On Friday, I pointed out that AMD’s 65nm fab won’t be for a while.
I didn’t have much to say about it then because I was pretty stunned at the time.
Well, now I’m less stunned.
For openers, ignore the spin that it is perfectly normal for a fab to start commercial production at a process level higher than the “real” one.
Yes, fabs can be refitted for process shrinks during their lifetime, but if you look at the statements AMD made about this fab from its creation to, well, the last few days, you will see tons of references to “65nm” for Fab 36, and none for “90nm.”
So this is another “don’t mention it until the last minute” drop of the ball by AMD, have no doubt about that.
There are two big possible reasons for the delay:
It’s probably both, leaning more towards the second.
One Unanswered Question
How much more will it cost AMD to start at 90nm, then go to 65nm as opposed to just starting off with 65nm? After all, whatever 90nm fabrication equipment is being put in to replace the intended 65nm equipment isn’t free.
Or is AMD just cannibalizing some equipment from Fab 30 to get Fab 36 going, then buy new equipment as they can? The number of wafer starts at Fab 36 by the end of 2006 (8-10K) is rather less than the 13K promised earlier on.
How Big A Deal Is This?
If this happened five or even three years ago, this would be a disaster. Intel would introduce 65nm chips that would run much faster than AMD’s 90nm, and AMD would go down again.
However, this is now, and for most of 2006, Intel’s 65nm desktop chips will essentially be Prescotts without the space heater feature. It remains to be seen if they’ll overclock better, but for the average person, they’ll run cooler, but not faster.
Any real threat from Intel will not come until the second generation of 65nm chips, realistically, towards the end of 2006, if even then.
The real difficulty will come from AMD not being able to expand production much until 2007.
And by the way, am I the only person who is wondering about a company that says essentially, “Uhh, gee, I dunno,” when asked about the future fate of Fab 30? This is currently their biggest asset, and they don’t know what they’re going to do with it once they get a second toy to play with?
After all, the tooth fairy didn’t deliver Fab 36. It really shouldn’t come as a shock to them that they’re going to have two fabs soon. They’ve only had two years to think about it, and it’s not like you can do a lot of different things with multibillion dollar fabs; they don’t easily convert to day care centers.
Of course, retrofitting Fab 30 is likely to cost a whole lot of money, which brings us to AMD’s core problem.
The reason why AMD always has been and always will be severely stunted competing seriously against Intel is that they don’t have money, and they’re in a game where you simply have to have it.
Essentially, there’s nothing wrong with AMD that five-ten billion dollars couldn’t fix, for a start, but that weighs too much for the tooth fairy or even stork to handle.
Right now, AMD seems to be hoping that the lawsuit fairy might drop that kind of change in their laps, but nobody’s holding their breaths for that to happen.
Let me put it this way, AMD just had record earnings from its CPU group. It was about $200 million. It would take AMD almost four years of those kinds of earnings, every quarter, just to pay for Fab 36.
In contrast, Intel’s earnings from CPUs are enough to pay for a couple fabs a year.
It’s a huge advantage, and it’s hard to see how anybody can profitably break Intel. Or sure, a handful of companies have the resources to try, but it’s safe to say that any “Kill Intel” move would cost tens of billions of dollars, and leave the CPU industry much less profitable, which defeats the purpose of spending that kind of money to begin with.
Besides, do you really want Bill Gates running the CPU industry, too?
IBM is one of the few others with the resources, and inclination to try, but all they’re trying to do is sneak around Intel with a low-risk, flank attack with Cell. I’m sure the head of IBM’s CPU unit would love to storm the Intel citadel, but the top bosses sure don’t.
AMD needs someone able to front it a big stake at the poker table, but when the chips cost a billion dollars a pop, that’s real money to anybody.
Until some sugar daddy shows up, AMD has little choice but to hustle and do everything on the cheap, and the results of that are delays like this one.