Part IThe Story
They put him in the trunk because there was nowhere else to put him. A neighbor did it, on Chicago's South Side, in the middle of the 1950s. The boy in the back of the car was eighteen and unconscious, and he had been beaten nearly to death at a party. He had stepped in when a group of young men went after a friend of his. The friend ran. The group turned on him. The neighbor closed the lid and drove to Little Company of Mary Hospital, where a priest gave Walter Jeremiah Sanders III the last rites. By one account he lay in a coma for three days.
He lived. More than half a century later, an interviewer from IndustryWeek was inducting him into its Manufacturing Hall of Fame and asked what the beating had taught him. Jerry Sanders answered quickly. "I learned two things. One, don't count on other people — not my favorite part of learning. And the second thing I learned was loyalty, because a neighbor threw me in the trunk of his car and got me to a hospital, so I survived."
The two lessons contradict each other, and he set them side by side without trying to reconcile them. Don't count on other people. Be the person other people can count on. The friend who ran and the neighbor who drove. Over thirty-three years as chief executive of Advanced Micro Devices, the company he co-founded in 1969 and ran until 2002, Sanders lived inside that contradiction. He was let down by an employer, by a partner, by a customer, and by his own shareholders. Each time he answered with a larger promise to the people who worked for him. Some of those promises were kept and some weren't. Silicon Valley mythology prefers its founders as engineers, ascetics, or prophets. Sanders was none of these. He was a salesman. He wore white suits and a rose in his lapel, drove a Bentley convertible, and was mocked in print more often than almost any executive of his generation. He also built the only Silicon Valley chipmaker other than Intel to reach its fiftieth anniversary as an independent company.
He began, as he would end, by standing at a door.
By the Numbers
The Least Likely to Succeed
8Fairchild alumni who incorporated AMD on May 1, 1969
$100,000Capital with which the company was incorporated
53 / 0Employees vs. dollars of sales at the end of AMD's first year, May 1970
$7.87MRaised in the September 1972 IPO: 525,000 shares at $15
1,000,000Am386 units sold between March and October 1991
33 yearsSanders's tenure as AMD's chief executive, 1969–2002
~$1BAMD's annual sales by 1988, the year Sanders predicted an industry-wide downturn
The Hollywood Office
In the creation story of Silicon Valley, the men who built the integrated circuit were physicists and process engineers. Robert Noyce, a minister's son from Iowa, co-invented the silicon chip at Fairchild Semiconductor in 1959 and had the calm of a man who expected the future to arrive on schedule. Gordon Moore, a chemist from the Pacific coast, made his name on a single forecast about the doubling of transistors, a forecast so dependable that it became a law. Charlie Sporck was the manufacturing man, a former foreman who later ran National Semiconductor as if it were a dock. Then there was Andy Grove, refugee and engineer, whom Michael S. Malone, the Valley's most persistent historian, listed alongside Noyce and Moore as "technologists."
Malone described Sanders, the youngest and most outlandish of these "Fairchildren," as "merely a salesman." He meant it as a compliment, though only partly.
Sanders had studied electrical engineering at the University of Illinois at Urbana-Champaign, on a scholarship that one account credits to the Pullman railroad car company, and graduated in 1958. He worked briefly at Douglas Aircraft and then at Motorola, which matters later. In 1961 he joined Fairchild Semiconductor as a sales engineer. At Fairchild he became what the semiconductor business had not yet learned it needed: a man who could explain to a customer, with charm and some stagecraft, why this sliver of silicon deserved a place in that customer's product. Malone says he worked as Fairchild's Hollywood-area salesman and "rarely even saw the home office." He lived in a house in the Hollywood hills that he could not afford. He drove a Bentley convertible. One legend, which Malone repeats while marking it as legend, has him wearing pink pants on a sales call to IBM.
None of this is how an engineer is supposed to behave, and it worked anyway. Sanders moved up through a series of marketing posts until he was director of worldwide marketing and, by most accounts, a likely future vice president and possibly CEO. He was also, as one company history puts it, someone who had worked for Robert Noyce. He had learned the business from the inventor himself and then sold it in a way the inventor never would have.
In 1968 Sherman Fairchild, the patrician investor whose name was on the door, decided the company needed adult supervision. He hired C. Lester Hogan, vice president of Motorola's semiconductor business, which happened to be one of the companies Sanders had already left. Hogan arrived with a group of conservative Motorola managers that the Valley quickly named "Hogan's Heroes." They wore their ties properly and kept their spreadsheets in order. They did not care for a marketing director who dressed like a lounge act and talked like a Chicago alderman.
They fired him.
— Jerry Sanders, on being fired from Fairchild SemiconductorMy whole life has been about treating people fairly, and I wasn't treated fairly.
The chronology made the firing worse. Noyce and Moore had walked out of Fairchild in July 1968, on their own terms, to found Intel. Sanders was pushed out the same year. The men he had sold for left with their reputations and went to a company that would become one of the most valuable institutions on earth. He left with a severance and a grievance. In time the grievance became his business strategy.
The first lesson of the trunk had come back. Don't count on other people.
Five Million Minutes
What he nearly did instead is not widely known. Charlie Sporck, who was close to Sanders at the time, recounts in his memoir SpinOff that after Fairchild, Sanders considered leaving semiconductors entirely. He thought about selling cars. He thought about becoming a travel agent. He thought about becoming a personal manager in the entertainment business, which, given the Hollywood office, the Bentley, and the pink pants, may not have been much of a career change.
Two phone calls kept him in the industry.
The first came from John Carey, a Fairchild engineer who would later run Integrated Device Technology and chair Wolfson Microelectronics. Carey was the operations man, and the kind of person Sanders was not: careful, interior, comfortable on a factory floor. The second came from Jack Gifford, another Fairchild man, who would later be the founding CEO of Maxim. Gifford had the practical sense of someone who builds things to sell, and he would soon persuade Sanders to start with linear chips because they made money right away. Both men told Sanders the same thing. Start a semiconductor company.
A group of Fairchild engineers was already putting a startup together. They asked Sanders to join. He had one condition: he would be president.
The group was divided over it. Engineers are rarely comfortable being led by a salesman, especially a flamboyant one who had just been fired. But they agreed, and that choice determined the character of everything AMD would become. On May 1, 1969, Advanced Micro Devices was incorporated by eight men: Sanders, Ed Turney, John Carey, Sven Simonsen, Jack Gifford, and three members of Gifford's team, Frank Botte, Jim Giles, and Larry Stenger. The company had $100,000. Its first headquarters was Carey's living room. Its second was two rooms at the back of a rug-cutting company in Santa Clara.
