SUNNYVALE, Calif. -- Dana Krelle hunches over a motherboard in his windowless office at Advanced Micro Devices (AMD) - Get Advanced Micro Devices Inc. Report. He's in a hurry. Krelle has a plane to catch and his boss, AMD President Atiq Raza, wants to see him pronto. But he can't stop talking.
Krelle, who heads chip marketing at AMD, is showing off the chip that his company hopes to unleash in the next couple of weeks as an
killer. Called the K7, the chip is designed to deliver a speed for PCs that Intel won't be able to match for at least a year.
"We've always been two clock rates behind Intel," says Krelle. "Intel is now at 550 megahertz; our highest speed on the K6 is 450 MHz. Now we want the highest performance and highest clock rate."
Despite misfires in its K5 and K6 lines, Krelle is convinced K7 will be a winner. At speeds that could reach 700 MHz, the K7 is expected to be faster than any PC chip Intel will have on the market until next summer. "The K7 can loaf and keep up with Intel's MHz rate," says Nathan Brookwood, a technology analyst with semiconductor research firm
AMD is depending on this chip to change its fortunes. The company has posted losses in nine of the 13 most recent quarters. In its latest quarter, AMD
lost a whopping $128 million, twice as big as any quarterly loss in the past three years. The stock this year has lagged behind the
Philadelphia Semiconductor Index
, falling 36% even as the index has posted a 6% gain. While it has recovered recently, rising 25% in the past six weeks, AMD stock is still down 44% from the 52-week high it reached in mid-January.
Chips Are Down at AMD
Things are starting to look so grim that some analysts say the company will be forced to sell out should losses continue, a scenario that AMD dismisses. But "we are a survivor," says Ben Anexter, director of external affairs, who has worked alongside founder and CEO W.J. "Jerry" Sanders for 27 of AMD's 30 years. "We aren't exactly a flash in the pan. This is a very long-term pursuit."
What's more, the K7 is AMD's chance to redeem itself from the K6 fiasco. Designed to supplant Intel's Pentium II chips, many of the K6-2 chips had to be
tossed away in the past two quarters because they were too slow for even dirt-cheap, no-name computers. That followed the K5, which came out a year late in 1996 and never made a profit.
"AMD has had an abysmal history of manufacturing and delivery problems related to the K5 and K6,"
U.S. Bancorp Piper Jaffray
analyst Ashok Kumar wrote in a May 24 research report. "The K7 is the third time that the company is asking investors to make a bet that it can execute." (Kumar has a neutral rating on the stock. Piper is not an underwriter of AMD.)
The K7 is different, Krelle says. The problems with its predecessor were design-based and borne out of a misguided strategy. The company first designed the K6 two years ago to beat Intel's chips on overall system performance, not just speed. But consumers focused only on megahertz, much the same way that sports-car shoppers obsess over how fast a model can go from zero to 60 mph. Intel, fighting to keep consumers from upgrading to the K6, pushed up its own chip speeds. To keep pace, AMD was forced to push the K6 to its breaking point.
With the K7, if consumers want megahertz, they'll get megahertz. Behind the new chip is Dirk Meyers, who led the
team that built the Alpha server chip. The Alpha is the world's fastest microprocessor. "The K7 has been designed and defined to obey these new market parameters," Krelle says.
The K7 faces some hurdles of its own. The chip is AMD's first that won't simply snap into an Intel-compatible motherboard. This means AMD has to line up the makers of motherboards, chipsets and buses, without whom AMD will be left with a chip that has no home. Krelle says many are already lining up. AMD has cleared such hurdles before. Two years ago, few thought AMD could crack the top-tier boxmakers, but it's now in nine of the top 10 -- all but
AMD must move quickly. Intel's next-generation Willamette chip is just a year away.
In the meantime, what keeps AMD going is the satisfaction that it's putting the giant chipmaker on the defense. Intel's Pentium III chip, after all, has shown disappointing sales despite a $30 million marketing campaign. In April, according to market research firm
, the P3 was in only 7.8% of all desktop computers sold. Intel chips had 53% of the total market to AMD's 41%. If AMD had been able to pump out large numbers of 450-MHz K6 chips, the P3 may have fared worse.
In spirit at least, AMD thrives on its rivalry with Intel. Even in tough times, AMD has managed to attract some of the best engineers and designers in the business. They join AMD because it is the anti-Intel, says Keith Diefendorff, editor of
and a former AMD employee. For a microprocessor designer, Diefendorff says, besting Intel is the ultimate challenge.
Publicly, Intel won't talk about AMD. Privately, it admits AMD is a formidable competitor. "We don't take them lightly," says one Intel official. "To do that would be a mistake."
With the K7, AMD is stepping up the attack and targeting its loftiest goal yet: the server market, where Intel chips sell for $3,600 each. "We want in," says Krelle. The way he sees it, those chips cost Intel little more to make than the ones it sells for several hundred dollars each. "It's pure margin," he says.
Breaking into the server-chip market would force Intel to drop its prices, just as it has in the market for desktop chips. That would give Intel less room to subsidize chips at the low end, where it has been brutalizing AMD with price cut after price cut. And even taking a tiny piece of that market would help AMD lift the average sales price of its chips, which
sunk to $78 in the first quarter, down from its goal of $100. Compare that with Intel's $200 price tag.
AMD has set for itself other long-term goals, such as sustained profitability and a 30% unit share of the total microprocessor market -- desktop and servers -- by 2001. "We will have to deliver 50 million units" a year, Anexter says. That's quite a task considering its quarterly record so far was 5.5 million units, or 22 million at an annualized rate. A planned fabrication plant in Germany should more than double its production by using technology that produces more chips per silicon wafer.
The K7 could give AMD a big push toward those goals. And if it bombs? In the words of AMD spokesman Scott Allen, "That would be really, really bad."