Does AMD want out of the chip making game?
Choice of next fab more about AMD's role than location
Analysis So where has AMD chosen to build its next-generation wafer fab? The company says it has a preferred location in mind, but won't - for obvious reasons - say exactly where it has chosen.
This week, Reuters ran a story claiming AMD had put its Dresden facility on the shortlist - hardly a staggering conclusion to leap to since that's where the company already makes its processors.
A spokesman for AMD in Germany told the news agency that the chipmaker was holding talks about a new plant, but declined to give details.
"AMD has a preferred site, but we have not finalised the negotiations," the spokesman said. "We'll be able to say more later this quarter."
So Dresden it is, then? Not necessarily. German business newspaper Handelsblatt reported this week that East Fishkill, New York State is also in the running.
Almost certainly that location means somewhere on IBM's East Fishkill facility, where Big Blue has been rumoured to be building a fab for AMD for some time. What's not known is if the plant - if it is indeed destined for AMD's use - is to be a working fab, or simply support for the two companies' joint R&D programme. That development effort is currently centred on IBM's Advanced Research Centre in the same area. Its goal is to devise 65nm and 45nm processes on 300mm wafers.
AMD's new plant, wherever it is located, will be a 300mm wafer fab. If AMD starts construction next year, the plant should be ready in 2006, tantalisingly close to the 2007 timeframe anticipated for the shift to 65nm.
The company's decision ultimately boils down to cost. What financial incentives will Dresden offer AMD to build its next-generation fab there? Is it cheaper to partner with IBM - perhaps even to bankroll IBM's fab construction costs with a foundry business commitment?
Building a new fab is a multi-billion dollar investment. Does AMD really want to spend this much? We argue that it may be better off working with a company with which it already has very close ties and a business model that supports the construction of ever more expensive fabs. That company is IBM, which numbers Apple and Nvidia - and soon Microsoft - among its 130nm, 300mm wafer customers.
IBM wants to build up its foundry business, and its desire to add more big name customers to its books might enable AMD to negotiate a very preferential rate. Potentially, IBM can make future Athlons and Opterons more cheaply that AMD can on its own.
AMD's finances are precarious - it continues to lose money - and its market share is around 16 per cent of the x86 market. As a global semiconductor company, it doesn't even make the top ten. It has just over $1 billion in the bank.
Even if AMD doesn't care to go the whole hog and completely outsource future chip production - as a long-time Intel rival it may not want to appear to 'lose face' by bowing out of manufacturing, or it could fear losing too much control over the manufacturing process - a joint construction/foundry deal with IBM makes sense, too. And it follow logically from the two firms' joint R&D efforts.
AMD has a potential partner in Dresden. It already runs an Advanced Mask Technology Center with Infineon, and may yet persuade the memory maker to partner on a new fab. IBM and Infineon also have ties over the development of 65nm fab techniques, so the plot thickens.
We have no special insight into AMD's negotiations or thinking, but we believe an IBM foundry deal makes the most sense. We'll have to wait until "later this quarter" - as the AMD Germany spokesman put it - to find out if we were right. ®
AMD hiring staff for IBM East Fishkill plant
IBM rumoured to be building 65nm fab for AMD
AMD, IBM pool chip making development
Microsoft picks IBM as Xbox 2 processor partner
IBM to partner with Infineon on 65, 45nm tech