Advanced Micro Devices
Re: 2004 Rumblings

Format for Printing

Format for printing

Request Reprints


By Roleplayer
January 5, 2004

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!

Duttermohl wrote:

AMD has bet the farm on SOI/65nm/300mm. IF they can fill that FAB with their own production, the profit/price per share of the diluted stock will more than make up for the dilution.... It's a risk AMD has to take, and one any investor in AMD has to assess.

Still, AMD has never shown "sustainable profits" for any length of time. When the times are good, they are very, very good for AMD. When the times are bad, they are very very bad. That is the nature of investing in a smallish player in the tech sector.
Lets look deeper into this issue.

I believe that the macroeconomy is a masking agent for what is really happening in the dynamics between AMD and Intel. Not everybody here has followed AMD as early as 1994 (IIRC, that was my first trade in AMD) and IMO there has been a definite pattern in AMD's market share. I specify market share because no other single factor matters as much to AMD's profits; it encompasses in one simple statistic AMD's ability to make CPUs that compete over at least some of Intel's product line, AMD's ability to make them in volume, and AMD's ability to sell them into the marketplace. In the best of times for AMD, their market share has exceeded 20%; in some sectors at some times they have been as high as 50%. I believe their medium term corporate goal remains 30% market share. Today, IIRC, they are running around 17%.

Now market share isn't everything. Pricing is extremely important. If AMD can meet Intel's historical blended ASP of around $150 per CPU, then even today's market share will enable AMD to pay for their new fab out of cash flow. However, market share also carries with it important elements of utilization of resources. Simply put, it is usually better for AMD as a business to sell 100% of its output at 95% or higher of its capacity to manufacture CPUs than it is to have higher prices and be underutilizing capacity. There are some unlikely price and volume points where it might make sense to raise prices across the board and idle some lines if necessary, but unless AMD's worst CPU is competitive with Intel's best CPU, that will almost certainly not happen. This is why Intel needs to have Prescott, or Tejas, or something, because AMD is going to go from relatively few Hammer CPUs to almost all Hammers in the next year (process having already started, natch) and AMD's worst Hammers are somewhat competitive with Intel's best P4s. Since the P4EE is a bragging rights CPU, and not really a moneymaker, I discount it. But that's a topic for a different discussion. I believe Intel will come out with a better 90 nm CPU. I believe Intel will do so in no worse than AMD's Barton/TB time frame, which puts it at about Q3 next year. And when it happens, I believe that even if AMD has the best CPU, there will still be significant overlap between the product lines. AMD may hold the best CPU (depending on whose reviews you choose) or even 20% or more of AMD's CPU sales may be better than Intel's best (unless there's a Prescott EE) but there will be significant overlap: the P4 is a good chip, and its 90 nm analog, when working, may fall behind but won't be left in the dust.

Lets get back to market share and the new fab. I believe AMD is no longer making anything older than TB. TB really got fixed about 18 months ago. How long will a CPU live on any given process and without significant revisions? The P4 Northwood, one of the best CPUs of all time, is now two years old, but a whole lot of its lifespan AMD really wasn't competitive. I'd say three years is the maximum expected lifespan for any non-niche CPU (things like XBox or whatever may keep some older processors alive) and history suggests its a lot less for AMD. TB really got fixed in H2 2002. Opteron shipped in Q2 2003. There is no chance IMO that any 130 nm CPUs will still be made by the time the new fab comes online. Some 90 nm CPUs may still be in demand by the time the new fab is online. Whether 90 nm CPUs are still in demand will depend on how well Intel competes with 90 nm Hammers.

So lets forecast 2006. We know that CPUs are going to have more transistors; they always do. We can reasonably infer that CPU size is going to remain roughly the same; the laser shrinks merely balance out the increased functionality. Since Fab30 is going to EOL (for CPUs) at 90 nm, it will probably convert to some other product once Fab35 comes online. If 90 nm Hammers, of some flavor or another (maybe T&L?) are still viable, then AMD will have significantly more capacity in 2006 than just the new fab. I counterpredict. I believe AMD will either convert Fab30 to something new and exciting for AMD, or more likely start making memory there.

