on each generation simultaneously: 64K teams above, 256K below, 1M below them, 4M below them. Intel had a single team trying to do all of it. Grove told colleagues he had come to think of the Japanese the way he had once thought of the Soviet army: bigger, slower, but never running out of people. Intel kept losing money on memory through 1983 and 1984. In the spring of 1985, Grove was still inside the loss.

The American executives knew by then what they were watching. The trade press through 1984 and 1985 read like a casualty list. Plants in Texas idled; lines in Idaho shrank. The semiconductor industry, which had hired aggressively for a decade, laid off more than 60,000 workers in 1985 and 1986. The Semiconductor Industry Association, founded as a domestic-policy club, had quietly turned itself into a Washington trade-war operation. By June 1985 the SIA had filed a Section 301 unfair-trade-practices complaint with the Reagan White House. Three weeks later Micron, a small Idaho memory company that had so far refused to die, filed an antidumping case against the Japanese. The political fight that followed would consume Washington for the next two years.

What mattered was the speed. In 1976, when MITI had convened its five rivals in Kawasaki, the United States made roughly 100 percent of the world’s leading-edge DRAMs. By 1981 it made under 30 percent of the new 64K generation. By 1985 it made under 10 percent of the 256K. A position built over twenty years had been lost in five. The technological lead had not been overtaken by a faster runner; it had been outflanked by a manufacturing system the runner had not realized was a different sport.

The lesson would be relearned in many industries and many countries over the following decades. A technological lead is not a cushion. It is a temporary lease on rented ground, paid for by next year’s investment. Quality is a manufacturing discipline, not a technical secret, and it compounds. Patient capital wins long games against impatient capital. None of those conclusions were popular in the American chip industry in the spring of 1985. They were the conclusions everyone was about to be forced to draw.

Sporck, in his book years later, would describe walking through a Japanese fab in 1981 and noticing, with the half-laugh of a man who had already lost, that the floor was so clean the workers wore the same slippers in the cafeteria as on the line. He had spent his career running American factories. He understood, by then, that he was not looking at espionage or subsidy or unfair trade. He was looking at a competitor that had decided to be better, and had organized itself, patiently, for a decade, to make that decision come true. The competition, he said, was tough. He was being polite. By the spring of 1985, with American memory chips selling at a fraction of their cost and the SIA’s lawyers drafting petitions in Washington, almost everyone on the American side had begun to wonder whether toughness was even the right word.