Intel Corp (INTC)

[walbert]

I don't see anyone else promoting your TSMC's-sky-is-falling thesis.

Really?

Intel’s 14nm milkshake: It’s better than yours

http://www.extremetech.com/computing/171477-intels-14nm-milkshake-its-better-than-yours/2

Excerpts:

"The Titan rises

If Intel delivers on this roadmap, one thing is absolutely certain — it’s going to have no problem maintaining high margins. If Intel chips on 14nm are nearly half the size of what TSMC can offer, then Intel’s nascent foundry business will be able to command very preferential prices. It won’t put TSMC or GlobalFoundries out of business, but it will attract those customers willing to pay top dollar for the latest and greatest in foundry capability.

I don’t see either of the other two foundries catching much of a break here. By every reasonable metric, Intel leads the foundry world. If it can continue pushing smaller nodes and providing greater benefits than its competitors, than the companies that want to compete with Intel will have to build cheaper, hotter, slower products to do so. This inevitably impacts selling price, which impacts income, which reduces the amount of money companies can afford to pay to their foundry partners.

The long term trend in the foundry business, meanwhile, is towards merciless consolidation. Ten years ago, nearly twenty firms still maintained cutting-edge fabs built on 130nm. Today, we’re down to four — Intel, TSMC, Samsung, and GlobalFoundries. STMicroelectronics’ 20nm work is done in partnership with GlobalFoundries. Intel doesn’t have to convince everyone to buy x86 parts — if it captures enough of the high-end ARM business at some future date to deprive TSMC of the revenue it needs to push forward, it eventually wins by default."

"And this is where the foundry gap we’ve talked about before seriously springs into play. Intel is actually a bit behind TSMC at the moment, but steep scaling from 22nm to 14nm for Intel gives it a huge edge by the next node, and a deeper advance by 10nm."

"If you care about transistor density, Intel does expect to still be leading the pack — and by a greater margin than it is today."

"We’ve talked about the problematic cost curve at 20nm before: Morris Chang has previously stated that he views 16nm FinFET and 20nm as basically the same node (which tends to confirm what Intel says here), and GlobalFoundries is planning a 14nm-XM process that connects a 14nm front-end to a 20nm back-end, offering power scaling and FinFETs — but not a die size reduction."