Carey summarized the division of labor in six words: "I was Mr Inside at AMD, and Sanders was Mr Outside."
Being Mr. Outside meant raising money, and there the contrast with Intel was painful. Noyce and his partners had needed days, by Malone's count, to fund their company. Arthur Rock wrote a short business plan, the investors signed, and Intel had its money before most of the Valley knew it existed. Sanders spent months sleeping on a friend's couch and walking the streets for investors during the day. Malone says it took a year. In 1985, with AMD by then on the Fortune 500, Sanders still couldn't tell the story without bitterness:
— W. J. Sanders III, 1985Bob Noyce said that it took him five minutes to raise $5 million — well, it took me five million minutes to raise five dollars... But I was dogged about it. I knew I had a story. I knew we could make money.
I knew I had a story. The rest of his career follows from that sentence. Engineers think the product is the story. Sanders thought the story was the product, or at least that it came first, and that it was what convinced investors, customers, and employees to bet on a company with no sales, no proprietary technology, and a president who had just been fired for being himself.
The trade press placed its bet against him. Innovation magazine called AMD "the least likely to succeed of the technology startups of the 68–69 timeframe." Sanders quoted that line for the rest of his life the way some men quote their wedding vows. "Actually, it just fired me up," he told IndustryWeek. "I just wanted to prove them wrong."
In September 1969 the company left the rug-cutting company for its first permanent home: a $500,000, 15,000-square-foot building at 901 Thompson Place in Sunnyvale. It held everything, from the executive offices to an eight-mask, 7-micron production line running two-inch wafers. That same year the founders broke ground on a building on DeGuigne Drive in Sunnyvale. AMD would keep its headquarters in Sunnyvale for forty-seven years.
Plug It In
At the start, AMD sold nothing it had invented. It began as a second source, which was a respectable trade in the early chip industry, though not a glamorous one. Large customers did not want to depend on a single supplier for a critical part, so they required that someone else also be licensed, or at least able, to make it. AMD made chips that Fairchild and National Semiconductor had designed, including Fairchild's 9300 TTL logic family, and redesigned them where it could to run faster and more efficiently. Its first product, the Am9300, a 4-bit shift register, came off the line in November 1969 and went on sale in 1970.
On the face of it, a copy is the hardest thing in the world to sell, because the customer already has the original. Sanders solved this with two moves, both marketing ideas and neither an engineering idea.
The first concerned names. In Sanders's telling to Sporck, every other chipmaker numbered its parts in its own system. Fairchild had the µA741 and National had the LM101. Sanders simply put his own prefix on the competitor's number:
"The really brilliant thing I did was everybody else was using their own numbering system. I just said: 'No, we'll make an AM-741.' Fairchild had a UA-741, National had an LM-101. I just said: 'We'll have an AM-101, and an AM-741.' And I got letters from every lawyer in town. But you couldn't copyright the numbering system."
The point was that a distributor holding an AM-741 did not need to explain anything. The customer already knew what a 741 was. In a single stroke Sanders had taken a competitor's design, its engineering reputation, and its catalog position, at no cost and legally. The second move was a guarantee: "My deal was, 'Plug it in. If it doesn't work, send it back.'"
That phrase captures most of what Sanders understood about selling. It moved the risk off the customer and onto AMD, and it treated the product's unproven reliability as AMD's problem. He then built a company that could keep that promise. At a time when unreliable chips were a familiar headache for computer makers, telecommunications companies, and instrument builders, AMD manufactured its commercial parts to United States Military Standard MIL-STD-883, the testing regime required for aerospace hardware. An April 1970 advertisement read: "Commercial-priced ICs are 100% MIL-tested." The parts went through temperature cycling, baking, hermetic seal testing, and centrifuging, and sold at commercial prices. The sales force reduced it to five words: MIL-STD-883 for free.
A 9300 shift register cost $6.05 commercial and $9.75 military. AMD was giving away the gap between those numbers, in reliability, to win customers' trust. A startup with no history could not ask anyone to trust it. It could only offer to absorb the cost if things went wrong.
The rest of the early strategy shows similar discipline under the flash. AMD stayed out of calculators and digital watches, the consumer boom of the early 1970s, because it judged those markets to be short-lived. It went after computer, telecommunications, and instrument makers, customers who grew steadily and would pay for reliability. Sanders split the company into product profit centers, so that engineers saw the commercial consequences of their design choices in a way their counterparts at larger firms did not. This was Mr. Outside rearranging the inside.
The numbers came slowly and then quickly. When AMD closed its first fiscal year in May 1970, it had 53 employees, 18 products, and no sales. In 1970 it shipped its first proprietary part, the Am2501 logic counter, which sold well. In 1971 its bestseller was the Am2505, then the fastest multiplier available, and it entered memory with the Am3101, a 64-bit bipolar RAM, a common part that was frequently in short supply. AMD's advertising made a point of that. Sales for 1971 were $4.6 million. In September 1972 AMD went public, selling 525,000 shares at $15 each for $7.87 million.
The employee posing for the photograph on AMD's early stock certificates was not a hired model. He was Rich Previte, AMD's twelfth employee, who stayed thirty years and became chief financial officer and then president.
By 1975 the company had 212 products, 49 of them proprietary. That year it entered the microprocessor business the way it had entered everything else, by copying. The Am9080 was a reverse-engineered clone of Intel's 8080. AMD also built its own Am2900 bit-slice family, which found buyers through extensive education and applications support and became the company's best-known proprietary line. For a period, by the Computer History Museum's account, AMD was selling more units of its improved 8080 than Intel was selling of the original.
When Intel began putting microcode into its processors in 1976, it signed a cross-license with AMD rather than fight it. The license gave AMD the right to the microcode in Intel's microprocessors and peripherals, effective October 1976. That ten-word grant would shape the next two decades.
The Asparagus and the Gorilla
Early on, AMD hired a small Los Angeles agency, Keye/Donna/Pearlstein, which later became one of the more inventive shops on the West Coast. For an engineering company, this was an unusual choice. Chip advertising in the early 1970s mostly consisted of spec sheets laid out on a page. AMD's ads made jokes.
