Things finally seemed to be going well for Intel Corp.'s archrival, Advanced Micro Devices Inc. Its K6 microprocessor, acquired when AMD bought boutique chip designer NexGen Inc., was unveiled in April to great acclaim. AMD Chairman W. Jerry Sanders III predicted that his company's new $2 billion Austin (Tex.) semiconductor plant could churn out 4.5 million K6 chips this year, on its way to earning AMD a 30% slice of the Intel-compatible processor market by 2001.
Oops. Manufacturing the K6 in high volumes has turned out to be yet another sobering challenge for AMD. Over the past two quarters, snafus in the process of layering silicon and chemicals has resulted in too few chips that are up to snuff. Analysts estimate that less than 50% of K6 chips etched onto each wafer of silicon pass muster, compared with an estimated 61% for Intel's Pentium MMX chips. What's more, those that do work are not as speedy as AMD had hoped, which means they sell for less money. Sanders is contrite. "We're behind on our learning curve," he says. "Ergo, a wider loss."
Indeed, AMD lost $38 million in the third quarter ended Sept. 28. Its stock, once buoyed by investor optimism over the K6's prospects, is now trading at 22, less than half its 52-week high of 48 1/2 last March. "AMD is exhibiting frightening incompetence," says Peter Christy, president of researcher Micro Design Resources.
Complicating the Sunnyvale, (Calif.) chipmaker's situation, Intel has slashed its own prices on Pentiums by up to 58%--and a lot faster than AMD expected. That has collapsed the price umbrella that AMD hoped would let it reap big profits off chips it sells for 25% less than Intel's.
NEW DIMENSION. But don't count AMD out. Sanders has a two-prong strategy for a comeback. First, he's betting that the company can solve its production problems by shrinking the K6 to get more processors out of each silicon wafer. This approach requires circuits only 0.25 microns across--some 1/400th the width of a human hair--and promises to yield twice as many chips per wafer as today's 0.35-micron process. Sanders hopes to convert entirely to 0.25-micron technology by June, 1998. That's why he's sticking with his goal of selling 15 million K6 chips next year, or about 14% of the Intel-compatible chip market. Analysts estimate that should help drive earnings to $194 million in 1998.
Can Sanders do it? Analysts concede that moving to a 0.25-micron process will increase the number of chips per wafer and boost their speed. But "if they can't get it right on 0.35 micron, they're absolutely going to have problems hitting 0.25," says analyst Jon Joseph of Montgomery Securities.
AMD has another chip up its sleeve, anyway. By the second quarter of next year, Sanders promises to roll out the K6 3D, which adds three-dimensional graphics processing onto the chip for eye-popping images. Sanders already has one taker: On Nov. 17, DreamWorks Interactive announced it will develop a game called Trespasser, a digital sequel to The Lost World: Jurassic Park, tuned for the K6 3D. Later in the year, AMD will deliver the K6 3D+, a higher-performance version that promises even more power.
It sounds good--if AMD can deliver in the volumes that personal-computer makers need. The company has signed up 100 customers for the K6, including IBM and Acer. While major PC makers such as Hewlett-Packard Co. already use AMD's older K5, they haven't jumped to the new chip. Analysts say that Compaq Computer and HP have discussed using the chip as soon as AMD can guarantee volume delivery of 233-Mhz K6 parts.
That would boost AMD's credibility at a crucial time. AMD isn't the only one challenging Intel these days. The new battleground: the fast-growing market for under-$1,000 PCs. This year, Cyrix Corp. has grabbed an 18% market share in the low end vs. AMD's 2.5%. And Integrated Device Technology's Centaur C6 chip could make inroads in 1998. With rivals nipping from all directions, Sanders had better get it right this time.