"Real Men Have Fabs"

The lunch was running long, and the waiters were beginning to clear plates around an audience that had stopped pretending to eat. It was the spring of 1992, at one of the In-Stat semiconductor forums that ran each year out of Phoenix and Scottsdale, and the speaker at the head table had spent half an hour working up to a punchline. Walter Jeremiah Sanders III, the founder and chief executive of Advanced Micro Devices, was a tanned, silver-haired man in a dark double-breasted suit with French cuffs the size of playing cards, the only person in the room who looked as if he had stepped out of a Vegas headliner ad rather than a chip-industry conference. He paced. He paused. He reminded the audience, by way of preamble, that he had been in this business since before some of them had been born, that AMD had been founded the year men first walked on the moon, and that in all that time he had learned one thing that mattered more than any other: a company that did not own its own factories was not really in the semiconductor business.

Then he leaned into the microphone and delivered the line that the trade press would still be repeating thirty years later. Real men, Sanders said, have fabs.

The phrase had not been his to begin with. A few weeks earlier, a reporter at the San Jose Mercury News named Valerie Rice had been interviewing T. J. Rodgers, the chief executive of Cypress Semiconductor, about why Cypress, unlike many of the chip startups springing up around the Valley, still operated its own manufacturing line. Rice, casting around for a way to summarize Rodgers’ position, asked whether what he was really saying was that real men have fabs, the same way the bestseller of the previous decade had insisted that real men did not eat quiche. Rodgers laughed and said exactly. Sanders read the line, fell in love with it, and decided to make it his own. By the time he uttered it in Phoenix it was no longer a joke. It was a creed, delivered to a room that included a growing number of fabless chip-company executives who had built their businesses on the opposite proposition. Several of them, the trade-press accounts later noted, scratched out parts of their afternoon presentations and rewrote them on hotel notepads.

The line worked because it captured something true about how Sanders saw himself and his industry. In Silicon Valley’s mythology of engineer-founders, Sanders was the great exception. He was not a physicist, not a process expert, not a circuit designer of any distinction. He was a salesman. He had grown up on the South Side of Chicago in the late nineteen-thirties and forties, raised by his paternal grandparents after his parents divorced, and at the age of nineteen he had been beaten so badly by a street gang that a priest was called to the hospital to administer last rites. He recovered, took a Pullman company scholarship to the University of Illinois, finished an electrical-engineering degree in 1958, and drifted through Douglas Aircraft and then Motorola before landing in 1961 at Fairchild Semiconductor, the company that would seed almost every important chip firm in the Valley over the next twenty years. At Fairchild he made his name not in the lab but on the road. He was the man who flew to Detroit to convince automakers that the silicon chip would replace the relay, the man who pitched the integrated circuit to skeptical Pentagon procurement officers, the man who could close a sale on equipment whose physics he could only loosely describe.

In 1968 a rival from Motorola named C. Lester Hogan was lured to Fairchild as president, bringing with him a phalanx of his own people whom the Valley quickly nicknamed Hogan’s Heroes. Hogan’s lieutenants were sober, button-down, and disinclined to forgive Sanders his white suits, his Hollywood lunches, and the occasional Rolls-Royce parked in the company lot. Within months they fired him. He absorbed the dismissal as a moral verdict. My whole life, he told friends, had been about treating people fairly, and now he had not been treated fairly. A few weeks later, eight engineers from Fairchild who had decided to leave together for a startup of their own asked him to come along on a single condition: that he run the new company as president. They had design talent and process knowledge. What they did not have was anyone who could walk into Honeywell or General Electric and persuade a purchasing director to buy a part nobody else had heard of. Sanders said yes. They incorporated Advanced Micro Devices in May 1969, in Sunnyvale, California, with a few hundred thousand dollars of startup capital and a business plan that one trade publication called the least likely to succeed of the year’s many semiconductor startups. Sanders kept that line framed in his office for decades and wielded it, in conversations with new hires, as motivation.

He built AMD the way an opera impresario builds a company. The flamboyance was deliberate. Where Robert Noyce, by then running Intel up the road, drove a battered station wagon and ate at the same Mountain View diner as his engineers, Sanders flew to industry conferences in chartered jets, kept a Bel Air mansion as well as a Saratoga estate, and arrived at the annual AMD sales meetings in costume. One year he held a raffle in which the prize was a thousand dollars a month for twenty-five years, paid out of his own pocket; the winner, a fab-line worker named Jocelyn Lleno, learned of her good fortune when Sanders showed up at her house with a television crew. He instituted a no-layoff policy and held it through downturns that crushed his rivals; in the middle of the 1984 recession, when the board pressed him to make cuts, he told them flatly that he was not going to preside over the dismantling of his life’s work. He gave hundred-dollar bonuses on the company’s first million-dollar quarter and signed personal checks for employees’ children’s college funds. The Wall Street Transcript named him the best chief executive in the semiconductor industry in 1983, then again in 1984, then again in 1985.

