Ephud:
Evidently you can't follow the links to download the wafer cost calculator. Hint, its an excel spreadsheet!
But you just proved how functionally disabled you are when you are faced with facts not to your liking. Your calculator assumed 30K WPW and a 130nm 6 metal layer copper process. Funny, I didn't think Dothan used that one. Prescott, according to sandpile.org uses a 90nm 7 layer metal with copper, low K dielectrics, high K gates and strained silicon. The $2860 per wafer doesn't come out in a 5-10K WPW fab on that higher cost process. That also assume high usage, which Intel did say they cut back this quarter (selective memory disorder problem as well). Intel does say that their 300mm fabs cost $4 to 5 billion (from various business news sites, I see no need to prove yet again that you can't follow links given your response above).
I believe when Intel process guys holler at claims that their 90nm process being bad. I believe Intel is running those fabs at low levels of WPW, well below 95% used in your model, to reduce finished goods inventory. Of course with that sky high leverage of unused fab space, that makes for expensive wafers and dies. Adding chipsets, wireless, embedded ARMs and other consumer chips may drop those high wafer costs, but at a price of lower GMs and higher losses for those other uses over the fully depreciated fabs they were running in. And it is a red herring since all of those other uses can't justify a single 300mm fab put together. There is just not enough revenue for that. Thus, the fabs must produce CPUs first and that is where they live and die.
At 7956 GDPQ at 1 WPW, you only need 5000WPW to supply the 40 million CPUs needed. Double that, if the typical CPU is a 112mm2 Prescott and its poorer bin splits instead of the Dothan which Sandpile states 86.7mm2 as the die size. Plug a usage of 25% into your model and I'm sure that the die cost and wafer costs nearly quadruple. So AMD's single Fab 36 at 2500 WPW may allow the 90nm K8 Semperon at 69mm2 to be produced at a conservative 745 GDPW for 24 million a quarter or about 60% of total demand. Figure that 69mm2 die at 90nm is 41mm2 at 65nm which gets us to 41 million per quarter from that small 300mm fab. Which may in fact cost $3K+ per wafer, but with 1224 GDPW actually cost even below your Dothan marginal price.
AMD right sized their new fab. Intel far oversized theirs. That is why their dies cost too much. And you are blind to that fact which probably causes your other "problems".
Pete
PS, I did answer the original premise that CPU sales are highly inelastic. Intel spokesmen even stated this outright, that lowering ASPs just didn't sell many more CPUs and certainly not enough to boost revenue. Like oil, a 10% decline in ASPs sells less than 1-2% more CPUs. Selling $10 CPUs just loses Intel $5-6 billion in revenue and $2-3 billion in cash each quarter. According to the calculations just $60 decrease in ASPs makes Intel lose money each quarter. AMD makes money at $90 ASP, Intel doesn't. |