One 1979 ad featured asparagus. Its argument was that AMD was investing in higher-margin products that would take years to deliver, the way an asparagus bed takes years to produce its first harvest. The ad asked customers, and probably Wall Street too, to be patient with a company that had never asked anyone for patience. The farming metaphor was an odd fit for a Chicago street fighter in a Bentley, and the oddness was part of the effect.
In 1983 AMD parodied Saul Steinberg's famous New Yorker cover, "View of the World from 9th Avenue," which shows Manhattan in the foreground and the rest of the country shrinking toward a sliver of Pacific. AMD redrew the map from its own position. Many of its ads made fun of competitors. When AMD could not get Intel's next-generation 8086 architecture, it signed in 1978 to second-source Zilog's Z8000 and advertised that choice as if it had been the plan from the beginning.
The Computer History Museum's curators, looking back, saw in all of this "charm, swagger and wit" belonging to a "bigger-than-life Jerry Sanders." The authorized history, Jeffrey Rodengen's The Spirit of AMD: The Legend of Advanced Micro Devices, records the period in company-anniversary style, but even a commissioned history cannot hide how unusual the tone was. Intel's culture was so serious that it later issued an annual memo reminding employees that Christmas Eve did not justify leaving before five. AMD's culture was performed: its founder was the main act, its ads were a running gag, and the joke was always that the small company was smarter than the large one.
The swagger was a bluff, and the bluff was strategic. Malone, who admired Sanders more than he usually admitted, put it plainly in 1999: "It was Sanders who kept AMD looking successful even when it was hurting, looking clever even when it was plodding, and most of all, kept it alive when the rest of the chip industry also-rans began to fold." He called Sanders "a marketer as great as Steve Jobs, but without the erratic behavior."
There was money behind the bluff as well. Siemens, the German engineering conglomerate, wanted American technology and an American foothold, and in 1977 it bought 20 percent of AMD's stock. The two companies started Advanced Micro Computers, with operations in Silicon Valley and Germany, to build microcomputers around the Z8000. The partners soon disagreed about what AMC should be. AMD bought out Siemens's stake in the American division in 1979 and shut the operation down in late 1981. Total sales for fiscal 1978 passed $100 million. In 1979 AMD listed on the New York Stock Exchange and began production at a new fab in Austin. It already had assembly plants in Penang and Manila, and it broke ground in San Antonio in 1981.
A company once rated least likely to succeed now had factories on three continents and was selling its future in ads about vegetables. The one thing it still did not have was the thing it wanted most, which was its own architecture at the center of the computer.
That came from IBM, though not in the form Sanders would have chosen.
People First
On the walls at AMD, and in Sanders's speeches, and in the citation later written for his induction into a semiconductor hall of fame, the motto was always the same: People first, products and profit will follow.
Plenty of companies say something similar. What made AMD's version different was that Sanders spent money on it publicly and theatrically, in the way a Rat Pack figure spends at a casino table. Every employee received stock options, which was a real innovation at the time. One story, repeated across several accounts and difficult to pin to an exact date, has Sanders standing at the door at the end of the first quarter in which AMD earned $1 million in profit and handing every employee a $100 bill on the way out. When AMD reached a sales milestone in 1980, it held a drawing. A production worker, an immigrant in Sunnyvale, won $1,000 a month for twenty years, or $240,000. In 1980 that could change the course of a family.
The point of these gestures was visibility. Sanders had a theory, which he explained to IndustryWeek by referring back to the trunk: "Today there's very little loyalty. There's a lot of jumping around, which is why I made sure that AMD was people-first. Because I knew that was the only way we were going to keep great employees from just taking a job up the street, because there are so many opportunities in Silicon Valley."
The phrase up the street is the key. In the 1970s a Silicon Valley engineer could change employers by crossing a parking lot, and many did, taking with them whatever was in their heads. Intel kept people with prestige: it was the place where the future was being invented. AMD could not offer that, so it offered belonging, along with something like a promise of employment for life. A hall-of-fame citation later called it a "cradle-to-grave" strategy, saying it "engendered loyalty by not laying off employees in downturns like others had," and that the resulting productivity "allowed AMD to match companies many times their size with a fraction of the financial resources."
For a while the record supported this. In 1984 AMD appeared in the book The 100 Best Companies to Work for in America. In 1985 it joined the Fortune 500 for the first time. Then the industry collapsed.
From mid-1985 into 1986, the American chip industry went through what was then its worst decline. Japanese manufacturers were selling memory chips below cost, and American companies were too crowded and too slow to respond. Two choices define how AMD came through it. The first was the Liberty Chip program, an engineering commitment presented as a marketing campaign: AMD would design and produce one new chip or chipset every week for the 52 weeks of fiscal 1986. Instead of cutting through the downturn, the company would produce its way through it. The second was lobbying. AMD pressed the U.S. government hard, and Sanders became one of the industry's loudest voices, until sanctions and restrictions were imposed on Japanese pricing. AMD also pulled out of the DRAM market, which no amount of salesmanship could rescue at those prices, and belatedly moved into CMOS after years of concentrating on bipolar chips.
Sanders came out of the crisis as an advocate. He co-founded the Semiconductor Industry Association and the Semiconductor Research Corporation, and asked late in life what his legacy was, those were the things he named first. "By having an association that can promote the wellbeing of this highly innovative industry — that's what I'm proudest of." The salesman had become a spokesman for his whole industry.
It is at this point that the myth and the record part ways, and an honest account has to follow both.
On December 5, 1988, the Los Angeles Times ran a profile headlined "High-Tech's Naysayer." It described Sanders, then 52, as "dapper," "fast-talking," and "given to wearing all-white suits that match his long, wavy hair." He was predicting the next downturn before anyone else would. "It smells like 1984," he said. "No one wants to believe this, but read my lips: Everyone is going to have layoffs."
That everyone included AMD. The same article reported that Sanders had already announced plans to lay off 2,400 workers worldwide and to close a factory before year's end. The company would still be profitable for 1988, on roughly $1 billion in sales, but it expected a loss in the current quarter.
Wall Street was not impressed. "Sanders shoots from the hip," said Andrew Kessler, an analyst at Paine Webber. "When things are bad for his company, he wants to drag the rest of the industry with him." Christopher Kirby of Rothschild agreed that the industry had a problem, "but it's hitting AMD harder than the entire industry."
