Interesting post from PTNEWELL - comments Zeev?
===========================================================
I am going to make a series of predictions, mostly about Rambus. If you look back on virtually any predictions from two years ago, you will find them all woefully wrong. Early in 2000, DataQuest predicted that Rambus would have >40% of the market by now, and rising fast. Semico predicted RDRAM would have 0.8%, and falling. The latest report in Digitimes put RDRAM at 6% of the market and holding (although if it really was 6%, this was almost surely a new high). In other words, no one was within a factor of 5 of the truth. Incidentally, Semico's Sherry Garber was NOT worse than DataQuest. She just erred on the side opposite to our hopes.
Bearing that in mind, I offer these predictions, secure in the belief that at least I cannot do too much worse than the official analysts, and hopeful that no one will take them too seriously.
Prediction 1: ATI will be chosen by Microsoft to provide a P4/RDRAM chipset for the Xbox2
Reason: Jack Robinson ran a story about 6 months ago stating that ATI was developing an integrated RDRAM chipset for the P4. This seemed odd, since ATI makes graphics cards, not CPU chipsets. However building the graphics system for a video game is a lot like making an integrated chipset ("integrated" meaning the graphics and video functions are incorporated). When NVIDA won the first Xbox competition, it decided to enter the chipset market. Also, it has been widely reported that the Xbox2 will use a P4, and that the P4 works best with RDRAM. Finally, The Inquirer reported last week that ATI and Via were the finalist for the Xbox2. (The Inquirer did not say anything about the memory types proposed – this is my inference). Finally, insofar as Intel has any say, they would probably prefer to work with ATI rather than Via.
Warnings: Obviously ATI could lose this competition, even assuming I am making the right inferences from these stories. It is a safe bet that Via's entry is DDR-based.
Prediction 2: The Nasdaq will fall below 1000 over the next two years
Reason:
The debate recently has been between those who believe that the Nasdaq will go higher because the economy is recovering, and those who believe stocks are still far overpriced. The much ballyhooed accounting issues have actually been around for years, and would not be taken seriously now if stock valuations were reasonable (that is, if companies were actually generating the cash to defend their stock prices).
It appears to me that the rule "Recovering economy equals rising stock prices" is generally a good one. The trouble is that for most of 200 years, the economy had both ups and downs equally often. Recessions and even depressions use to occur every two years or so. But throughout the 90s, the economy grew, and stocks grew faster (much faster than earnings). Since the Fed will not allow a prolonged recession, either stocks will outstrip earnings for all time (impossible, since everyone would become a millionaire, yet the number of goods produced by the economy does not allow everyone to live a millionaire lifestyle) or stocks will revert to their true valuations, recovery or not. When stocks are overpriced, it becomes a game of whoever gets out first keeps the cash, and everyone else loses. Of course "everyone else" will be quicker on the sell button during the next rally. The longer that game is played, the lower stock prices go. The limit is hit when the cash generated by the company is used to defend the stock price, by dividends, or stock buybacks. Very few tech companies have the free cash flow to defend anything like there current stock prices. The bear has a long ways to go yet.
Prediction 3: MU, PMCS, AMCC, VTSS, will have fall below $5/share over the next two years. Intel will fall below $20/share, probably to $15/share or lower.
Reason:
The first four companies all had awesome years in 2000, when spending on information technology in general exploded as never before. However interestingly, none of them earned enough in 2000 to justify their present stock price (as opposed to say, investing in a T-bill). All lost money last quarter. None are likely to ever repeat 2000, in my estimation. MU, for example, earned an average of $0.40/share over the last 4 years. For a capital intensive commodity business (the other major example is airlines), where one must always worry about buying more expensive equipment, and new potential competitors, those earnings makes the company worth about 5*$0.40=$2/share. MU has had the advantage of a high stock price, which has let it generate billions from the capital markets. But of course if the stock price drops to single digits, that becomes a harder sale. If Micron ever loses it ability to print stock and various options to raise money, it will face the possibility of a bankruptcy in a downturn.
Intel is a harder call, since it remains profitable. It just does not have enough real profit or fast enough growth to justify the current exorbitant price.
