Advanced Micro Devices
AMD's Future Fab Capacity

Format for Printing

Format for printing

Request Reprints


By ValueNut
January 22, 2002

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

I have been thinking about some of the things AMD has been saying and doing lately concerning added future fab capacity -- especially concerning foundry fab space.

Like many of you I find it curious that Jerry said that AMD is not planning to ever allow ASP to rise over $150, in fact, he seemed to imply he liked ASPs right around $100.

That's a bit of a strange thing to say, because if AMD really begins to capture the high end of the mobile and (possibly) server market in 2002, while switching more of its production capacity to desktop Athlons and away from Durons, AMD would have to begin dramatically lowering their mobile and server ASPs and possibly their overall desktop ASPs to keep their total ASP below $150. It doesn't really make sense because that would have the effect of forcing up demand for AMD processors when they can only supply about 25%-30% of the market. In other words, they would be selling their processors too cheap. If total ASPs don't drift above $150 this year, they are even more likely to next year. What is AMD thinking?

Of course, AMD could simply be trying to warn Intel off from a price war. They might be "telegraphing" to Intel, "keep those prices have nothing to win by dropping them". But I don't think that can be the whole story.

I think AMD's public mis-direction exists, but goes a different way. I think keeping ASPs low only makes sense if you have the ability to bring on a lot of new capacity; otherwise the demand generated by having low prices is wasted. I don't mean just outsourcing 20% of AMD's total processor shipments -- I mean something a lot more dramatic and not in 2004, but perhaps in 2002 or 2003.

Here is the problem I am now beginning to think AMD is very stealthily targeting. AMD is going to have a good year in 2002, with an even better year, probably a great year, in 2003. I don't think many people seriously doubt that on this board -- and, more importantly, AMD upper management believes it I am sure.

So, earnings are going to rebound in 2002 and AMD may save up several billion from cash flow in 2003 -- enough to pay for all the costs of the 300mm fab they are beginning to build and are probably nearly ready to announce. In fact, I don't think AMD will need a partner to build this fab and produce parts by or before 2005. I think AMD management knows that they can pay for that fab out of cash flow.

However, I think that AMD knows that if they do only what they have announced in terms of their capacity expansion road map, they will allow Intel to retreat into the part of the market AMD can't supply, lick their wounds, and buy and/or finishing developing technology that can compete with AMD in a year or two. That means in the long run, they might have a stronger competitor, even though Intel is relatively weak now and next year.

Let's take a fresh look at CPU unit sales for the next few years. If we assume global x86 unit sales of 150 million in 2002 and growth of an average of 12% a year for the next five years, we get expected units like this:

Year    Global x86 Unit Sales
2002          150 million
2003          168 million
2004          188 million
2005          211 million
2006          236 million

What does AMD expect its own production of x86 CPUs to be? Let's assume AMD ramps Austin to flash this year and begins to produce some externally fabbed CPU parts. That should result in, say, 40 million CPUs this year. In 2003, Dresden alone will be able to produce 50 million, but with 20% of capacity coming from an outside fab, that could be as high as 50 million total CPUs for the year (assuming most of them are Athlon). When AMD fully ramps 0.09 micron by the end of 2004, AMD should be able to produce 120 million Athlon CPUs including what comes out of their external fabs. However, their average die size will have grown, so they are probably only going to be able to produce 80 million units assuming a 30/70 Hammer/Athlon mix. Finally, in 2005, Fab 35 will begin producing some parts and by 2006, could be producing as much as twice as many 300mm dice as Dresden -- meaning around 120 million Hammer processors, versus Dresden's 60 million Hammers. In 2006, both fabs will hit 0.065 micron and eventually be able to produce 200-300 million K9 processors.

Here is my guess at AMD's capacity expansion in units and percent of the x86 market:

Year       AMD's x86 Unit Sales         AMD's unit market share
2002             40 million                       27%
2003             50 million                       30%
2004             60 million                       32%
2005             80 million                       38%
2006            150 million                       64%
2007            200 million

The problem is that without really high volumes in 2002 and 2003, AMD cannot take much market share. The most obvious AMD strategy would be be to raise prices as marketshare begins to hit 30% and start saving money for a new fab, so as to take on Intel as an equal in 2006.

