Tell You So
Hal is a hardware engineer who sometimes programs. He is the former editor of DTACK Grounded and can be contacted at halwh@ddj.com.
Once upon a time, Bill the software entrepreneur decided that he'd migrate MS-DOS towards UNIX. This happened many years ago, and most folks have forgotten. But, in addition to being Forbes magazine's richest man in the world, Bill has a very long attention span.
AT&T gripped UNIX with a fierce intensity. Bill had to hire away DEC's operating-system guru so Bill's company could develop a GUI-based UNIX-wannabe. We know this operating system as "Windows NT." (Wouldn't you know it, after NT was developed, AT&T sold UNIX to Novell cheap. Novell then dumped it to SCO even cheaper.)
Anyhow, NT is here and it works, but it isn't selling as well as Bill would like. How is the blame to be placed? Last year, it was decided that the problem was the high price of memory. At a street price of $35/MB, 32 MB would set users back $1120. And if the 32 MB were purchased as part of a Compaq system, for instance, the cost would be much higher. (Gotta pay them dealer markups.)
About the time Bill's marketeers proclaimed, "NT isn't selling because DRAM costs too much," his company released Windows 95, which needed 16 MB, or so said Bill and all the Ziff publications. Non-Ziff publications discovered that, while Windows 95 ran with 16 MB, it ran much faster with 32 MB.
Windows 95 has caught on with home users, but is being boycotted by corporate MIS types, who are sticking with Windows 3.x. Why? One reason is memory cost--Windows 95 needs more DRAM than Windows 3.x.
The thing is, beginning in 1991, DRAM prices remained constant for over four years. DRAMs and CPUs are made in the same clean rooms, use the same silicon wafers, and the same semiconductor production equipment--steppers, diffusion ovens, and such. While DRAM prices held, CPU makers (such as Intel) provided more and more complex CPUs without raising prices. Translation: Toward the end of that four-year period, the DRAM makers had a license to print money. They were fabulously profitable.
Fabulous profits attract competition, but it takes two years to build a modern semiconductor fab. The fabs that came online in 1995 cost $2 billion each; those started in 1995 are budgeted at $2.3 billion. The fabs in the Far East are about 10 percent cheaper for some reason (maybe no campus and fancy landscaping).
That's one reason why DRAM production has only recently caught up with demand--with a vengeance! For a DRAM manufacturer, Figure 1 is scary. This isn't a free fall, it's a power dive. In the old days, the tail would come off the P-38 in a dive this steep.
Figure 2 places recent DRAM price trends in historic perspective. Our very own Feds ("I'm from the government, I'm here to help you.") interrupted a nice, steady 17-year price decline by forcing Japan to sell us DRAMs at higher prices in the late '80s. (The Japanese reluctantly acceded.)
History repeats. Micron, the Utah-based DRAM maker, recently petitioned the government to keep Korea from selling 16-MB DRAMs in this country too cheaply. I am not making this up.
Suddenly, DRAM supply far exceeds demand. The classic solution to this problem is to cut back production. But the interest payments on $2 billion fabs continue even when the fabs aren't being used. The life of a fab is short, because advancing technology obsoletes it in a very few years. And if you don't build DRAMs, you quickly lose your trained workforce. This is strong motivation to continue producing DRAMs.
Suppose I have a computer system that has too little memory for NT. Today (June 30), I can drive down to Fry's and buy two 16-MB 60-nsec 72-pin SIMMs for $210 plus tax. That's $210 for both SIMMs; 32 MB of DRAM for the price of a mid-level IDE hard drive. I might have to toss out the old memory (maybe the old memory can be combined with the new SIMMs, maybe it can't). But, now that the memory needed to run Windows NT is downright cheap, everybody is going to switch to NT--right?
In a previous column (Dr. Dobb's Sourcebook, March/April 1996), I mentioned my former newsletter DTACK Grounded. Coverage was not limited to 68000 stuff; I wrote about whatever interested me. In 1982, UNIX prognosticator Jean Yates predicted that, in 1983, Microsoft would sell $2 billion in Xenix licenses. (Xenix was Bill's first try at cloning UNIX, in this case to run on 4.77-MHz 8088 systems!) I thought Yates' forecast was utterly ridiculous, and I said so. I added that UNIX was most certainly never going to take over the personal computer operating-system market.
Talk about kicking over a hornet's nest! This became a continuing thread in my newsletter, and I got more and more letters and phone calls from self-appointed "experts" telling me that I was flat wrong. Few noticed that Xenix's annual license revenues fell about $2 billion short in 1983. In the early 1980s, I seemed to be the only person in the PC world who was asserting in print that UNIX wasn't going to take over. (Later, I learned of a computer-science professor who was publishing the same opinion, but we never crossed paths. No, I don't know his name.)
About the time it was beginning to dawn on some of my readers that I might be right, AT&T announced the 68000-based UNIX PC. (Remember the BYTE cover story?) In the months leading up to the introduction, I read article after article in the computer and business press about AT&T pressuring Convergent Technologies, which was contracted to build the UNIX PC, to prepare for production of enormous numbers of systems. The production capability was eventually set at 125,000 UNIX PCs per year (according to published reports).
