I came across a very interesting article/blog written over the weekend by Jean-Louis Gassée on Intel and mobile. It covers some similar ground to several of my blog posts on the topic but also has some new facts. And it has additional credibility since Jean-Louis was head of product development and worldwide marketing at Apple (pre-iPhone).
He also refers to an interviewfrom last year that I hadn’t read with Paul Otellini, Intel’s then-CEO on his last day. Apparently Jobs gave Intel the chance to build the processor for the first (and so presumably subsequent) iPhones. Intel turned them down. Otellini said:We ended up not winning it or passing on it, depending on how you want to view it. And the world would have been a lot different if we’d done it. The thing you have to remember is that this was before the iPhone was introduced and no one knew what the iPhone would do. At the end of the day, there was a chip that they were interested in that they wanted to pay a certain price for and not a nickel more and that price was below our forecasted cost. I couldn’t see it. It wasn’t one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100x what anyone thought.
That was the day Intel failed in mobile. This would have been an ARM-based chip. But at the time Intel had an ARM-based division called Xscale (the old Digital StrongARM) who were desperate for business. Eventually they sold off the whole division to Marvell in 2006.
When I was at VLSI I worked for a guy called Cliff Roe who said “I wouldn’t want to be Intel’s next CEO [this was in the Andy Grove era]. He has to find a way to educate the company and Wall Street that they can’t go on making those kind of margins.” Well they still have those kind of margins (Intel’s gross margin is an astounding 65%). But it is also a genuine problem. If Intel had won Apple’s business they would now be building about 250M application processors per year at a cost of, say, $20-40 which is a $5-10B business, significant given Intel’s total business is around $60B. But it would be at a much lower margin since the ARM market is competitive and the x86 market is not (AMD is the only competitor and not a strong one). Whether Wall Street would treat this as good or bad is unclear. They love Intel’s margins.
Intel’s entire strategy has been that they are 3 years ahead of everyone else in manufacturing and process. Indeed they highlighted this at their recent analyst day. The chart is also a bit self-serving. Intel introduced FinFET at 22nm, compared to TSMC at 16nm. But TSMC introduced double patterning at 20nm compared to Intel at 14nm. I have no insight into whether Intel’s 14nm yield problems have anything to do with double-patterning learning ramp, but it is certainly the biggest different from 22nm.
But they need to ship $50 of negative revenue with each tablet processor because they are so competitive! And even in their main business a lot of Broadwell is being pushed out to the middle of next year. I think what actually happens is that Intel has a process lead but only for microprocessors where their price is so high that they don’t need the process to be truly competitive. Everything else has to run in old processes making them non-competitive too.
Herethey were in 2012:Intel dismisses ‘x86 tax’, sees no future for ARM or any of its competitors
How’s that going a couple of years later? ARM is still solidly entrenched in mobile and the rumors that Apple will switch their MacBook Air and maybe Pro to ARM are getting stronger.
Now Intel will roll mobile in with the non-server processors so the fact that they are currently losing $1B per quarter on it (perhaps $7B over the last two years) will be be hidden from view. Or maybe they are quietly giving up, who knows?
But the next big thing is Internet of Things (IoT) and that is a perfect market for Intel with lots of differentiation, because…nobody has a clue. And margins in IoT are likely to be even lower than mobile and ASPs lower still (these will not be mammoth chips in 10nm). I believe most IoT devices will be build on older nodes so whether Intel is 3 years ahead or not is irrelevant.