Underneath the showmanship was a strategic conviction that hardened into dogma. The chip business, Sanders believed, was at heart a manufacturing business. Design mattered, of course, and AMD spent enormous sums on circuit work. But the leverage, the secret sauce, the thing that separated a real semiconductor company from a glorified consulting shop, was control over the silicon. A company that did not own a fab depended on someone who did. A company that depended on someone else for its bottlenecks would, sooner or later, find that the someone else had become its master. He had absorbed this lesson early, watching how Fairchild’s customers came to depend on Fairchild’s process recipes, and he saw no reason to think anything had changed. In the nineteen-eighties he poured AMD’s profits back into bricks and cleanrooms. The company expanded its Sunnyvale campus, broke ground in Austin, built the so-called Submicron Development Center on the original campus in 1988 to bring next-generation lithography in-house. By the early nineties, when the Phoenix audience heard the famous line, AMD had three operating fabs and was planning a fourth.

The conviction was not unique to Sanders. For a long generation of American chip executives, owning the fab was not a strategic preference; it was the definition of the industry. Texas Instruments owned its fabs. Intel owned its fabs. Motorola, National Semiconductor, IBM, Hewlett-Packard, Hitachi, NEC, Toshiba, Samsung, and every other serious player owned its fabs. The few experiments with separating design from fabrication had been treated as oddities. The model of the integrated device manufacturer, the IDM, was so overwhelmingly the norm that no English shorthand for it existed; you simply said semiconductor company.

What had changed, by the time Sanders climbed onto the stage in Phoenix, was the price tag. A leading-edge fab in 1980 had cost on the order of two hundred million dollars in current money. By 1990 the number had passed five hundred million. By the mid-nineties, when AMD broke ground in Austin on the facility it would call Fab 25 and then began planning a sister plant in Dresden in eastern Germany, the cost of a single state-of-the-art line was approaching two billion dollars; the Dresden plant, dedicated by Sanders himself in October 1999, came in at one-point-nine billion. Each new technology node, the industry’s term for a generation of miniaturization, demanded a wholly new factory, with new lithography tools, new etch chemistries, new metrology, and a new clean-room class. Moore’s Law, the doubling of transistor density every two years that had defined the industry’s economics since the sixties, was producing a parallel curve no one talked about as much: a roughly comparable doubling of fab cost. The deep mathematics of compounding meant that, by some point in the future, the cost of a single new fab would exceed the entire annual revenue of all but a handful of chip companies. People in the room with Sanders in 1992 could already see the trajectory.

That trajectory was producing a counter-philosophy. By the time Sanders was preaching in Phoenix, Morris Chang’s five-year-old Taiwan Semiconductor Manufacturing Company had assembled a small list of clients made up of startups that could not have afforded a fab and specialty designers who had decided that bricks and mortar were no longer the point. The foundry option, once theoretical, now existed.

Inside the Valley, Chang’s model was finding articulate evangelists. T. J. Rodgers, who had originated the quiche line and now watched Sanders weaponize it, was perhaps the noisiest. Rodgers had worked briefly at AMD in his first job out of Stanford and had left convinced that Sanders’ cult of manufacturing was an overcorrection from another era. Cypress, his own company, kept a fab, but Rodgers wrote and spoke incessantly about the new model in which design was the value-creating step and fabrication was a commodity service. Other executives went further. Irwin Jacobs, a communications engineer, had founded Qualcomm in 1985 and decided early that his company would never own a fab; the chips that powered Qualcomm’s wireless designs were built by partners. Jen-Hsun Huang, a young engineer who had cut his teeth at LSI Logic and AMD, would in 1993 cofound a graphics-chip company called Nvidia on the same premise. There was no fab in Nvidia’s plans, and there never would be. The list of design-only companies was still short, but it was growing, and at every fabless-vs-IDM panel of the early nineties the room was a few seats fuller on the fabless side than it had been the year before.

Sanders watched all of this with the disdain of a heretic-hunter. The fabless companies, he told reporters, were not real chip companies; they were design houses dressed up in chip-company costume. A serious enterprise controlled its own destiny, and destiny in this business meant atoms, not just electrons. He compared Rodgers, in one widely circulated remark, to Joseph Goebbels: the master of the half-truth. Rodgers, never one to back down, called the IDMs dinosaurs and accused them of whining for political protection rather than competing. The two men disliked each other personally and disagreed strategically, and the trade press got years of copy out of the feud.

What Sanders could not see, or would not, was that the economics were turning against him. AMD spent the nineties in an exhausting two-front war: against Intel, whose x86 monopoly AMD was trying to crack with a series of in-house microprocessor designs, and against the rising cost of every successive fab generation. The K5, AMD’s first internally designed Pentium-class processor, was late and underpowered; in 1996 Sanders bit down hard and acquired NexGen, a Silicon Valley startup whose Atiq Raza had been working on a competing design, paying $857 million in stock and folding the NexGen team into AMD as the engineering core for what became the K6. The acquisition saved AMD’s processor line and revealed, in passing, that even the most committed IDM in the Valley was perfectly willing to outsource its design intellectual property when its own teams could not deliver. The line that real men have fabs, on closer inspection, said nothing at all about whether real men had to do their own design.

