The Phantom of the Fab
There was a story a few days ago about AMD probably building a fab at Dresden, but it didn’t make a whole lot of sense. For instance, the price tag was only $1.3 billion, which just isn’t enough to build a new fab these days.
Those who put out the original article have since written a new article on the proposed deal which makes rather more sense.
We’ve said in the past that AMD has no money for a new fab. That was true, that remains true. This new fab will be made possible only by a financial package backed and/or paid by the Sazony and national German governments covering almost all the cost of the fab.
The exact details are unclear, but it looks like the Germans will pay for some of the plant, and offer loan guarantees for most of the rest, though I could be wrong on this, I’m working from Babelfish translations for some of this.
In other words, if AMD can’t pay the loans back, the German governments will. That’s the only way AMD will be able to borrow that kind of money.
If I have it right, this little transaction will end up eventually more than doubling AMD’s already high debt, from roughly two to over four billion dollars (perhaps approaching five). To be able to pay it back, AMD is going to have to make much more money and be much more successful than they’ve ever been able to in the past.
Put another way, strictly financially speaking, if AMD has another run of losses like they did in 2001-2002, they’ll fold, period. On the other hand, should AMD be in danger of doing that, the regional Saxony and federal German government would have to pony up something in excess of four billion dollars if they did.
This kind of money would make an AMD default a German political matter.
A while back, we talked about the possibility of AMD becoming a SOI (Slave of IBM). With this deal, AMD enters into a sort of slavery agreement with Germany, and especially Saxony, though it’s hard to tell who is more enslaved to whom.
FX Opterons and Sun
AMD announced 2.2GHz Opterons. Since the Athlon 64 FX is a 2.2GHz Opteron, this is hardly earthshattering news.
What is rather more interesting in the long-run is AMD’s roadmaps for Opterons the next few years.
The next speed bump won’t come until Q2 2004, and if the current methodology of “add 200MHz for every “2” we add to the model number” holds up, we won’t see a 3GHz version until the second half of 2005.
Then a miracle happens, well, maybe.
It will take AMD about two years to go from 2.2 to 3GHz or a bit more. With the introduction of the K9 in 2006, the numbers will jump to the 4.4GHz range (of course, that presume the numbering format remains the same).
If this is so, it makes one believe the K9 is going to be dual-core.
A Setting Sun?
There was a big to-do about AMD and Sun getting into a strategic alliance and good stuff like that.
That sounds really important until you realize that Sun sold less than 60,000 servers last quarter, and now has less than 5% of the server market.
Sun hasn’t been doing too well lately, as a matter of fact, they’re doing terribly. They keep losing market share to the big three (HP, Dell and IBM, who between them have two-thirds of the server market).
BTW, IBM only has about a 16% market share in the server business. total. So IBM plus Sun add up to just a little over 20% of the server market, and neither IBM nor Sun are going to just sell Opterons. IBM will continue to sell servers with PowerPCs, and yes, even Intel chips (matter of fact, they just introduced an Itanium2 server). Sun will continue to sell UltraSPARCS and Intel servers.
While we’re on the subject, just how big is this server market, anyway?
Just How Big Is It?
People tend to grossly exaggerate the importance of the server market to the CPU industry.
Server CPUs are gravy. Really tasty gravy. You modify your desktop chip a bit, and get a lot more money for it.
It’s a great sideline, but that’s all it is. It’s not a meal, and a CPU can’t survive on just gravy.
How many servers were sold in the world last quarter? As the link on the last page said, 1.37 million
How many desktop systems were sold in the world last quarter? From the same counter as for the servers (Gartner), 42.5 million, or over thirty times more.
Yes, servers can come with more than one CPU, sometimes quite a few more than one, but most servers have just one or two processors. Even assuming three processors per server, the desktop market would still be ten times the size of the server market.
Desktop sales remain the meat and potatoes of the CPU industry. Server sales are a very nice supplement to desktop sales; they aren’t substitutes for server sales.
Let me put it another way: if AMD made every single server CPU in the world (that means zero Xeons, zero PowerPCs, zero SPARCS, zero Itanim2s) and nothing else, their CPU production would go down, quite a bit.
This is something AMD is in great danger of forgetting with this new emphasis on servers. They can’t price their desktop chips skyhigh in an effort to protect their server sales. All that does is kill your desktop sales; you save one server chip sale and lose twenty desktop sales.
It really isn’t even necessary; so long as your server chips are MP-capable and your desktop chips aren’t, that separates the two out well enough.
AMD is in no danger of capturing the whole or even a huge chunk of the server maket anytime soon. If they have 10% of the market a year from now, they’ll be doing quite well. Getting into the server market is going to be a long, hard slog that’s going to take time, even for a good product.
What AMD does right now with its pricing with Opterons and FXs doesn’t matter much, they can’t make many of them at the moment anyway. Sky-high pricing will only hurt once AMD starts pumping out production of desktop chips starting around three months from now, and that’s what I’m afraid they’re going to do.
Do You Ask Cheerleaders How Good The Team Is?
I know I’m going to get emails complaining about this, and what they’ll boil down to is “Why aren’t you being positive about AMD (like everyone else)?” Often, the message becomes, “AMD can do nothing wrong. Why are you so biased?” and they haven’t a clue how ridiculous they look.
The answer is simple, “Because I’m not a $%$#$%$# cheerleader, that’s why. My job is to inform you, good and bad not to make you feel good about “your” team.”
I don’t just swallow what AMD PR gives me; I try to look a little bit deeper into the matter, and when everyone else is waving pom-poms and just telling you the good, one does feel the need to emphasize the bad. Why repeat what you’ve already seen elsewhere?
If you don’t think that isn’t even-handed, how come you never complain to the people who just give you good news? 🙂
Yes, building a new fab is good news, but adding billions of dollars of debt when you’re already on financially shaky ground is not. You already know about the fab, but you need to know about the debt, too. It’s a potential time bomb a couple years down the road.
Having a recognized name in servers get on board is good, but when that name has more rep these days than sales, you ought to know that, too.
The world is rarely black and white. Very few things are all good or all bad. They’re usually somewhere in the middle, leaning one way or the other.
Besides that, to be properly informed is not just a matter of knowing facts, but understanding the role of time. The world, even the computer world, does not change instantly and radically. It often, usually, takes a few years even in the computer world for a new product or concept to be successful.
People expect instant results, and it just doesn’t happen.
In the case of Opterons, it will take years for them to be embraced by that particular marketplace. I don’t say that because I hate Opterons, I’d say that about any new type of processor. AMD is going to have to work long and hard to convince people to take a chance and try them out. It’s not like the same people are all running out and buying Itanium2s, either; even the Intel-lovers.
That’s just the way it is.