Intel, AMD neck and neck in chip costs
Each chipmaker claims it can crank out chips cheaper than the other guy. But after doing the math and studying divergent studies, analysts conclude it's a dead heat.
 | ||||
| | | ||
| Reader Resources Athlon-powered desktops  Pentium 4-powered desktops | | ||
| ||||
| ||||
Intel and AMD, however, tell a different story. For the past year, AMD has said the smaller size of its Athlon and of its upcoming Hammer chips give the company a physically intractable advantage.
Intel Chief Financial Officer Andy Bryant threw gas on the fire at the company's analyst meeting in April by declaring that Intel's new factories, which process larger wafers, with 300-millimeter diameters, make it the low-cost leader. He added that Intel's yield, or number of good chips per wafer, was 50 percent higher than that of an unnamed competitor.
"Through these investments we have cost leadership," Bryant said. "Don't buy a story that says Intel has a cost problem."
Days later, AMD Chairman Jerry Sanders gleefully responded to Intel's claim of superior yields.
"How do you spell 'bull****'?" Sanders asked investors and analysts at Merrill Lynch's Hardware Heaven conference in San Francisco in late April. "The only way they could do that would be to invent a perpetual-motion machine...We will put our yields up against anybody's."
A cost advantage would be a crucial weapon for either company. Astronomical fixed costs are a brutal fact of life in the semiconductor market. Manufacturers have to spend billions on factories, equipment and employees before the first chips even come out.
Success or failure largely hinges on making as many chips as possible to spread the "hard" costs as thinly as possible. Smaller chips, or larger wafers, lower expenses by letting makers produce more chips from each wafer. Low manufacturing costs also ensure low prices for consumers.
Tough talk aside, neither Intel nor AMD holds a major edge, a situation that won't likely change with the next generation of chips. Still, the equations are filled with so many variables and accounting assumptions that the debate will likely never be settled.
"I don't think either one has a significant advantage" in terms of cost, said Kevin Krewell, an analyst with Microprocessor Report, an influential industry newsletter, who added that calculating costs "is an inexact science."
Instead, chip prices will be the overriding factor in the competition, and once again AMD's lower prices will likely hurt the company.
Sizing up the competition
The battle between AMD and Intel largely comes down to the differing strategies each company took with the design of its chips.
Sunnyvale, Calif.-based AMD focused primarily on reducing the size of its chip, called the "die." (The die is tested, wrapped in a ceramic or plastic package, and then sold as a processor.) "Thoroughbred," a version of the Athlon chip manufactured using the 130-nanometer manufacturing process--which refers to the average size of the features on the chip--occupies 84 square millimeters, 36 percent smaller than the latest Pentium 4 chip, which takes up 131 square millimeters.
"AMD has a distinct, differentiated strategy," said Joe Osha, an analyst at Merrill Lynch, who added that the manufacturing costs are close. "They don't have the ability to spend money (on new factories) at the same rate as Intel"--new factories that would allow AMD to produce larger wafers. "The strategy is about as good as you could have come up with."
| Intel CFO Andy Bryant said Intel's yield, or number of good chips per wafer, was 50 percent higher than that of an unnamed competitor. |
"Prescott," the next version of the Pentium 4, will occupy 90 square millimeters, according to AMD estimates, while Clawhammer, AMD's next-generation chip, will take up only 63 square millimeters. By that time, Thoroughbred will take up only 50 square millimeters.
Intel's die sizes "are just too damn big," said Sanders, adding that AMD enjoys a 43 percent to 80 percent cost advantage, depending on the chips being compared.
But while Intel's die is larger, the company claims the difference is more than compensated by something AMD doesn't have: factories processing 300-millimeter wafers.
The 300-millimeter wafers--the measurement is that of their diameter--drastically cut costs through the magic of Euclidean geometry. Approximately 2.25 more chips can be popped out of one of these wafers than can be produced from traditional 200-millimeter wafers.
Put another way, even after the investments in new buildings, equipment and more expensive wafers, chips produced on 300-millimeter wafers cost 30 percent less than chips produced on 200-millimeter wafers. "It takes everything you can factor in," said Nathan Brookwood, an analyst at consulting firm Insight 64.
