Hector Ruiz, CEO of Advanced Micro Devices (AMD), slowly hung up the phone as he reached for a piece of paper in his top-desk drawer. Quietly studying the handwritten word “MAID” on it, Ruiz scratched out the letter “D” for Dell. It was mid 2006, and Ruiz had just talked with Michael Dell, the founder and chairman of Dell Inc (Dell). He had called Ruiz with the news that Dell would purchase Opteron, AMD’s server microprocessor, for its highest-end server line. Dell had long been an Intel-only shop. Landing Dell as a customer was the culmination of a four-year effort that AMD had codenamed Project MAID.

Sunnyvale, California-based AMD designed and manufactured microprocessors for the computing, communications, and consumer electronics markets. With roughly 10,000 employees, the semiconductor company had 2005 revenues of $5.8 billion, a 17% increase over 2004. The 2003 launch of Opteron and the company’s AMD64 technology ushered in a new chapter in AMD’s history. Traditionally, AMD had been a distant follower to Intel, which had a dominant position in microprocessors for the server and personal computer (PC) markets. However, Intel’s dominance was eroding as Opteron gained acceptance and AMD focused on “customer-centric innovation” under Ruiz, who was appointed CEO in 2002.

Driven by Opteron’s success, AMD’s unit share in servers for the second quarter of 2006 rose to
26%, up from 11% in Q2 of 2005.1 The top four computer-makers that sold the vast majority of servers—Hewlett-Packard (HP), Sun Microsystems (Sun), IBM, and Dell—now offered at least one
Opteron-based server. Furthermore, AMD’s presence in the lucrative corporate segment was growing: 90% of the top 100 Forbes Global 2000 were using AMD64 technology by the end of 2005.2
And, AMD reported higher margins than Intel in the first quarter of 2006.

AMD also felt it had built enough credibility to lead the industry in new directions. The company had recently launched a marketing initiative called the “Power Campaign” to focus the industry on the importance of energy efficiency and systems designed to maximize performance at the minimum power consumption. Accordingly, AMD was leading an effort to use “performance-per-watt”—in which it held an advantage—as the best benchmark to compare competing microprocessors.

Yet, Ruiz saw challenges that could pose a threat to sustainable growth for AMD. Intel’s dominant market position could limit AMD’s ability to make inroads into key market segments beyond servers, such as corporate desktops and notebooks. Furthermore, Intel had just announced its “roadmap to recovery” that included a new line of microprocessors that balanced performance, power consumption, and cost. Intel believed its new processors would be at parity, or even superior, to AMD’s on a performance-per-watt basis. Intel also planned a major restructuring, vowing to cut

Professor Elie Ofek and Research Associate Lauren Barley prepared this case. HBS cases are developed solely as the basis for class discussion. Cases are not intended to serve as endorsements, sources of primary data, or illustrations of effective or ineffective management.

Copyright © 2007 President and Fellows of Harvard College. To order copies or request permission to reproduce materials, call 1-800-545-7685, write Harvard Business School Publishing, Boston, MA 02163, or go to http://www.hbsp.harvard.edu. No part of this publication may be reproduced, stored in a retrieval system, used in a spreadsheet, or transmitted in any form or by any means—electronic, mechanical, photocopying, recording, or otherwise—without the permission of Harvard Business School. 507-037 AMD: A Customer-Centric Approach to Innovation

$1 billion in costs and to renew its focus on microprocessors. There was also concern that Intel would cut prices aggressively to regain lost market share and sell its soon-to-be obsolete inventory.

Ruiz reflected on the upbeat call with Michael Dell, but his thoughts returned quickly to the issues his company faced. Would AMD’s success in servers translate into greater acceptance by computer- makers and end users in segments beyond servers? Specifically, could AMD leverage Dell’s initial order to win Dell’s volume business, the lower-end server and commercial client businesses that included corporate desktops and notebooks? How would Intel’s new product line affect AMD’s efforts in the market? Was benchmarking by performance-per-watt relevant? And, in the short-term, how should AMD respond to Intel’s price cuts?

AMD—The Early Years

Jerry Sanders, formerly of Fairchild Semiconductor, founded AMD in 1969. AMD’s focus on microprocessors began in the mid 1970s. A microprocessor—also referred to as a “processor” or a “chip”—was usually a single silicon chip (also known as an integrated circuit) that served as the brain or central processing unit (CPU) of a computer. The microprocessor executed the instructions and processed the data in computer programs. A computer’s underlying architecture, instruction set, and operating system had to be compatible with the microprocessor and affected how well the processor performed its tasks.