Sanders did not back down. He pointed out, correctly, that when he first warned in September he had been alone. Since then Intel had admitted that sales of its 80386 were slowing, National Semiconductor had reported a sharp decline, and Texas Instruments had announced furloughs at its Sherman, Texas, plant. "If I ever questioned my prescience, I no longer do," he said. "I don't mean to sound arrogant, but it's clear to me we're in a decline."
Later retellings say the cradle-to-grave promise was kept, with no layoffs in hard times. The 1988 newspaper says otherwise. Both versions are part of the record, and the tension between them is part of who Sanders was. The promise was sincere and expensive, and it had limits. When those limits arrived, Sanders did not deny them. He announced them first, loudly, before anyone else in the industry would, and then argued with the people who said he was only covering for his own company. The second lesson of the trunk was loyalty. The first lesson came back as well.
The Sword of Damocles
In 1981 IBM was building a personal computer, and it wanted Intel's x86 processor at its core. IBM's purchasing rules, which were close to doctrine there, required at least two sources for any critical part. Intel needed a second source to win the most important design contract of the decade. AMD was the obvious choice.
The two companies signed a ten-year technology exchange agreement in October 1981 and formally executed it in February 1982. The structure was elegant on paper. Each company could earn the right to manufacture the other's products by trading the rights to a product of comparable complexity, with royalties flowing to whichever company had designed the part. The agreement extended the 1976 microcode cross-license through 1995. It included the right to arbitrate disputes. After five years, either side could terminate with a year's notice. One clause was easy to miss: if AMD went bankrupt or was acquired, the cross-license would effectively be void.
The deal made AMD. Starting in 1982 it mass-produced Intel-licensed 8086, 8088, 80186, and 80188 processors, and by 1984 it was shipping the Am286, its own version of Intel's 80286, into the fast-growing market for IBM PCs and their clones. Intel supplied database tapes for the 8086, 80186, and 80286. In 1984 AMD also produced the world's first 512K EPROM. For a few years Mr. Outside had what he had always wanted: a place inside the industry's most important architecture.
In 1984 Intel decided, internally, to stop sharing product information with AMD in order to protect its market position. The court records describe what followed in neutral language. Intel delayed transferring the technical details of its 80386, and then refused. The 386 was the chip that would carry the PC into its next era, and Intel kept it to itself. In 1987 AMD invoked the arbitration clause. Intel responded by canceling the 1982 agreement entirely. In 1990 Intel sued AMD for copyright infringement over its use of the 287 microcode.
Once again Sanders had been let down by people he had counted on. This time it was a partner, and a partner run by the same men who had left Fairchild on good terms while he left on bad ones.
The litigation lasted seven years. The arbitration alone took three years of testimony. During that time AMD could not tell whether it would be allowed to keep making the products that paid its bills. It hedged by building clean-room versions of Intel's code for its 386 and 486 processors, deliberately rebuilding the work from scratch so that no one could argue it had been copied. This was slow, and in the case of the 386 it came long after Intel's own chip, which had been released in 1985. AMD's Am386 finally shipped in March 1991. By October it had sold one million units.
The arbitrator ruled for AMD in 1992. Intel challenged the ruling. In 1994 the Supreme Court of California upheld the arbitrator and awarded AMD damages for breach of contract. That same year a jury found for AMD in the microcode case, affirming its right to use Intel's microcode through the 486 generation. In 1993 AMD released the Am486, which sold well enough that Compaq signed an exclusive agreement to use it. The Am5x86 followed in November 1995. A settlement effective in 1996 confirmed AMD's rights to the 386 and 486 microcode and to nothing after it. Going forward, AMD would have to design its own x86 processors.
When Malone asked him, decades later, whether he had ever paused to enjoy his successes, Sanders's answer described this period:
— Jerry Sanders to Michael S. Malone, Forbes ASAP, May 2000No, actually, I missed that opportunity. I had the chance. We'd been engaged in legal battles with Intel from about '86 to '94... In a three-year period, AMD made over a billion dollars in operating profit. And I was very proud of that billion dollars. But always over my head was the sword of Damocles, that if we lost the court case, AMD was history. So there was tenseness, a tautness in my stomach through all those good years.
In this account a man who earned a billion dollars could not enjoy any of it, because the whole time he was waiting to learn whether he would be allowed to keep it. The flamboyance, the white suits, and the Rolls-Royce look different after reading that passage. Some of it was the performance of a man spending in large, obvious ways because he did not believe the money was secure. Some of it may have been a way of convincing himself.
He won. The bankruptcy-or-takeover clause, which would have voided the license, never came into play. But the win produced a harder problem: AMD now had to stop copying and start inventing.
Kryptonite
AMD's first x86 processor designed entirely in-house, launched in 1996, was named K5. The K stood for Kryptonite, the one substance that could hurt Superman. Superman was Intel. The 5 referred to the fifth generation of x86. Intel had called its fifth-generation chip the Pentium, partly because the U.S. Patent and Trademark Office had ruled that plain numbers could not be trademarked. This was the same rule that had let Sanders call his chip the AM-741 in 1969. A quarter-century on, the rule that had once favored AMD now worked for Intel, which had turned a number into a brand that AMD could not copy.
The name Kryptonite admitted something. A company that names its product after the one weakness of an otherwise invulnerable hero is telling you how it sees the contest. Sanders never claimed AMD was the hero. He claimed it was the rock.
The K5 was not enough on its own. In January 1996 AMD bought NexGen, a small company with its own x86-compatible design, the Nx686, for $857 million in stock. What AMD did next was out of character for a company that ran on its founder's personality. It gave the NexGen team its own building, left them alone, and gave them time and money to rework their chip. The result was the K6, introduced in 1997. Some variants, such as the K6-III/450, were faster than Intel's sixth-generation Pentium II while still using the older Socket 7. The company that had prospered by copying others' work had learned that the most valuable thing it could do with talented people was to give them room and not interfere.
AMD still could not sell to everyone. Michael Dell, who had started his company in a University of Texas dorm room and built it by squeezing every cost out of the supply chain, kept Dell exclusively on Intel. At an industry event in November 1997, Sanders said it outright: "I can't sell a K6 to that guy no matter what I do." For a man who once got distributors to stock chips purely on a promise of refunds, this was a notable admission. There was one customer he could not reach.