Lower costs, higher volumes and other factory efficiency programs more than outweigh any disadvantage resulting from die size, Bryant said. Two 300-millimeter chip-fabrication factories, or "fabs," will crank out chips for Intel this year, and there will be four 300-millimeter fabs--three production and one research--by the first half of 2004.
"We have the ability to build everything but Celeron on 300-millimeter by the end of next year," Bryant said, adding that the cost per chip will be around the historical low by the end of 2003.
For its part, AMD counters that the 300-millimeter factories are a liability because they cost billions of dollars to erect and more to staff.
"We don't have to spend billions and billions of dollars on new factories," Sanders said.
AMD won't have a 300-millimeter facility of its own until 2005, when it opens a jointly owned fab in Singapore with Taiwan's United Microelectronics Corp. (UMC). Hiring outside foundries such as UMC typically costs more than doing it in-house, analysts have said, so any cost savings will be diluted to some degree.
So who's in the chips?
When all the math gets tabulated, however, the results are too close too call, according to, among others, Dean McCarron, principal analyst at market research firm Mercury Research.
AMD can get 326 Athlons out of a 200-millimeter wafer. Intel, meanwhile, can squeeze 497 chips out of the more costly 300-millimeter wafers, McCarron said. Assuming that AMD pays $2,500 per wafer, Athlons cost $11 per die, he estimated. Adjusting for 300-millimeter wafers, Intel spends $10 on each Pentium 4 die.
|
"The only way (Intel could beat our yields) would be to invent a perpetual-motion machine."
--Jerry Sanders, AMD chairman
|
Similarly, Krewell, who uses more conservative assumptions, said Athlons cost $22 to produce. Pentium 4s, Krewell said, cost about $30 on standard wafers, but with the 30 percent cost savings of 300-millimeter wafers, the price drops to around $21.
Both analysts, though--like others--add that their estimates are riddled with further estimates. Wafer costs can vary by how investments in real estate and manufacturing equipment get depreciated. Equipment reuse, tax breaks and financial incentives struck with local governments further cloud the issue.
The exact yields are also shrouded in mystery, but both Intel's Bryant and AMD's Sanders are arguably right on this issue. Because its wafers are larger, Intel can probably yield 50 percent more chips per wafer as Bryant said. But in terms of percentage, the companies are likely close, as Sanders said.
There are other variables as well. Intel likely has a slight cost advantage in packaging, Krewell added. The company buys more chip packages, which are made in Asia by other manufacturers, so it probably gets greater discounts. Volume likely also decreases testing costs for Intel. AMD, too, is just starting its switch from ceramic to organic packages, an expensive crossover.
On the other hand, Intel makes only some of its Pentium 4s on 300-millimeter wafers, so the vaunted benefits of the larger platters are only just starting. Similarly, AMD is still making a lot of the older, larger and more expensive Athlons. The transition to Thoroughbred won't be complete until the end of the third quarter, said an AMD representative.
UMC will begin to make Athlons at the end of the year on its own 300-millimeter production line, but on a contract basis, which attenuates some of the savings.
In the end, a more important factor in the competition between the two companies will be chip price, and Intel still holds the upper hand. The Pentium 4 ranges from $133 to $637. Meanwhile, Athlons sell for $130 to $330 officially, but often go for far less because of liberal, unpublished discounts offered to large customers by AMD.
Lower manufacturing costs are "important to your margins, but it doesn't impact the sale price of your processors," Krewell said. "Intel sets the price, and AMD prices theirs against Intel."
For AMD that's not good news. Intel is in a position to raise the speed rapidly on the Pentium 4. In the third quarter, Intel will sell the 2.5GHz and 3GHz chips for high prices, but deeply discount the 1.8GHz to 2GHz chips, which compete against AMD's Athlon, Osha said.
"They are going to price those in a manner that will make it very difficult for AMD to make money," Osha said. He predicts AMD will lose 21 cents a share this year, "but it could be a lost worse than that."