To: kpf who wrote (219841 ) 12/9/2006 10:23:33 PM From: economaniack Read Replies (2) | Respond to of 275872 kpf, sometimes I forget I am writing for a general audience and express things in terms familiar to economists but not layman. Efficient scale normally refers to the level at which minimum average cost has been achieved, and one generally assumes that can be scaled linearly by replicating the production process. Obviously there are fixed costs of design and process development which are amortized over the entire output of the firm and these are never entirely averaged out, so Intel has some scale advantage. On the other hand, at AMD's scale, (two fully functional megafabs) their production is large enough that manufacture costs (fabs, production and packaging) are most of the cost of a finished CPU, and the amortized development costs do not represent an insurmountable advantage for Intel. I certainly didn't mean that AMD had an actual advantage because of size and that Intel was too large, just that most scale economies were exhausted at AMD's current output, and Intel did not have substantially lower average costs on account of being 3 times larger. Similarly, by scaling linearly, I mean that AMD's costs of silicon (adjusted for SOI and other technical factors) are comparable to Intel's. Fab 36 pretty much exhausts the economies of scale for fabs, and Intel operating 3 fabs will (to a reasonable approximation) get 3 times the output that AMD gets from 1 rather than some higher factor which would reflect a higher efficient scale. Clearly given a certain capital base (fab capacity) marginal cost will be quite low if the fab is partially idle, and only rise when the fab is pushing its capacity limits, so in that sense it isn't linear at all, but comparing running one full 300mm megafab at 65nm, to 2 or 3 or so on, the output increase is linear to a first approximation. With software, in contrast, marginal cost of production is very low, so most of the cost at any output is still the fixed cost of development. That means that a dominant firm can have per unit costs that are small fraction (<20%) of an entrant with 10% market share. Obviously that is not the case with processors.