"Hmm," I thought, "If I were running Convergent, I wouldn't set up such a production line unless AT&T had guaranteed to buy at least 50,000 units." And I didn't think that anywhere near that number of UNIX PCs could be sold, even with AT&T being its own best customer--something I speculated on in my June 1985 newsletter, just before the UNIX PC was introduced.
A year later, The Wall Street Journal ran a feature article on how sales of the UNIX PC had progressed. My prediction turned out to be dead on the mark (surely you aren't surprised?), and I wrote this up in my newsletter, which had by then devolved to the free Junk Mail Flyer. An unknown Dr. Dobb's Journal editor picked up and expanded this in Dr. Dobb's Journal (August 1986, page 8) under the heading "FNE Well Grounded."
In a few months, AT&T will surely notice that the Unix PC (built by Convergent Technologies) is not selling. Let us assume that AT&T succeeds in stopping production after only 50,000 Unix PCs have been built. Of that 50,000 about 10,000 will have been sold. AT&T will purchase 8,500 itself, the recently divested [Bell] operating companies, many of which have not yet wised up, will buy 1,483 units, and the rest of the world will buy 17.--Hal Hardenbergh (a.k.a FNE), dTACK Grounded, June 1985.
Last year, Convergent shipped some 40,000 to 50,000 of the Unix PC units. But AT&T has sold only 10,000...and many of those were sold to AT&T's own divisions. AT&T has said that the Unix PC...is "ahead of its time."--Brenton R. Schlender, The Wall Street Journal, May 30, 1986.
Since the UNIX PC debacle, almost nobody thinks UNIX is going to take over the personal-computer world. AT&T had to unload all those unsold systems into the liquidator market. 40,000 of them. I gleefully subscribed to the late UNIX World for years afterward so I could see the small-print ads offering UNIX PCs at lower and lower prices in a desperate search for customers. I think the last price I saw was $495 for a complete system with the UNIX operating system, which was normally sold separately. I wonder how much the liquidators paid AT&T for those turkeys?
I was recently asked what I got for my correct UNIX prediction. I replied, "Not a (darn) thing. I lost a lot of readers who otherwise would have stuck with my newsletter and maybe even bought one of the 68000 boards we were selling."
I hope this gives me some credibility in predicting sales or market penetration of complex operating systems in the personal-computer world. Windows NT is a slightly cleaner, more modern UNIX with a graphics interface. But NT has the same baggage as UNIX: It's so complex that a Systems Administrator is needed, something individual computer users rarely have. NT isn't going to make it in truly personal computing. Not even with cheap DRAM.
NT will find its success in the same corporate MIS arena as UNIX, VMS, CMS II, and other, similarly complex networking OSs. Right now, NT, with the Intel Pentium Pro, is generally regarded as serious competition for UNIX. There is some evidence of large applications being converted from UNIX to NT. This is not a mass marketplace.
I miss getting mail telling me how wrong I am. (Well, my e-mail address is listed here nowadays.) So I'm going to tell you another reason why DRAM supply suddenly exceeds demand. Simple. PC system sales are down. Way down.
Why? Take a look at the PCs all them smart folks are trying to push at us: multimedia, loudspeakers, 3-D, TV displayed on your SVGA (or VGA) monitor. Wow! Terrific! Uh, who needs this crud? I don't. And I don't think bosses want to buy super game machines for employees, either.
So if I (and those like me) don't need this stuff at home, and bosses won't buy this stuff at work, who needs it? Why, those adults who are willing to pay big-time so they, and/or their children, can play really neat arcade-type games on their PC.
The problem is, the PC market is a mass market. Vendors can only sell what the market wants to buy. And while the arcade game-playing PC market is large, it's a small part of the mass PC market. After this limited market is satisfied, there's still an awful lot of fancy multimedia PCs gathering dust in the showroom.
PC makers have cut production way back and are selling their DRAM inventory, driving down street prices. And once DRAM prices start moving down, you want to unload your excess inventory fast. Figure 1 shows what happens then.
After last Christmas, one company got so many returns of its high-end PC systems that it had to put up a tent here in Silicon Valley to store them. Not surprisingly, this company--okay, it's Acer--has just announced a $500 PC with 100-MB disk storage and DOS/Windows included in that price!
These events in the PC world are mirrored in the automobile industry. The auto makers tried to force the public into fancier and fancier cars, at ever-higher prices. It didn't work; the average automobile selling price (including leases and sales to rental agencies) never even reached $20,000, and is actually turning down as consumers go for Saturns rather than Lexi.
There's more money in Big Macs than in fine French cuisine. Prudent companies (that is, those that want to be profitable) will pursue the mass market. This past Christmas proved that Tiffany multimedia machines ain't it. Intel is terribly disappointed, because it was counting on multimedia to sop up excess CPU cycles.
What's Bill up to now? His company is preparing a stripped-down operating system for the upcoming cheap, mass- market, non-multimedia PCs. No, it's not a mini-NT.
Figure 1: Recent DRAM price trends. (FPM = fast page mode, or standard DRAM, and EDO = extended data output.)
Figure 2: Historic DRAM price trends.