The pressure was coming from other directions too. In 1998 the California Public Employees' Retirement System, the largest public pension fund in the country, brought a proxy fight to AMD's annual meeting demanding an independent chairman. CalPERS calculated that $100 invested in a peer group of stocks on December 31, 1992, would have grown to $369 by December 31, 1997. The same $100 invested in AMD would have been worth $98. Proxy filings showed that Sanders had been paid $1.7 million in 1997, including $256,428 for personal security and a $473,812 cost-of-living increase, in a year AMD lost $21.1 million. The year before, he had received $1.3 million while AMD lost $69 million. His options had been repriced after a stock decline. Analysts noticed that AMD shares rose whenever there were rumors he might step down.
Harvard Business School had already turned him into an eighteen-page case study for its organizational behavior curriculum, published January 20, 1998. Professors teaching it presumably did not have to search hard for discussion questions.
The best defense of Sanders that year came from a CNET columnist, Michael Kanellos, and it was only partly a defense. Kanellos called him "a one-man Rat Pack," described the designer ties, bracelets, and rose in his lapel, and noted that in a town where "self-inflicted torment is the ultimate status symbol," where newly rich option holders "still go to Supercuts," Sanders was "one of the few reminders left that money should be spent in large, obvious ways." The column ended: "Sanders may not be an answer to all our ills, but he is all we have."
He survived the proxy fight. The next year was worse. In mid-1999 AMD posted a quarterly loss of $200 million, and Malone wrote that the company had "been at death's door a dozen times — as it is right now." Malone also wrote that AMD, more than any other company, explained the Valley: "No other chapter in the Valley story exhibits as many triumphs and tragedies, outrageousness and grinding hard work."
The product that would rescue AMD had already shipped. On June 23, 1999, AMD released its seventh-generation processor under the name Athlon. For the first time, it did not fit Intel's motherboards. Intel's Slot 1 connector was licensed, so AMD designed its own Slot A, based on the bus from Digital Equipment's Alpha processor. This was a statement of independence built into the hardware. In March 2000 AMD introduced a 1-gigahertz Athlon. Malone called it Sanders's "latest torpedo aimed at Intel" and "his greatest play," and wrote that "once again, he is betting it all."
That spring Malone sat down with Sanders, then 63, for a PBS series on entrepreneurs. Malone described him as "tired but still razor-sharp" and called him "the last man standing." Of all the Fairchildren, he was the only one still running a major chip company. Malone suggested he was stuck in "entrepreneurial hell." Sanders first pushed back by citing the company's size, $2.5 to $3 billion, and then partly agreed: "There's a certain kind of entrepreneurial quality about AMD. I hope there always will be."
He also described how he planned to fight a company much richer than his own. In May 2000 he called AMD "a virtual gorilla," meaning it would combine its strength with partners instead of trying to match Intel's spending on factories alone. AMD worked with Motorola, and later IBM, on manufacturing technology. With Alpha Processor Inc. it developed HyperTransport, a high-speed interconnect, and then handed it to an industry standards body. A company that could not afford to be a gorilla would assemble one from partners.
Around that time, ZDNet UK's Richard Barry asked whether he could promise that AMD's coming 64-bit processors would be as fast and as well supported as Intel's and would ship in volume. Sanders answered with something close to a fatalist's joke:
"Richard, I can't say, categorically, that I'm going to make it through the day! I can't categorically say that we are going to vanquish the largest company in the western world."
Then he made his case. Intel's 64-bit chip, Itanium, required new software, and he said customers did not want new software. They wanted to run what they already had, only on larger data sets. AMD's answer, code-named Sledgehammer, would extend the existing x86 instruction set to 64 bits rather than replace it. When Barry asked whether Intel had made a mistake, Sanders could not resist: "I would never say that Intel made a real cock-up, Richard. Intel spent the money on advertising and turned the cock-up into Celeron!"
He was right about Itanium. He would not be CEO when that was proved.
Real Men Have Fabs
Sanders stepped down as chief executive in 2002 after thirty-three years. His successor, Hector Ruiz, was an engineer who had grown up near the Mexican border and come to AMD from Motorola. He would later write a book titled Slingshot: AMD's Fight to Free an Industry from the Ruthless Grip of Intel. The title shows that the David-and-Goliath framing outlasted its author.
Sledgehammer shipped on April 22, 2003, as the Opteron server processor. It added a 64-bit extension to x86, which became known as x86-64, AMD64, or x64, along with an on-chip memory controller and HyperTransport. Shortly afterward the same technology went into the Athlon 64 for desktops. On April 21, 2005, AMD released the first dual-core Opteron. A month later came the Athlon 64 X2, the first family of dual-core desktop processors. For a few years the company once rated least likely to succeed led the market in performance on the measures that mattered most.
On June 27, 2005, AMD won an antitrust case against Intel in Japan, and the same day it filed a broad antitrust complaint in federal court in Delaware, alleging that Intel had used secret rebates, special discounts, and threats to keep AMD processors out of the market. On May 18, 2006, Dell announced it would ship servers with Opteron chips by the end of the year, ending its exclusive relationship with Intel. By September, Dell was offering Athlon X2 desktops. The customer Sanders could not win while he ran the company was won by the product he had bet on, after he had left.
In November 2009 Intel agreed to pay AMD $1.25 billion and to renew a five-year patent cross-license, settling all outstanding disputes. The fight that began with a 1984 decision inside Intel ended twenty-five years later with a check.
There is a phrase linked to Sanders, repeated so often that it has become as much legend as quotation: real men have fabs. It summed up his view of the business. A chip company that did not own its manufacturing was not really a chip company. He had built fabs in Sunnyvale, Austin, and San Antonio, and he regarded owning them as both pride and survival. The virtual gorilla could share technology with partners but not its factories.
The industry did not accept that view. In October 2008, as each new process generation became more expensive, AMD announced it would spin off its manufacturing into a joint venture with the Advanced Technology Investment Company, an investment arm of the government of Abu Dhabi. Ruiz stepped down as CEO in July 2008 to reassure the new investors, and he became chairman of the new company, GlobalFoundries, in March 2009. The Emirate bought out AMD's remaining stake that year. AMD became fabless, a company that designs chips for others to manufacture. In 2018, after GlobalFoundries stopped developing its 7-nanometer process, AMD moved its leading-edge CPUs and GPUs to TSMC in Taiwan.
