SAN FRANCISCO -- The roller-coaster ride of
is turning into one of the must-watch events on the Street.
On Wednesday, the stock of this controversial high-speed memory-chip designer shot up 83 51/64, or 31%, to close at 350 3/8 after
Morgan Stanley Dean Witter
analyst Mark Edelstone raised his 12-month price target to 500 from 110. His reasoning? He expects Rambus to earn between $3 to $5 a share in calendar 2002, and $7 and $10 per share in 2003. That compares with 43 cents a share for calendar 1999.
This is one crazy stock. On the four trading days before Wednesday, it plunged 156, or 37%, after popular technology review Web site
reported that nine out of every 10 Rambus chips come off manufacturing lines as duds, that Rambus chips cost 750% more to buy than standard, slower-speed memory chips and that their performance gains don't justify the extra cost.
That stock plunge reversed earlier good fortune for the shares. On Feb. 15,
announced that its upcoming
chip for high-end desktop computers would use only Rambus designs, sending the stock soaring 280% over the next four weeks.
For those who haven't read the
dozens of stories
has already written on this topic, here's a quick summary: Rambus was the hottest IPO of 1997. It makes no chips, just licenses designs, potentially gaining the company huge royalties with no manufacturing costs. The memory industry has fought Rambus adoption for years because the chips are expensive to make.
, the world's biggest chipmaker, supports Rambus. Memory is a gigantic market:
Cahners In-Stat Group
estimates $40 billion in so-called dynamic random access memory, or DRAM, will be sold next year, with the number rising to $59 billion in 2002 and $82 billion in 2003.
Whether Rambus, with a market capitalization of $7.9 billion, is worth $500 a share depends on how much share of the memory market its designs will take. Because the stock has been the target of heavy short-selling -- bets that the price will fall -- many investors are pessimistic about the prospects. But not Edelstone, who predicts Rambus will account for half the memory market in 2003.
Intel's memory czar, Pete MacWilliams, tells
that Rambus memory will be in all Willamette machines, as well as some budget-priced computers and notebooks. So what does that mean? Chip-market research firm
estimates that high-end desktops make up 34% of Intel's chip market -- so that somewhere around 38 million computers a year will use the Willamette chip and Rambus memory. If Rambus-based memory made it into all of Intel's low-end machines, we're talking around another 6.8 million or so computers. Notebook computers account for about 19% of their chip production, or about 23 million computers. Ask Rambus execs what they expect, and they parrot Intel.
Where Intel's Chips Go
Source: Semico Research
Sources in the memory industry say that for computers that sell between $500 to $2,000 retail, Rambus would earn royalties of about 1.5% of the cost of the memory chips inside each machine, or between 50 cents to $2 per machine. If Rambus made it into all Intel's high-end notebook and budget computers, that's about 70 million machines a year, based on Semico's market estimate.
Chip engineers outside of Intel and Rambus say that scenario works only in a world without competition, and for Intel, that world no longer exists. Chip technology analyst Burt McComas, of the research firm
, says Intel now has to compete against notebook chipmaker
, which specializes in chips that burn so little power they don't need a fan to cool them. Rambus chips, in comparison, are so hot, McComas says, they would melt a notebook that didn't contain an additional power-burning fan.
And Rambus memory is simply too expensive for Intel's low-end machines, he says, especially because Intel will have to compete against a partnership of
and graphics chipmaker
. Via has said it will use a PC-133 high-speed memory chip, seen as an alternative to Rambus, and S3 has a better reputation for graphics chips.
As for the high-end Willamette? Intel's chief rival,
Advanced Micro Devices
, is expected to come out this summer with its super-fast Athlon chip hooked up to another Rambus memory alternative, known as double-data-rate DRAM.
Chip analyst Nathan Brookwood of
says that while Rambus will give the Willamette more memory bandwidth than the Athlon, it will be an advantage only if consumer demand is strong. And that's in question, he says. "If it turns out ... you have to pay too much for the Rambus support, then that will give AMD a real advantage in terms of the cost of systems with competitive levels of performance," Brookwood says.
When asked about competitors and their products, Intel won't comment. As far as power consumption in notebooks is concerned, MacWilliams argues that it can design its notebook chips and chipsets to keep the power consumption down while using Rambus memory. And in the long term, he says, a switch to Rambus memory in notebooks makes sense.
Meanwhile, Cahners In-Stat memory analyst Steve Cullen says that the market likely will peak in 2003, and memory prices can crash (and along with them, Rambus royalties) in 2004. MacWilliams admits that past 2003, he can't predict what memory technology will be standard. "At some point,
Rambus will run into limitations. Then its important to figure out what is next," he says. "We have an established technology
Rambus. For it to be displaced, we need to come up with something better and more compelling. A lot will depend on what evolutionary improvements we can add onto existing Rambus DRAM."
Rambus now trades at 190 times 2001 earnings. Is it worth $500 a share? Ask psychic