AMD & ATI: The Acquisition from all Points of View

Our Thoughts: Will AMD manufacture ATI GPUs? 

One of the most intriguing aspects of this proposed acquisition is the potential for ATI to cease to be a fabless GPU manufacturer.  In the past, fabless companies like ATI and NVIDIA have had to rely on foundries like TSMC or UMC in order to manufacture their chip designs.  By relying on a 3rd party to manufacture all of their chips, ATI and NVIDIA avoid having to invest in multi-billion dollar fabs that require continued multi-billion dollar investments to stay up to date.  As a point of reference, AMD's Fab 36 cost $2.5 billion dollars and that's before even fully being converted to 65nm. 

The upside of being a fabless manufacturer is obviously that you leave manufacturing to those who are good at it; the downside is that you lose manufacturing as a competitive advantage.  The other downside is that foundries like TSMC aren't as quick to transition to new process technologies as companies like AMD and Intel.  While Intel has been shipping 65nm product for quite a while now, ATI and NVIDIA GPUs are still being built using 90nm transistors.  If the buyout does go through, ATI could potentially gain a manufacturing advantage over NVIDIA since AMD has already invested in its own fab plants. 

On the surface, the potential for ATI to have access to its own fab is tremendous; however, there are a number of limitations to success that are worth talking about.  For starters, AMD isn't Intel, and AMD's manufacturing processes have always lagged behind Intel's, which is why AMD is still currently shipping 90nm CPUs while Intel is in the middle of its 90nm to 65nm transition.  So the advantage that ATI would gain from having access to its own fab would not be as significant as if Intel were the buyer.  AMD should still be slightly ahead of TSMC and UMC, but Intel is almost one year ahead of AMD right now on the process transitions.

The other problem is that AMD is significantly capacity constrained as is.  AMD currently has two fabs of its own, Fab 30 and Fab 36, both in Dresden currently producing 90nm CPUs.  Fab 30 is set up for production on 200mm wafers while Fab 36 uses 300mm.  AMD just recently started shipping revenue generating parts through Chartered Semiconductor, a 3rd party fab that is manufacturing Athlon 64 CPUs for AMD in order to help relieve some of its capacity constraints.  So with Fab 30, Fab 36 and Chartered all working to just meet demand for AMD's CPUs, it's not like AMD has a lot of extra capacity to use to manufacture ATI GPUs. 

AMD's new fab in New York State will help deal with some of the capacity issues, but construction won't even begin until July 2007 at the earliest and we're looking at another 3 - 5 years before it's operational.  Without tons of excess capacity, it doesn't seem like ATI will be able to benefit from AMD's manufacturing facilities in the production of GPUs.  Chipsets are a different story as it will give AMD something to do with older fabs, as chipsets are no where near the size of modern day GPUs and are already produced on an n-1 manufacturing process (e.g. Intel's Core 2 processors are 65nm while Intel's P965 chipset is built on a 90nm process).  GPUs could also be manufactured at older fab plants as newer ones are upgraded, though the sheer size of modern day GPUs makes this an uncertain option.  More than likely, the only GPUs  manufactured at AMD's plants would be those integrated into chipsets or found on-die/on-package with AMD CPUs, meaning that they'd be very low end, low margin parts. 

AMD has already announced that at least for the next 1 - 2 years, there will be no changes in manufacturing with AMD or ATI, meaning that ATI will continue to produce GPUs and chipsets at TSMC and UMC, while AMD will continue to produce CPUs at Fab 30, Fab 36 and Chartered.  After that period of time, we'd venture a guess that AMD would start bringing some manufacturing in house (e.g. chipsets) or start producing CPUs with integrated graphics (either on-die or on-package). 

What AMD is doing with the proposed ATI acquisition is taking one step towards becoming more like Intel.  Intel currently has four 200mm 130nm fabs that are producing chipsets (anything older than Intel's 965 chipset); those fabs would have been useless for CPU production so the fact that Intel can get some additional life out of them by using them for chipset production helps amortize their high construction costs.  When the chipsets are ready to transition from 130nm to 90nm, these fabs can then be upgraded to 65nm to get ready for the next wave of chipsets they will be producing. 

While ATI will give AMD something to do with older fabs, there's also the argument that AMD could have become a chipset manufacturer on its own without having to pay $5.4B for ATI.  AMD has manufactured chipsets in the past, and one would think that it would be cheaper to hire engineers and construct your own chipset team than it would be to purchase ATI.  Obviously there's additional value that ATI brings to the table above and beyond the chipset, so it may just be that the sum of all of ATI's advantages are what make the acquisition sensible to AMD.