The company nearly died anyway. In 2009 it cut 1,100 jobs. In November 2011 it announced it would cut more than 10 percent of its workforce, about 1,400 people. In October 2012 it announced another 15 percent. Its Bulldozer architecture, released October 12, 2011, was in some benchmarks slower than the chips it replaced. AMD was saved by a business Sanders could hardly have pictured: custom chips for game consoles. The PlayStation 4 and Xbox One, both powered by AMD silicon, were later credited with keeping the company out of bankruptcy.
In October 2014 Lisa Su, an electrical engineer who had been chief operating officer since June, became CEO. In 2017 AMD launched Ryzen, based on the Zen architecture that a team under Jim Keller had built from the ground up. AMD set itself a target of a 40 percent improvement in instructions per clock and reported 52 percent. By 2019 Ryzen was reportedly outselling Intel's consumer desktop processors. In February 2022 AMD's market capitalization passed Intel's for the first time in the companies' histories. In 2025 AMD reported revenue of $34.6 billion with about 31,000 employees.
Sunnyvale is gone, and the way it went is worth recording precisely. Two of AMD's former fabrication sites there were placed on the federal Superfund list, in 1986 and 1990, after volatile organic compounds were found in the groundwater. When excavated, the acid neutralization tank vaults were found to have holes. After the fabs were spun off and the layoffs had emptied One AMD Place, the company signed a lease in August 2016 for a 220,000-square-foot building in Santa Clara, ending forty-seven years in Sunnyvale. In April 2019 the Sunnyvale City Council approved the Irvine Company's plan to demolish One AMD Place and redevelop the 32-acre site as townhomes and apartments.
AMD's new headquarters at Santa Clara Square faces Intel's headquarters across the Bayshore Freeway and San Tomas Aquino Creek. After fifty years, the two companies are directly across from each other, separated by a freeway and a creek.
Sanders was not there for any of this as an executive. When IndustryWeek called in 2011, the questions went back to the beginning: the party, the beating, the neighbor. He mentioned his cars, a Ferrari, a Rolls-Royce, a Bentley, an Aston Martin, the way other men mention grandchildren. "I'm a very competitive guy," he said, "or at least they tell me I was, I like to think of myself as mellower now." He said what he was proudest of was the industry association, not the company. He said his legacy was innovation.
The story he told most often was simpler than any of that. A neighbor had closed a trunk lid on a boy who might not live and driven him to a hospital, and that was all the neighbor did. Years later, at the end of the first quarter in which a company with no sales, no money, and no reason to exist finally made a million dollars, its president, a salesman who had been told he was least likely to succeed, went to stand at the door.
The employees came out one at a time, and he handed each of them a hundred-dollar bill.
Part IIThe Playbook
Jerry Sanders is often mistaken for a style: the white suits, the Bentley, the jokes, the swagger. The style mattered, but underneath it were methods that can be repeated, developed by an underdog that spent thirty-three years facing an opponent with more money, better engineering, more patience, and a more prestigious founding story. What follows are the operating principles that can be drawn from his record. Some of them worked well. At least one cost him a great deal. All of them come from the evidence above and not from his own mythology.
The principles
- 1.Sell the guarantee, not the specification.
- 2.Borrow the incumbent's vocabulary.
- 3.Fund the story before you fund the product.
- 4.Choose markets by their half-life.
- 5.Pay loyalty in visible currency.
- 6.Be the first to say the bad number.
- 7.Write the contract for the day your partner betrays you.
- 8.Build the fallback while you fight for the original.
- 9.Buy a team, then leave it alone.
- 10.If you can't be the gorilla, assemble one.
- 11.Name the giant and become its weakness.
- 12.Know when the founder's flourish becomes the company's discount.
Principle 1
Sell the guarantee, not the specification
An unknown supplier's biggest liability is that customers have no reason to trust it. AMD in 1969 was all liability: no record, no proprietary parts, no reputation. Sanders did not try to persuade customers that his chips were better. He offered to carry the risk if they were not. "Plug it in. If it doesn't work, send it back." He then built manufacturing that let him keep the promise. Running commercial parts through MIL-STD-883 testing (temperature cycling, baking, seal tests, centrifuging) at commercial prices turned an expensive quality program into a sales message: MIL-STD-883 for free.
What makes this work is the order of operations. Sanders identified what customers were most afraid of, which was unreliable chips, made that the center of the offer, and then reorganized the factory to support it. Most startups do the reverse. They build what the engineers find interesting and then hunt for buyers who care about it.
The gap between AMD's commercial and military prices, a 9300 at $6.05 versus $9.75, was the amount of trust AMD was handing out. It was a customer acquisition cost paid in testing rather than in advertising.
Tactic: Write down the one failure your target customer most fears, and redesign your offer so your company absorbs that risk visibly, in writing, before you discuss features.
Principle 2
Borrow the incumbent's vocabulary
The AM-741 trick, putting AMD's prefix on Fairchild's part number, was a move about distribution disguised as a naming decision. Distributors did not have to explain an unfamiliar product. Buyers did not have to look anything up. AMD inherited the incumbent's catalog position, its documentation, and the customer's existing understanding of the part, all legally, because, as Sanders noted with satisfaction, "you couldn't copyright the numbering system."
The same thinking drove AMD's strategy for the next three decades. The Am9080 copied the 8080. The Am286, Am386, and Am486 were deliberately compatible. Even Sledgehammer, AMD's most important independent bet, extended the x86 instruction set instead of replacing it, because Sanders saw that customers "do not want to go to new software and new operating systems just because they're using a new chip." The challenger's best move is often to make switching seem like nothing has changed.
The limit to this approach arrived in 1996, when the Pentium name showed that an incumbent could trademark its way past a copier. The approach works until the incumbent stops speaking in shared, uncopyrightable terms and starts speaking in brands.
Tactic: Find the shared, unprotectable vocabulary of your category, such as standard formats, part numbers, or protocols, and make your product read as an obvious drop-in before you ask anyone to learn something new.
Principle 3
Fund the story before you fund the product
"It took me five million minutes to raise five dollars... But I was dogged about it. I knew I had a story." Sanders spent about a year sleeping on a friend's couch and canvassing for money. Intel raised its money in a few days. The contrast is usually told as an injustice. It is better read as practice. Raising money the hard way forced Sanders to keep refining the story until it worked, and the story he ended up with, the scrappy underdog against a monopoly, ran the company for thirty years.