The K6 sold well. The Athlon, its successor, came out in 1999 and for the first time in AMD’s history outran Intel’s flagship part on benchmarks that the trade press considered serious. Sanders dedicated Fab 30 in Dresden that October. Standing on the cleanroom floor in a tailored suit, surrounded by white-coated technicians and German politicians, he called the plant the first fab of the new millennium and the most advanced semiconductor manufacturing facility in the world. He meant it, and for a brief moment he was right. AMD’s stock climbed; the Athlon shipped at a gigahertz; the company crossed five billion dollars in revenue.

It was the apex. The company never quite scaled the same height again with Sanders in charge. By 2002 he was sixty-five and the board was uneasy about succession. He stepped down as CEO that April and handed the reins to Hector Ruiz, a Mexican-born engineer he had recruited from Motorola two years earlier as president and chief operating officer. Ruiz could not have been more different from Sanders if he had been cast for the contrast: the son of a baker in Piedras Negras who had walked across the border every morning as a teenager to attend high school in Eagle Pass, Texas, learned English in three years, taken degrees from UT Austin and a Ph.D. from Rice, and built a methodical, low-temperature career through Texas Instruments and Motorola Semiconductor. He wore quiet suits. He drove ordinary cars. He spoke softly. The Wall Street Transcript was unlikely to make him a CEO of the year three times running. But Ruiz had inherited a company whose annual capital-expenditure line was beginning to dwarf its operating profit, and he was not sentimental about Sanders’ creed.

Sanders kept the chairman’s title for two more years and then, in April 2004, retired from that as well. He left behind a company that owned more silicon real estate than at any prior moment in its history and a board that was beginning, quietly, to ask whether owning all of it had been wise. The Athlon’s success had not been free. AMD had borrowed heavily to build Dresden, then borrowed again to keep the plant on each successive node. Intel, with its vastly larger volumes, could amortize the same fab investment across many more chips. AMD, with a fraction of the volumes, could not. The asymmetry was structural and getting worse.

Ruiz’s response, executed across the second half of the decade, was a series of moves he and his lieutenants began calling the asset-smart strategy. The phrase was deliberately neutral, calibrated to avoid either Sanders’ moral language or his counterparties’. What it meant in practice was a slow, painful inversion of everything Sanders had built: AMD would shed the fabs it could not afford to keep current, raise outside capital to do so, and reposition itself as a design-led company that bought its silicon, increasingly, from outside foundries. In 2003 AMD spun off its memory business into Spansion, the joint venture with Fujitsu. In 2006 Ruiz acquired the graphics-chip designer ATI, a fabless company by definition, which gave AMD a substantial design portfolio that had never seen the inside of an AMD cleanroom. By 2007 the company was in open conversation with sovereign wealth funds about a way to take its remaining fabs off its own balance sheet entirely.

The deal closed in pieces. On October 7, 2008, AMD and an Abu Dhabi state investment vehicle called the Advanced Technology Investment Company announced that they would jointly create a new manufacturing entity, provisionally called The Foundry Company, that would absorb AMD’s Dresden operations and a planned new fab in upstate New York. ATIC would put in $2.1 billion of cash and commit several billion more over the following years; AMD would contribute its fabs, related debt, and operating personnel. Ruiz would step down as AMD’s chairman to become chairman of the spinoff. By the time the transaction closed in March 2009, the new company had been renamed GlobalFoundries, and AMD, the company whose founder had spent three decades insisting that a chip firm without fabs was not a chip firm at all, no longer owned a single one.

The man who signed the final papers on AMD’s side was not Ruiz. Ruiz had resigned as chief executive in July 2008, after the seventh consecutive quarterly loss; the actual stewardship of the spinoff fell to Dirk Meyer, the engineer who had designed the Athlon and now ran the company. Meyer’s announcement on the day the deal closed was carefully bland. Today, he said, was a landmark day for AMD, creating a financially stronger company with a tightened focus. He did not invoke Sanders. He did not need to. Every reporter in the room had the old line ready, and the next morning’s coverage in the trade press wrote itself: real men, the headlines all noted, evidently no longer needed fabs.

Sanders, by then in his seventies and largely retired, declined most interview requests about the spinoff. When he did speak about it, he kept his counsel. He did not concede that he had been wrong. He observed, mildly, that the industry had changed and that companies needed to adapt. The remark surprised everyone who had expected fireworks. It should not have. The man who had once compared T. J. Rodgers to Goebbels could read a balance sheet as well as anyone in the Valley, and the balance sheet, by 2009, said that the cost of staying in manufacturing had finally outrun the cost of leaving.

The fabless companies he had once dismissed were, by then, the largest growth engines in the industry. Qualcomm was a giant. Nvidia was a giant. Broadcom and Marvell and a hundred smaller designers were giants in their categories. None of them had ever owned a fab. All of them bought their silicon from a single Taiwanese foundry whose existence, twenty years earlier, Sanders’ generation had treated as a curiosity. The line about real men had begun as a joke borrowed from a journalist, hardened into a creed, gotten its own audience laughing in Phoenix in 1992, and survived in the trade-press archives for a generation. The industry it was meant to describe had, in the meantime, walked out from under it.