Semiconductors Don't Count on a Turnaround at AMD By Bob Faulkner Street.com Contributor 5/15/2008 9:42 AM EDT
Since the middle of April, Advanced Micro Devices (AMD) has seen its stock run up about 20%, driven in large part by the anticipation of a new manufacturing strategy, which the company refers to as "asset smart." Rumors suggest that as part of that strategy, AMD will spin off its fab operations as a separate entity or sell them to an existing foundry partner. The fantasy on the part of AMD investors is that such a transaction would strengthen AMD'sbalance sheet, reduce its need to raise cash for capital expenditures and enhance its competitive positioning vis-a-vis its rival Intel (INTC) . This strategic change has been in the works for the better part of a year, and many had anticipated that the details would have been revealed at the company's annual meeting last week. However, CEO Hector Ruiz remained mum on the issue, suggesting that the announcement will come when it's ready. But the real question is, will any changes along those lines do much for beleaguered AMD? While I'm certain you'd see the stock pop if the company announced the full or partial sale of its wafer fabs, in the long run it's really not going to help much. Too Far BehindWhile it's true that AMD does not have the cash to chase Moore's Law on a head-to-head basis with Intel, selling its fabs or migrating to some form of asset-light strategy isn't going to change its competitive position. AMD's fundamental problem is that it is, at best, a full year behind Intel on process technology. Intel commenced shipping 45 nanometer parts in the fourth quarter of 2007, and AMD will not start doing so until the fourth quarter of 2008, if all goes according to plan. If 100% of AMD's production capacity came from foundries today, this problem would not change. To get a better understanding of the competitive disadvantage I created a hypothetical example in the table below. The numbers are fictitious but close to a realistic example. More importantly, it's their relationship from one process to the next that is critical. Let's start with your basic 300mm wafer fab that includes about $2.4 billion in equipment (we'll skip the bricks-and-mortar expenses for the sake of simplicity) that processes about 45,000 wafers per quarter at the 65 nm node. Most wafer fab equipment is depreciated over about four years, so we come in with depreciation expense each quarter of about $150 million, or roughly $3,333 per wafer. Now, throw in another $2,000 per wafer for the variable expenses such as labor, materials and utilities. If our fictitious microprocessor die has an approximate area of 280 square millimeters, we should be able to get about 252 of these little things on this big old silicon wafer. Nothing is perfect in this world, least of all manufacturing semiconductors, so every possible die is not going to function as expected. However, since this is a fairly mature process (i.e., we've been doing it for a while), our experience tells us that if we adhere to the recipe religiously, nine out of 10 of the die on that wafer should work just fine. From there, the math is quite simple. The expected "good die per wafer" have to bear the costs of processing the entire wafer. Each one of those die are going to be sold for $50 to Dell (DELL) or Hewlett-Packard (HPQ) or whomever. At that price, we have a nice 53% gross margin. Not bad, huh? Now, let's say my arch enemy is making the same kind of microprocessor, and I know that some time over the next year he or she will move to the same 65nm process node (i.e., lower cost structure) that I am now using. Rather than wait for that to happen, I elect to push my manufacturing capability down to the next level, in this case 45nm; that is, I'm going to be drawing the circuits and the transistors with a much finer "pen." To do this, I have to add some equipment that generally runs about 35% more than what I had been using at the prior node. That bumps up my depreciation rate, but since I'm still producing 45,000 wafers each quarter, my variable costs don't really change. However, because I'm going to draw my circuits in a finer geometry, each die is going to be smaller. Exactly how much smaller will depend upon whether or not I decide to add features. To keep this simple, we're just shrinking the identical part we had before, because it's really the be-all and end-all of the microprocessor world. Essentially, its length and width will each be about 30% smaller then the 65nm version, so our die area is about 134 square millimeters. With that much smaller die, I can get a lot more on each wafer, but there's also a downside. My yields (good die per wafer) will probably decline somewhat, because I'll have to work the kinks out of this new recipe. Despite the lower yield, the bottom line is that if I still sell them for $50 each, my gross margin has jumped to 67%. Now, I might want to keep my price there and really irritate my customers, or maybe I'll reduce my price by $5 or $10 per unit, thereby making my customers really happy and still make more money than my competitor. Remember, this is just a crude example to make a point. In reality, there is much more that goes into the calculations, but the basic principles are the same. Cost is a function of silicon area, and if I am manufacturing at a larger process geometry then my competitor, I am at cost disadvantage. That is the crux of AMD's problem. Capex Isn't the IssueSimply removing the capital expenditures from the equation is not going to change AMD's competitive or financial position. Why? Most fabless semiconductor companies buy "good die" from their foundry, and the price of that good die includes all of the aforementioned costs. That way, their gross margins are fairly predictable and consistent. The buyer neither bears the risk nor gains the benefit of yield swings that he or she has no control over. That's where the foundry gains leverage by betting it can generate higher yields than those on which the pricing was based. If the foundry gambles incorrectly, the customer doesn't take the hit. Another aspect of the problem from AMD's perspective is that their foundries are no better positioned to deliver 45nm wafers than is AMD. Each is looking at late 2008 or later. More importantly for AMD, by taking this approach it would cede its destiny to the foundries from a timing perspective. Foundries bring up a new process when they're certain they'll have sufficient customer demand to support it. That timing may or may not be appropriate for any one customer. For all the talk about leading edge processes, Taiwan Semiconductor Manufacturing (TSM) , the world's largest foundry, has only 15% of its revenue coming from its 65nm process. At Chartered Semiconductor (AMD's microprocessor foundry), it's only 7%. While Intel doesn't disclose specifics by process node, its entire microprocessors production was at 65nm in more than a year ago. On the last AMD conference call, CEO Hector Ruiz responded to a questioner that he didn't believe that ownership of production capacity was a requirement for technology leadership. If that's the case, why are AMD's fundamentals in such a sorry state? If that's the case, then why has the CPU business generated nearly $6 billion in revenue over the last five quarters yet managed to lose $830 million at the operating profit line during that same period? AMD is structurally disadvantaged, and nothing that the company has talked about will change that. [Die Frage ist, ob dies im Aktienkurs schon hinreichend eingepreist ist. Die noch wichtigere Frage ist, ob AMD mit dem Kostennachteil, ein Jahr hinter Intel zurückzuliegen (d.h. aktuell 65 nm bei AMD und 45 nm bei Intel) überhaupt langfristig überleben kann. Ginge AMD pleite, ist natürlich gar nichts im jetzigen Kurs eingepreist. Die Aktie bleibt riskant. - A.L.] Advanced Micro Devices is one of the many technology and telecommunications stocks that Bob Faulkner monitors and discusses in The Telecom Connection.
|