At AMD the story came before the hardware and often stood in for it. The asparagus ad of 1979 asked for patience before the higher-margin products existed. Malone's assessment that Sanders kept AMD "looking clever even when it was plodding" names the cost honestly: the story can get ahead of the product. It also names the benefit. Without the story, there would have been no company left by the time the product arrived.
Tactic: Before raising capital, write the one-paragraph narrative your company will still be telling in ten years, and test it on people who have every reason to say no.
Principle 4
Choose markets by their half-life
In the early 1970s the obvious money was in consumer electronics: calculators and digital watches. AMD stayed out on purpose, judging those to be short-lived markets, and went after computer, telecommunications, and instrument makers, customers who would pay for reliability and keep buying. This is the asparagus logic applied to market selection. The question was not how big a market was this year but how long it would keep producing.
Being disciplined also meant leaving markets. AMD left DRAM in the mid-1980s rather than keep fighting Japanese pricing on a commodity. It shut down the Siemens microcomputer venture in late 1981 once the partners' goals diverged. A salesman's instinct is to say yes to everything. Sanders's portfolio decisions show a willingness to say no.
A
AMD's Bets Under Sanders
Outcomes as recorded in the source material.
| Bet | Years | Outcome |
|---|---|---|
| Second-source logic with MIL-STD testing | 1969–1972 | Scale $4.6M sales by 1971; IPO 1972 |
| Advanced Micro Computers (Siemens JV) | 1977–1981 | Exit Closed after visions diverged |
| DRAM | to mid-1980s | Exit Withdrew amid Japanese dumping |
| Am29000 RISC processor | 1986–1995 | Pivot Survived as an embedded part |
| Clean-room Am386 | 1991 | Scale 1M units in seven months |
| NexGen acquisition → K6 | 1996–1997 | Scale K6-III/450 outran Pentium II |
| Athlon / Slot A | 1999–2000 | Scale 1 GHz by March 2000 |
| Sledgehammer (x86-64) | announced c. 2000 | Legacy Shipped as Opteron, April 2003 |
Tactic: For each market you are considering, estimate how many years of steady purchasing it offers, not how large it is today, and drop the ones whose demand looks like a fad.
Principle 5
Pay loyalty in visible currency
Sanders understood that in a labor market where engineers could change jobs by crossing a parking lot, an abstract promise of loyalty was worth almost nothing. So he made the rewards concrete and public: stock options for every employee, a $100 bill handed over at the door, a production worker winning $1,000 a month for twenty years. These were not the biggest possible rewards. They were the easiest to see and to talk about. Each one became a story employees told their families and recruiters heard about.
The underlying idea, which Sanders spelled out to IndustryWeek, was defensive: people-first was "the only way we were going to keep great employees from just taking a job up the street." Culture functioned as a retention policy. A hall-of-fame citation would later credit this with letting AMD "match companies many times their size with a fraction of the financial resources."
The visibility mattered as much as the amount. A bonus deposited quietly is forgotten. A hundred-dollar bill handed over in person becomes part of company lore.
Tactic: Set aside part of your compensation budget for rewards that are public, physical, and attached to a specific moment, and make sure the founder delivers them in person.
Principle 6
Be the first to say the bad number
The 1988 episode shows the most complicated side of Sanders. He had promised people-first and cradle-to-grave employment, and he announced 2,400 layoffs and a factory closure anyway. He also did not hide it. He went public before any competitor had, warning "read my lips: Everyone is going to have layoffs," and accepted the analysts' criticism that he was "projecting AMD's own distinct problems onto the rest of the industry."
He turned out to be right about the industry. Within weeks Intel, National, and Texas Instruments had all admitted to inventory problems. Being early did not spare AMD the pain. What it did was keep his credibility intact. A leader who announces bad news first keeps the right to be believed later. One who waits until it can no longer be denied loses that right.
The broader point for operators is that a values statement will eventually run into economic reality. The question is not whether that happens but whether you will be the one to say so.
Tactic: When your data shows a downturn coming, announce it, along with your own company's share of the damage, before your competitors do, and tie the announcement to a specific number.
Principle 7
Write the contract for the day your partner betrays you
The 1982 Intel agreement made AMD's fortune and very nearly ended it. Two clauses decided what happened. The arbitration clause gave AMD a forum when Intel stopped cooperating in 1984 and canceled the agreement in 1987. That forum produced the arbitrator's 1992 ruling and the California Supreme Court's affirmation in 1994. The termination clause, allowing either party to end the agreement with a year's notice after five years, gave Intel its way out. A third clause, voiding the cross-license if AMD went bankrupt or was taken over, sat quietly in the document as a potential death sentence.
Sanders's own summary, "the sword of Damocles," along with the billion dollars of operating profit he could not enjoy, shows what it costs to depend on a partner's goodwill. The arbitration clause was what kept AMD alive. Without it, Intel's 1984 internal decision would have ended AMD's x86 business with no recourse.
⚖
Twenty-Five Years With Intel
From reverse engineering to settlement.
1975
AMD ships the Am9080, a reverse-engineered Intel 8080.
1976
Intel cross-licenses its microcode to AMD, effective October.
1981–82
Ten-year technology exchange signed (Oct.) and executed (Feb.) so IBM has a second source.
1984
Intel decides internally to stop sharing product information; withholds the 386.
1987
AMD invokes arbitration; Intel cancels the 1982 agreement.
1990
Intel sues over AMD's use of 287 microcode.
1991
Clean-room Am386 ships in March; one million sold by October.
1992
Arbitrator rules for AMD.
1994
California Supreme Court upholds the award; jury finds for AMD on microcode.
2005
AMD files antitrust complaint in Delaware; wins in Japan the same day.
2009
Intel pays AMD $1.25 billion; all disputes settled.
Tactic: For any partnership your company depends on, ask a lawyer to draft the termination and dispute clauses on the assumption that the partner will act in bad faith in year five, and negotiate those clauses first.
Principle 8
Build the fallback while you fight for the original
AMD did not wait for the courts. While litigating for its rights to Intel's designs, it built clean-room versions of the 386 and 486, reconstructing the work independently so that whatever the verdict, it would have a product to ship. The Am386 came late, years after Intel's 1985 chip, but it sold one million units in seven months. The fallback became the business.
