Intel's fab flexing may pay off big amid capacity shortages

There's a semiconductor manufacturing capacity problem and Intel appears to be one of the few players that can control its own destiny.

For years, Intel has preached that its manufacturing prowess---it is one of the few chip makers that has its own fabs---is a competitor differentiator. Indeed, Intel and its tick-tock cadence produces smaller more powerful chips at a steady rate.

Intel's bet on manufacturing appears to be paying off now. Why? A bevy of competitors are wrestling with their contract manufacturers to increase capacity.

For instance, Nvidia can't get enough 28-nanometer Kepler GPUs to meet demand. Nvidia and its fabrication partner TSMC failed to plan correctly.

On a conference call with analysts, Chris Evenden, director of investor relations, was asked repeatedly about Nvidia's ability to meet GPU demand.

Evenden's reply:

There's just not enough capacity and the planning process when we think about these new nodes just simply needs to be better and I don't think there's any way around that. In combination between us and TSMC, we under-planned for the supply of 28-nanometer and we need to fix that in the future.

Sound familiar? It should. Nvidia sounds a lot like Qualcomm.

Paul Jacobs, CEO of Qualcomm, said in April that the company was also faced with a 28-nanometer chip capacity problem.

We are seeing very strong demand for our industry-leading MSM8960 and other 28 nanometer products. Although the manufacturing yields are progressing per expectation, there is a shortage of 28 nanometer capacity, and at this stage we cannot secure enough supply to meet the increasing demand we are experiencing. We are working closely with our partners to bring additional capacity online; however, the constraints on 28 nanometer supply are limiting our potential revenue upside this fiscal year.

The common thread: These technology companies in most cases only have a few manufacturers to run to---TSMC, Samsung and Globalfoundries, which is AMD's old manufacturing arm.

Complicating the capacity issue even more is Apple. Matt Cleary, an analyst with Primasia, recently outlined Apple's foundry options. To date, Apple has largely relied on Samsung to make its semiconductors. However, Apple is looking to diversify away from Samsung, said Cleary. Samsung and Apple have repeatedly dueled in court.

The problem for Apple is that it only has so many partners to work with. Intel, TSMC and Samsung are the only ones capable of making Apple's next-gen chips in volume. According to Cleary, Apple would have to rewrite some of its iOS to use Intel's manufacturing process since the chip giant is a no-ARM zone.

That reality means that TSMC is the likely Apple partner going forward. TSMC will then have to build enough capacity to accommodate Qualcomm, Nvidia and Apple. Guess who wins in that pecking order?

Cleary said in a research note:

We believe that only Intel, TSMC and Samsung have a viable shot at being selected as Apple's next foundry partners. UMC clearly lacks the scale and leading-edge process level in order to compete in this league, while GlobalFoundries probably still has to prove itself following the public divorce from AMD. TSMC is pulling out all the stops in order to win Apple's business, including hiring teams of IC design and system level engineers, accelerating its 20nm process development....

It seems most likely that TSMC will have the inside track to become Apple's next foundry partner—once TSMC gets 20nm up and running. The broader question will be whether Apple will prove lucrative enough as a customer to justify all of these efforts.

Add it up and you may have the bulk of the mobile industry moving through one manufacturing partner---TSMC---in 2014.

Intel's advantage here---even though it is a bit player in mobile today---is that it controls its own manufacturing and therefore its destiny. Architecture squabbles with ARM aside, Intel could gain traction in mobile simply because it can deliver the manufacturing goods.

The chip giant did a lot of talking about its manufacturing cadence during its analyst meeting this week. Evercore Partners analyst Patrick Wang said that Intel was "fab flexing."

Indeed, Intel CEO Paul Otellini spent a lot of time talking about manufacturing. Otellini argued that the semiconductor industry was at an inflection point. The "fabless" model---companies that design chips and then outsource manufacturing was faltering. Otellini's main point: Design and manufacturing should be one.

Speaking at Intel's analyst meeting May 10, Otellini said:

When design and manufacturing are deeply integrated you get a spiral of advantage. We can decide where we want to optimize (to lower costs). If you're a foundry and fabless company there's an economic tension there.

You've seen a consistent cost per transistor improvement from Intel. This curve from the foundries is difficult as invention gets harder. My contention is we'll increase our capabilities and advantage going forward.

Otellini said a new state-of-the-art fab will run companies $10 billion. That investment will raise costs for foundries and prices for fabless chip makers.

Not all analysts were completely sold. Piper Jaffray analyst Auguste Gus Richard said:

Intel highlighted its manufacturing prowess as its key differentiator at its analyst meeting. Management spent time emphasizing its fab capabilities in an era where it believes the foundry model is faltering. However, we believe that the law of diminishing returns is beginning to overtake Moore’s Law. Moreover, Intel is expected to shoulder more of the development costs as one of the few remaining practitioners.

Richard makes a solid point on Intel's development and capital costs. However, if the foundry model does falter, it is very easy to pick the winner---Intel. Perhaps, Intel's manufacturing spend becomes an issue, but given the shortages faced by Nvidia and Qualcomm it looks like the chip giant's bet could work out nicely.