Prediction 3: AMD will outperform the SOXX over the next two years
Reason:
When AMD beat Intel to the 1 GHz mark in early 2000, most of us assumed it was a fluke. In fact it soon became clear that AMD was not only making faster CPUs than Intel, but that the Athlon (at least from June 2000, with the introduction of the Thunderbird) did more per clock cycle than Intel's CPUs. With the P4, Intel supposedly introduced an architecture that would allow it to ramp up to 10 GHz over a few years, and leave AMD in the dust. Yet so far AMD, although not able to keep up in GHz, has kept up (or even has a small edge) in performance. AMD computers do more work per clock cycle. (This is the real reason AMD has not needed RDRAM as soon as Intel did – when each instruction does more work, the memory does not need to feed instructions to the CPU as fast. The usual explanations about the relative importance of latency are bogus, and ignore factors like page misses and the single addressing strobe rate that raise the effective latency of DDR. Anyway, PC1066 will have lower latency than DDR).
It is an extraordinary story to have a perennial "chimp" producing computers that outperform the gorilla, and to continue to do so for two solid years. Intel has briefly taken the lead back on a couple of occasions, but it has not lasted. I still would not bet on AMD as a good investment, given its very high research and capital costs just to stay in business. But it has a P/S far lower than other members of the SOXX, and it is starting to develop a reputation for something besides cheap Intel clones. There is at least the possibility that Intel faces real long term unsquashable competition with a good public reputation. AMD should at least beat the rest of the vastly overpriced SOXX.
Prediction 4: The Hammer will include a PC1200 RDRAM version, and that will be the mainstay for the Hammer
Reason:
The recent reports are the main reason for this prediction. However it is also a simple fact that there will not be any plausible likelihood of any type of DDR that can remotely approach 1200 MHz RDRAM anytime soon. It seems clear on basic principles that the faster computers are, the more they benefit from fast memory. If the Hammer is going to be impressive, and a major advance on what is available today, it will almost surely need something much better than DDR can offer. There is only so much that can be done with making each instruction do more work; eventually you need to fill the cache faster.
Prediction 5: Rambus will win a broad court victory on appeal
Reason:
Words used in a patent, such as "bus", are supposed to have their usual meaning, a principal the appeals court will uphold. Moreover the court appeals court will declare that patent claims which do not even mention the word "bus" do not require any specific type of bus to be applicable (seems obvious, doesn't it?) The appeals court will also rule that Rambus could not have possibly defrauded anyone in JEDEC by applying for patents on technology it had already been certified as inventing. The obvious fact that key Rambus patents date from long before Rambus entered JEDEC, and that some additional claims were made long after SDRAM was on public sale with specifications available over the web, make a mockery of the claim that Rambus in any way used what it learned in JEDEC.
The appeals court will also rule that the infringement was intentional, and that Rambus is therefore entitled to extensive damages.
In other words, justice, not to mention common sense, will prevail.
Prediction 6: Rambus will collect little or nothing from MU, Hynix, and Infineon.
Reason:
By the times these companies exhaust the avenues to continue the court fight, it will be 2004 or 2005. By then if any of the three are above are still in business, they certainly will not have the money to pay Rambus. They may be forced to pay in stock shares, but those will have relatively little value by that time. There is simply no way for Rambus to ever get back the vast sums stolen from it by these companies.
Prediction 7: The DRAM market will total $20 billion in 2002.
Reason:
Wild guess. I don't pay any attention to model calculations, since they never seem to do the official forecasters any good whatsoever. When they prove they can do better than pulling a number out of a hat, I'll start paying attention again. In 1999, 2000, and 2001, the industry was $20B, $28B, and $15B respectively. With the way prices have shot up recently, this must be a better year than last year. The same as 1999 seems as good a guess as any.
Prediction 8: Rambus earnings will rise to $0.14 share per quarter by January 2003
Reason:
For two years now we have had lots of predictions that Rambus was about to collapse, and a few predictions that sales were about to explode (actually they had started to do the latter before the Richmond case went against Rambus, but the sharply increased revenue was not for RDRAM). It is time to simply look at the recent past and project it into the future. In the January report, Rambus stated that RDRAM sales were rising at a 30% quarterly rate (which actually covered the difficult July-September period, when the i845 was introduced). I will simply suppose that RDRAM bit sales roughly rise by 1.3**4=2.86, while the market as a whole rises by 60% (in bits) over the next year. That might give RDRAM 10 to 11% of the market (by bits). Unfortunately, prices will also fall during that time, at least for RDRAM. Also, Intel will not pay anything more for using RDRAM in their chipsets. The $10 million/quarter from Intel is fixed. So altogether I am (wildly guessing) that revenue might rise to $34 million/quarter, while expenses stay about the same.
Warning: The author is NOT Nostrambus, although he plays one on the Internet. Take any of the above to the bank at your own peril. |
