Fab-Less AMD?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

All sorts of rumours have been doing the rounds recently about AMD going fab-less following their announcement of an ‘asset light’ strategy for the company. After giving the matter some consideration, I’ve arrived at the conclusion that the rumours are wrong, and I thought I’d spend a moment to share with you the reasons why.

AMD has spent somewhere around $2.5 billion constructing and equipping Fab36, which has been in operation since 2006. AMD is also in the process of spending $2.5 billion updating its Dresden Fab30 to deal with 300mm wafers. Once it has been upgraded, the facility will be renamed Fab38. Companies shouldn’t really be in the habit of throwing good money after bad, but it is difficult to see AMD selling off such expensive assets such a short time after investing so heavily in them.

AMD is also in the early stages of planning a new $3.5 billion Fab in New York state. AMD has taken the unusual step of looking for a partner to help it finance the new facility. If AMD was so eager to flog-off its production facilities, surely it would be running away from the NY deal rather than looking for partners that can help it go ahead?

My suspicion is that AMD is looking to (eventually) bring ATI’s chip production in-house and is looking for an opportunity to cheaply increase potential production volumes in the meantime. Look at the problems ATI has had bringing the R600-based GPUs to market. Could some of those problems have been avoided if ATI had direct control over the manufacturing process rather than relying on a foundry?

AMD has had a long culture of fabbing its own chips. The quote oft attributed to AMD’s founder, Jerry Sanders – “Only real men have fabs”. Admittedly, Jerry isn’t in charge any more, but is Hector Ruiz really going to turn his back on AMD’s historical position and turn the company into an IP-only outfit? Nine years is a long time in the computing industry, but back in 1998 AMD was busy pushing the virtues of having fabs: “The trouble with foundries is that they want their 50 per cent profit margin” said AMD VP Dana Krelle. Has the reality of the market really changed so much that using a foundry to produce CPUs is viable today?

There is a great deal of benefit to be found in fabricating your own CPUs. It allows companies to tailor their chip design and the manufacturing process to complement each other. Look at the drop in temperatures and improvement in speed AMD was able to engineer with its Thoroughbred-B CPU. The addition of an additional metal layer in the process over the A core allowed them to add another 500 MHz to the chip’s top speed.

If they hadn’t been in charge of the process as well as the design, would they have been able to make such improvements? It also takes just one look at the great ‘success’ that Nexgen and Cyrix had outsourcing their chip production to realise that the best CPUs have always been made by companies that were in charge of the entire production chain, from design through to process.

Overall, I think the rumours surrounding AMD going fab-less have been started by people who have only looked at the short-term picture. AMD is bleeding money, and an easy way to cut those losses is to sell off a big chunk of the company.

But looking at the bigger picture, as AMD integrates ATI more effectively it will need access to more fab capacity. If that capacity comes from foundries, it will lower AMD’s profitability on those chips. Combine that with the strategic disadvantage of relying on foundries to adapt and perfect new and smaller processes as quickly as Intel does, and I think AMD would have to be mad to go fab-less at this time.

Liam Edgeley


Leave a Reply