A Second Source for Intel

In 1981, IBM transformed the personal computer industry when it launched the IBM PC with the Intel 8088, a 16-bit processor conforming to Intel’s newly developed x86 microprocessor architecture. Hoping to achieve rapid penetration, IBM adopted an open standard to encourage software developers to design applications for the IBM PC. As the market exploded, IBM required Intel to license its patents to other chip suppliers to ensure a reliable supply of microprocessors and to spur price competition and innovation. In 1982, AMD became a licensed second-source for Intel’s x86 family of processors, originally named because the earliest processors—excluding the 8088—had model numbers ending in “86.”

AMD later manufactured Intel’s 80286 (286) in 1986, but relations grew strained as the two companies disagreed over the cross-licensing agreement. AMD pursued arbitration in 1987 to gain access to Intel’s design for its next-generation processor the 80386 (386), a 32-bit processor. In 1992, the arbitrator awarded AMD more than $10 million in compensation and a permanent, royalty-free license to the 386. AMD released the Am386 in 1991 and Am486 in 1993, low-priced clones of Intel’s 386 and
80486 (486) processors. In 1995, the two companies reached an agreement that recognized AMD’s rights to Intel’s microcode—the software code inside the processor—for the 386 and 486. However, AMD agreed it would not use Intel’s microcode beyond the 486 processor. Furthermore, AMD was prohibited from using Intel’s next-generation technology for connecting the CPU to the computer’s memory and other components, and the companies’ product development paths diverged.

The “K” Series

AMD optimistically dubbed its next-generation, in-house x86 processor “K” for Kryptonite, the only substance harmful to Superman. The K5, the first in the series, was released in 1995 to compete with Intel’s Pentium—also an x86 processor—that was released two years earlier. With its late arrival and lackluster performance, the chip failed in the marketplace. To accelerate the release of its next

2 AMD: A Customer-Centric Approach to Innovation 507-037

product, AMD acquired NexGen, a company that also designed x86 processors. This move was credited with saving AMD’s x86 processor business, and the resulting K6 line was a more robust competitor to Intel’s. The K6 fit into existing Pentium motherboards and offered comparable performance, but at a much lower price. It proved a modest success with consumers, and in part prompted Intel to launch its lower-end version of its Pentium called the “Celeron.” By 1998, AMD held 60% of the chip market for PCs priced lower than $1,000 to Intel’s 24% share (16% Celeron, 8% Pentium II). However, AMD’s success was uneven as the company suffered manufacturing yield problems and was not able to meet customer demand at times. Furthermore, AMD was left battling with Intel in the price-sensitive consumer segment and was never able to gain a foothold in the lucrative corporate market with its higher volumes and longer product life cycles.

The “Virtual Gorilla”

Founder and CEO Sanders realized that AMD needed a change to survive, which resulted in his “Virtual Gorilla” strategy that leveraged industry technology partnerships to compete against the much bigger Intel. AMD’s Athlon (also called K7), an x86 processor released in 1999, reflected a much- improved design. Its performance beat Intel’s comparable Pentium processor, which surprised both Intel and the industry. Intel rushed a redesigned processor to market, resulting in low manufacturing yields. In contrast, AMD—with help from its strategic partner Motorola—had greatly improved its manufacturing processes and enjoyed a clear lead over Intel.

AMD’s overall unit share climbed briefly over 20% in 2001 as it capitalized on Intel’s supply shortages and its own advantages in pricing, performance, and manufacturing. AMD’s marketing strategy emphasized comparisons that demonstrated AMD’s superior performance. Yet, AMD chips were still largely found in the low-end of the retail business. Intel countered in 2002 with its Pentium 4 processor, and by 2002, AMD’s unit share was down to less than 15%.3 (See Exhibits 1 and 2 for product lines and market shares.) As the U.S. recession lengthened into 2002, hitting the highly cyclical semiconductor industry particularly hard, AMD announced plans to cut 15% of its workforce and faced an uncertain future. (See Exhibit 3 for AMD selected financial data.)

Intel—The Early Years

Gordon Moore and Robert Noyce, also formerly of Fairchild Semiconductor, founded Intel in 1968.4
In 1970, Intel produced the first 1-kilobit DRAM (dynamic random access memory), which replaced magnetic tapes for storing computer data. By the early 1980s, the DRAM market had matured with cost-effective Japanese companies capturing the bulk of the market, and the memory business was draining Intel’s resources. It was fortuitous that then-president Andy Grove was able to exit the
DRAM business and shift Intel’s focus to microprocessors just as the PC market took off.

x86 Microprocessors

Although Intel engineers created the first microprocessor in 1971, the company was slow to realize its potential in microcomputers, the precursors of personal computers. This changed as Apple launched the Apple II, the first commercially successful microcomputer, in 1977. Intel lost the design bid for the Apple II processor to Motorola, but won the microprocessor business for the IBM PC launched several years later. As its processor became the PC standard, Intel was able to force IBM and other PC computer-makers (also known as original equipment manufacturers or “OEMs”) to drop their

3 507-037 AMD: A Customer-Centric Approach to Innovation

second-source requirement. OEMs assembled standard components and built them into finished computer systems, which they marketed and sold under their own name to end-users.