The pattern recurs. When the 1996 settlement ended AMD's rights at the 486 generation, the company already had the K5 in development and NexGen about to be acquired. Litigation and engineering ran side by side, each covering for the other. The lesson is not to rely entirely on winning in court. Use the time a lawsuit gives you to make the lawsuit matter less.
Tactic: Whenever you are in a dispute over rights you depend on, assign a separate team to build the version of your product that does not need those rights.
Principle 9
Buy a team, then leave it alone
AMD's purchase of NexGen for $857 million in January 1996 is the clearest sign that Sanders knew where his talents ended. A company built on its founder's personality did something contrary to that personality: it gave the acquired engineers their own building, gave them time and money, and did not interfere. The K6 resulted, and some versions beat Intel's Pentium II.
The tactical point is that an acquisition's value often comes from leaving the acquired team alone. Integration, the usual post-merger ritual of combining reporting lines and replacing tools, can destroy exactly what made the purchase worth making. Sanders bought a roadmap he could not have built himself, and then trusted it.
Tactic: When you acquire a team for its technology, physically separate it, fund it explicitly, and set one delivery date as the only management check-in for the first year.
Principle 10
If you can't be the gorilla, assemble one
Sanders's "virtual gorilla" idea was an honest description of AMD's position. AMD could not match Intel's capital spending, so it pooled with others: Motorola and IBM on manufacturing technology, IBM specifically on silicon-on-insulator and 90-nanometer know-how, and Alpha Processor Inc. on HyperTransport, which AMD then gave to a standards body so others would adopt it. The approach treats scale as something that can be put together from partners rather than owned outright.
The approach had a limit, and Sanders set it himself: the fabs. Real men have fabs. He was willing to share technology but not factories. His successors went further, out of necessity, and spun manufacturing off into GlobalFoundries in 2008–2009 and later moved to TSMC. The fabless AMD that eventually passed Intel in market value was the virtual gorilla taken further than Sanders ever would have taken it.
J
Conventional Wisdom vs. the Sanders Method
Where he departed from Valley orthodoxy, and where his successors departed from him.
| Conventional wisdom | Sanders's approach |
|---|---|
| Engineers lead chip companies. | A salesman demands the presidency as a condition of joining. |
| Differentiate with proprietary names and designs. | Adopt the incumbent's part numbers; compete on guarantee. |
| Frugality signals seriousness. | Spend "in large, obvious ways" to signal survival. |
| Match the leader's capital spending. | Assemble a "virtual gorilla" of partners. |
| Integrate acquisitions quickly. | Give NexGen its own building and leave it alone. |
| Outsource manufacturing when costs rise. | "Real men have fabs," later reversed by his successors. |
Tactic: List the three capabilities where your largest competitor outspends you most, and for each, name one partner whose own interests line up with closing that gap.
Principle 11
Name the giant and become its weakness
The K5 was named after Kryptonite. The product name was a positioning statement: AMD presented itself as the one thing that could hurt an otherwise invincible rival. Sanders consistently cast AMD as the alternative to a monopoly, from the lobbying against Japanese dumping in the 1980s to the 2005 antitrust complaint that his successors filed. The underdog story gave customers who wanted a second supplier a reason to choose AMD beyond price.
This framing works because it turns the competitor's size into an argument for buying from you. A buyer purchasing from the challenger is also protecting itself against the incumbent's pricing. The costs are real too. Naming yourself after the giant's weakness defines you in relation to the giant. For three decades AMD described itself mainly in terms of Intel, which may help explain why its turnaround eventually came from a market Sanders hadn't foreseen: custom silicon for game consoles.
Tactic: Describe your product as the specific remedy for the incumbent's most resented behavior, then keep a separate roadmap for markets where the incumbent does not matter.
Principle 12
Know when the founder's flourish becomes the company's discount
The CalPERS proxy fight of 1998 was a referendum on Sanders himself. A $100 investment in AMD over five years had become $98, against $369 for peers. Compensation filings included $256,428 for personal security and a $473,812 cost-of-living increase in a year with a $21.1 million loss. The most damning detail was that AMD stock rose on rumors he would leave. When the market prices a founder's departure as good news, the founder has become a cost to the company.
The same flamboyance that had kept AMD alive in 1970 by projecting confidence it hadn't yet earned had turned, by the late 1990s, into a governance problem. Sanders survived the vote, and the Athlon and Sledgehammer that came after show he still had his strategic instincts. But the episode is a warning that applies to any founder whose image and company have become hard to tell apart. The style that wins customers early can cost investor confidence later.
Tactic: Once a year, ask your board directly whether any part of your personal conduct or compensation is reducing the company's valuation, and treat a yes as an operating problem, not a personal insult.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— Jerry Sanders, to IndustryWeek, 2011, on surviving a near-fatal beating at eighteenI learned two things. One, don't count on other people — not my favorite part of learning. And the second thing I learned was loyalty, because a neighbor threw me in the trunk of his car and got me to a hospital, so I survived.
— AMD's founding motto under SandersPeople first, products and profit will follow.
— Jerry Sanders, as recounted in Charlie Sporck's SpinOffMy deal was, 'Plug it in. If it doesn't work, send it back.'
— Jerry Sanders to ZDNet UK, on AMD's 64-bit roadmapRichard, I can't say, categorically, that I'm going to make it through the day! I can't categorically say that we are going to vanquish the largest company in the western world.
— Jerry Sanders, to the Los Angeles Times, December 1988If I ever questioned my prescience, I no longer do. I don't mean to sound arrogant, but it's clear to me we're in a decline.
Maxims
- Absorb the customer's fear. A new company earns trust by taking on the risk the buyer is most afraid of.
- Speak the incumbent's language. The easiest product to sell is one that seems familiar from the start.
- The story is the first product. Capital, customers, and employees all follow a narrative before they follow a spec sheet.
- Pick markets that keep producing. Choose demand that lasts over demand that is merely large this year.
- Make loyalty visible. A reward that people can see and retell is worth more than a larger one deposited quietly.
- Announce the bad news first. Credibility goes to whoever says the hard number before it can no longer be denied.
- Draft for betrayal. In any partnership you depend on, the dispute clause matters more than the revenue clause.
- Fight in court and build anyway. Use the time a lawsuit gives you to make the lawsuit matter less.
- Acquire, then step back. A team bought for its talent needs its own building and minimal interference.
- Watch what the market thinks of you. When the stock rises on rumors of your departure, your style has become a cost to the company.