So what does Fab35 mean for AMD? Since the laser shrink's a wash IMO, its increased capacity comes from 300 mm^2 wafers. There are some other benefits as well: 300 mm^2 wafer fabs are more efficient users of resources (water and power), so cost per CPU goes down by more than just the increase in CPUs per wafer. So I figure that Fab35 will mean AMD will have the ability to support roughly 35% of the world's demand for CPUs.

Let me ask: assuming that the infrastructure was in place to test, install, and sell that many, how many people think that AMD could not sell that many Hammers today? In a worst case, AMD could simply drop prices so that the volume Hammers were running under $100, maybe even far enough under so that AMD's blended ASPs were around $80. Hammers are premium CPUs today, and certainly could drive that level of sales. 12M CPUs at $80 ASP puts AMD at between $1 and $2 EPS per quarter. And this is why I think the macroeconomy has been masking things. When AMD was doing its best in comparison to Intel, 2000, the market was scorching hot. But IMO AMD would have done well even if the market was cool because their products were better. When AMD fell in 2002, the market was cold, but IMO AMD would still have done poorly if the market were warmer, because they just didn't compete. In a cold market, the CPUs that don't get sold are the oldest ones, because it's always in a company's best interest to push performance forward. So Intel and AMD eliminated their slowest speed grades, and cut prices on several more products, deciding based on whatever demand remained how many processors would retain good time prices. Those older CPUs get sold to emerging markets to clear out inventory, or even get written off.

I believe those same elements will apply in 2006. If AMD has the performance crown, then they will make money as never before. If Intel has the performance crown, AMD will become somebody else's subsidiary. Which leads me to the hundred thousand dollar question (most of us Fools would be happy with $100,000 profit on our AMD investments): what the hell is AMD thinking? I mean, they are working very happily with IBM, sharing expertise and fab space. IBM is building a great big new fab, 300 mm^2 wafers, that seems like it could make more than all the CPUs AMD is going to be able to sell. AMD could lay off Billions of dollars of risk by using the IBM fab, or at least getting a big partner on their new fab (remember they tried to get a partner on their new fab putting it in the far east, but that deal fell through).

I come to only one conclusion that makes sense, at least since Jerry, the eternal optimist, is no longer running the shop. AMD is convinced that in 2006 they will easily retain the performance crown. That's the only scenario where AMD's actions are sensible. If AMD has the speed crown, then building their own fab means greatest profits. If AMD might have the speed crown, then sharing risks and fab space maximizes profits. If AMD might not have the speed crown, then using somebody else's fab maximizes profits.

We haven't heard a lot about K-9. We don't have the best of information yet on how well SOI works with smaller lasers. There is a risk that AMD is looking at how poorly Intel has kept up with AMD since Opteron was introduced and infers wrongly that Intel will continue to compete poorly; I am quite certain Intel had performance problems relative to AMD in 2000 in large part because they underestimated Athlon. But Intel has had so many years at the top that underestimating has to come easily. AMD just doesn't have the corporate mindset that makes it seem likely to me that they are counting on Intel performing poorly. So I think that AMD has looked at what they have, looked at where they need to go, and feels extremely confident that they will arrive at 35% worldwide capacity with products that will drive 35% worldwide sales and customers that will put them into boxes.

I look to 2004 to see which if any of the flying rumors come true. One I've heard is that AMD is going to get another big Hammer customer: I don't know if that will be something like HP selling Opteron servers, or Gateway or Toshiba or whoever starting to buy desktop or notebook CPUs. Another I've heard is that some new partnership will be forming to help AMD with their new fab. Personally, I'd like to see that: nobody has ever done well in the long run expecting Intel to fail. If AMD gets a partner on their new fab, it's not all bad: it still saves AMD a lot of risk, nobody's going to want to partner up with a company that might go bankrupt, and it still shows more confidence on AMD's part to control the new fab than it would be to (say) be a partner in IBM's new fab.

But for now, I look at the new fab as a very bullish statement on AMD's part, a capital expense equivalent of a stock repurchase program. And as always, I will keep my eyes open, and thank all you other Fools for teaching me enough of the little details (especially engineering stuff) so I think I make some sense of the bigger picture.

Fool on,

Become a Complete Fool
Join the best community on the web! Becoming a full member of the Fool Community is easy, takes just a minute, and is very inexpensive.