In the 1990s, Intel began designing and manufacturing other system components that worked with its processors, including logic chipsets and motherboards, to encourage more rapid adoption of its next-generation processors. With its dominant market position, Intel exerted significant influence over the industry standards that affected virtually all the components in the computing platform.

Intel faced many threats in the 1990s not only from clone manufacturers such as AMD, but also from companies that supported alternative proprietary architectures. In 1991, Apple, IBM, and Motorola formed the “AIM Alliance” to produce the PowerPC that was based on the RISC (Reduced Instruction Set Computing) architecture. Intel responded to these competitive challenges by investing even more heavily in R&D. Furthermore, the company moved to an overlapping product development cycle in which the next two generations of chips were developed simultaneously—rather than serially—leading to large advances in performance and even shorter product life cycles. Intel’s aggressive product development strategy combined with its manufacturing prowess allowed the company to ship new chips approximately every other year.

“Intel Inside”5

As Intel’s product innovation accelerated, the company realized the ongoing need to generate demand for its next-generation microprocessor. Research after the launch of the 386 showed that end- users were satisfied with the 286 and unaware of the reasons to upgrade. This led to Intel’s umbrella marketing campaign—”Intel Inside”—that was launched in 1991. An example of ingredient branding, “Intel Inside” focused on persuading end-users of the processor’s importance in the computer purchase decision.

Part of the campaign involved co-op funds that reimbursed up to half the advertising costs of participating OEMs if they purchased Intel chips and used the “Intel Inside” logo in their ads and on their computers. Intel also launched its own advertising campaign focusing on the brand attributes of “reliability” and “leading technology.” In an industry driven by Intel’s short product life cycles, “Intel Inside” prompted IT managers and CIOs to upgrade to the latest processing technology, rendering current Intel and competitive technology obsolete. By 2000, at least 2,700 PC OEMs had participated in the program, and Intel with its co-op partners had spent more than $7 billion on “Intel Inside” advertising, creating one of the most globally recognized and valuable brands.

By 2000, Intel covered the PC and server markets with its family of x86 processors: the 32-bit Celeron for the low-end consumer PC segment; the 32-bit Pentium for the higher-end consumer PC and corporate desktop and mobile segments; and the 32-bit Xeon for the low- and mid-range corporate server and workstation segments. Servers were powerful, networked computer systems that managed tasks such as database queries, web site hosting, and file and email storage. Intel introduced its server microprocessor Xeon in 1998, which started a migration—particularly in the low-end server segment— to x86 servers and away from proprietary RISC-based systems. Behind the hugely successful “Intel Inside” campaign and the company’s dominant share in x86 microprocessors, Intel’s profits soared until the U.S. economy stumbled into recession in 2001. (See Exhibit 4 for Intel selected financial data.)

Itanium

In the early 1990s, Intel had begun development with HP on a new 64-bit processor, dubbed
Itanium. 64-bit chips processed data in chunks twice the size as 32-bit processors, making them

4 AMD: A Customer-Centric Approach to Innovation 507-037

suitable for memory-intensive and complex tasks such as maintaining huge corporate databases or creating industrial designs. Intel expected its Itanium chip would help the company enter the higher- end server market, previously dominated by Sun and IBM with their proprietary 64-bit RISC architectures, SPARC and POWER respectively. (See Exhibit 5 for schematic of server market.)

With an entirely different architecture than an x86 processor, Intel’s Itanium promised significant performance advantages over existing 32- and 64-bit server products. However, to take advantage of Itanium’s performance, the large installed base of x86 32-bit server software would need to be rewritten. Itanium ran x86 applications, but at a very slow speed. Itanium cost Intel $1 billion to develop and took over a decade to complete. It launched in 2001 and was several years late to market.

Remaking AMD

When I came to AMD, it was like watching a lot of six-year olds play soccer. They were motivated, talented, and enthusiastic. But, it was hard to coach them—they just wanted to kick the ball. We needed some discipline. And, we needed a rallying cry to try a different path.
— Hector Ruiz

In the late 1990s, AMD was assessing Intel’s Itanium strategy. Randy Allen, corporate vice president of AMD?s server products division and a 22-year AMD veteran, summarized:

For years, Intel had by far the largest share of the x86 marketplace and AMD was the only viable competitor. From time to time, we hurt Intel—sometimes in price, sometimes in performance. I think this created a desire within Intel to distance themselves from us once and for all when they saw computing going to 64 bits. The Itanium architecture was a lockout play, and nobody else—including AMD—would have the rights to it. Intel knew it alone had the resources to move the entire software infrastructure from x86 to a new instruction set.