However, that doesn't sound like what Jerry and Hector are planning. And I am not sure that strategy makes sense. Intel may not have a solution to Hammer in 2003 or even 2004, but if you give them enough time, Intel will come up with one. They are not stupid and have lots of money and if they only drop to 70% market share next year, it won't kill them. They can beat a strategic retreat, raise their prices in their 70% of the market and come out with a real Hammer-killer in 2004 or 2005, chastened but a wiser and more dangerous opponent.

I think Hector and Jerry are trying to come up with a knockout punch that is, at the moment, off-the-roadmap, to take advantage of their growing cost and performance design advantage in 2002 and their likely overwhelming advantage in 2003.

What is that "knock-out" punch? I don't have AMD management's ear, but I can tell you what I would do if I were Hector. I would try to find a "hungry" fab partner with lots of spare 0.13 micron copper fab capacity (preferably at 300mm), and eager to sell the capacity for a comparable price to Dresden's operating costs who will greatly increase AMD's capacity as demand for AMD's products exceeds Dresden's supply.

In fact, UMC, the rumored partner AMD supposedly has an agreement with, has all of this. UMC's costs of manufacture seem to be comparable to or lower than Intel's, since they already have 0.13 copper production mature on two different in-production 300mm fabs. Each of these fabs has a design capacity of 40,000 300mm wafers per month, though I imagine they are nowhere near full capacity yet. These are not small fabs. See the website for details (click on 0.13 micron and 300mm).

UMC needs someone like AMD to provide orders to keep their fabs busy. UMC's revenues are reportedly down over 50% from last year and foundry utilization has dropped from 95% last year to under 50% (perhaps only 30%) this year -- even though they have shut down a lot of non-130nm capacity. (See

It looks to me like UMC should have no trouble providing any quantity of Athlon dice AMD were able to order. They should even be able to supply the highest frequency Thoroughbred products with a little help from AMD, though SOI products like Barton and Hammer may not be so easy.

Furthermore, UMC really needs a partner with high volumes. A high-profile partner like AMD would also be good for UMC's image (they make a big deal about fabbing Sun's Ultrasparc), so they are likely to give AMD a preferential rate.

So far, so good. I am only saying what everyone already expects. AMD will be able to buy as many dice as they want -- at least within the limits of AMD's contractual obligations with Intel -- whatever those are (20%? 40%?).

However, what if AMD negotiated a deal with UMC to get guaranteed capacity out of one of UMC's 130nm fabs now? In other words, what if they bought a fractional share of one of UMC's fabs? If I were Hector, I might try to buy a portion of their capacity using AMD stock. This would likely be a good deal for both UMC and AMD. AMD would be able to perhaps double capacity next year from 50 million units to 100 million units and take Intel completely by surprise right when they AMD has the design capability to do so. Since they owned the fab (or at least their portion of it), they would be able to work around agreements with Intel that limit them to some fraction of their output in contract fabs (if there are any such restrictions).

AMD would likely also get a good deal if they could convince UMC to take a small enough fraction of AMD stock to not dilute it too much. Another alternative is to buy the fab on a financed basis, where UMC loans AMD the money to buy their fab and AMD pays it off out of future revenues. Even if financing rates were 12% or so, this would be a great deal for AMD. If they can get several billion in additional revenue, they could pay off the loan early using alternate financing. UMC could get a lot of capacity committed, get a guaranteed revenue stream and have a place for a lot of their people to stay at work. This may be what Hector meant when he said "I'm not sure what to call our new agreement. It's not really's more like an extension of our family". That also fits with what Jerry cryptically said; "My lawyers say that I am allowed to say that an agreements with Intel do not restrict us from doing what we have announced we are doing."

If AMD can arrange to buy more fab capacity, they have the potential to supply a lot more than 30% of market share next year. Potentially, they would not have any manufacturing limits on growth at all, without having to spend a dime on new fabs except the small amount they would have to save for Fab 35 to be completed in 2005. And AMD would have the potential to not only take a lot of Intel's revenue share away by keeping ASPs low, but they might force Intel into losses in 2003.

Intel is very much at risk from this strategy. And I don't have to tell you guys how much AMD's stock price would surge if they had greater capacity, lower costs and superior products to Intel at the same time. You could see market cap go from $6 billion to $20 billion in a fairly short period of time. I don't think Intel's market cap would remain over $200 billion if that were to happen.



TMF Money Advisor
Got money questions? Your answers are just a phone call away! TMF Money Advisor puts you in touch with an objective Financial Planner whenever you need it.