Hammer

In 1999, after lengthy debate, AMD decided to respond with an approach to 64-bit computing that was markedly different than Intel’s. AMD set out to develop an x86 64-bit processor, which was codenamed “Hammer.” Allen explained AMD’s motivation:

We heard from many in the industry that they viewed a transition such as the one Intel proposed with Itanium as a major disruption, fraught with peril. Our only viable option to go to
64-bit computing was to extend the current x86 architecture from 32 bits to 64 bits. We weren’t going to get a license for Itanium, and we didn’t have Intel’s power to move the industry to a whole new software infrastructure.

Tom McCoy, executive vice president of AMD’s legal affairs and chief administrative officer, added his thoughts:

We had to get into the server business. By the mid to late 1990s, Intel and Microsoft had marched up the server food chain, where the tier-one OEMs—IBM, Sun, Dell, and HP—made their money. The OEMs needed Intel’s processors, and it was too big a club over AMD’s head. Whenever we had a design win in one segment, the OEMs had to retreat for fear of retaliation from Intel in a different segment. We needed to show that we could design and bring to market a server-class product—a great technology that was something the world really wanted.

5 507-037 AMD: A Customer-Centric Approach to Innovation

Dirk Meyer and Hector Ruiz orchestrated AMD’s response. Sanders hired Meyer in 1996 from
Digital Equipment Corporation (DEC) where Meyer had been co-architect of the Alpha 21064 and
21264 microprocessors, well-regarded 64-bit proprietary RISC processors. In 2000, Sanders persuaded
Ruiz to join AMD as president and chief operating officer. Ruiz came from a 22-year career with
Motorola, where he had risen to head its semiconductor products operations. Ruiz, skilled in building teams and processes, brought a more methodical approach to AMD, particularly useful as the company defined its strategy to penetrate the server market.

AMD’s design team built Hammer from the ground up, and the new processor brought several innovations to the x86 server market:

• The first was AMD64—also known as x86-64 or x64—the company’s new 64-bit extension to the industry-standard x86 instruction set architecture. AMD64 had two primary capabilities. It ran legacy x86 32-bit applications, in many cases more quickly than existing processors. Furthermore, it also ran 64-bit applications as users migrated to 64-bit computing. One industry participant estimated it took only two days to transition existing software to AMD’s new architecture AMD64, but four to six months to transition over to Itanium’s.6

• The second was its “Direct Connect” architecture that greatly improved communication to the processor, thereby increasing performance. With “Direct Connect,” an AMD processor was directly attached to its own memory through an integrated memory controller. In contrast, Intel’s processors routed instructions through a shared chipset (also known as the front-side bus), which could create system–level bottlenecks that impeded performance. In addition, when a system contained more than one AMD processor, all the processors were directly connected to one another with AMD’s “Direct Connect” architecture.

• The third was “HyperTransport,” a technology AMD developed with industry partners including Apple, Broadcom, Cisco Systems, NVIDIA, and Sun. With “HyperTransport,” all the direct connections occurred at much greater speeds using high-speed links.

Project MAID

AMD was well on its way to designing its server-class product, which it later called “Opteron,” but large impediments stood in the way of a successful launch. Henri Richard, executive vice president and chief sales and marketing officer at AMD, explained AMD?s go‐to‐market strategy:

We anticipated resistance in launching an enterprise-class product, because we lacked credibility and had no history in the segment. We couldn’t be a player in the server market without the support of the tier-one OEMs that controlled it. We were unlikely to land Dell because of its longstanding exclusive relationship with Intel. The other three conduits to the market had their own reasons not to engage. IBM and Sun developed their own proprietary architectures. And HP, the largest x86 server player, was the most politically charged because of its collaboration with Intel on Itanium. We also needed Microsoft to develop an operating system for Opteron. (See Exhibit 6 for tier-one OEM server shipments and market shares.)

In spite of the challenges, AMD rallied behind Project MAID. Ruiz explained how the company planned to remove the barriers systematically:

If you looked at the profile of our top customers six years ago, you wouldn’t have recognized their names. They were certainly not HP and IBM. You wouldn’t have recognized our shareholders either. This was a problem, and we knew we had to change it, so we came up with

6 AMD: A Customer-Centric Approach to Innovation 507-037

Project MAID before we launched Opteron. We knew if we executed on Project MAID, we’d change the company. The “M” in MAID stands for Microsoft. We had to convince Microsoft that AMD’s success would be good for them. The “A” is AMD and our mission of “customer- centricity” (explained later). The “I” is IBM. They are a very good technology company, and we needed IBM to put Opteron in one of its servers and be with us at launch to tell the world it was a really good processor. The “D” stands for Dell. We thought if we executed on the first three elements of MAID, we’d get Dell, the market leader.