Monday’s post on Intel’s work to win Apple business generated a lot of traffic and Tweets. One follower on Twitter keeps coming at me with good questions. So Frog Man (@accretivegrowth), whoever you are, this one is for you.
The main thrust of the questions I have been getting is what should Intel’s strategy be. Not just for mobile, but for their entire business.
To be clear, Intel is still incredibly strong. As much as their PC business is slowing, their server business is doing incredibly well. While there is a lot of interesting work taking place out there on ARM servers, the software ecosystem built on top of Intel’s x86 chips is 30 years deep. Maybe ARM servers will find their place, but that is not going to be determined for many years. In the interim, Intel’s most profitable chips are in much demand, essentially powering all the data centers and cloud apps being built out there. Intel has the dominant share in the fastest growing segment of computing. That will fund years of investment in other categories, and by investment I mean losing money on R&D until volumes kick in.
Nonetheless, Intel has to take a multi-year approach to their business, which is what happens when you build $5 billion factories. And if you look out over five years, it is clear that Intel needs to do something.
So far, one of their biggest bets has been on mobile. That has cost them a lot of money, but most of those losses come from marketing subsidies they have been using to seed their products in the mobile and tablet markets. Only secondary has been the cost of the “1,000 engineers working on products for the iPhone”. As difficult as this process has been, I think there is a reasonable argument to say that Intel is right to make a bet on mobile. It is a big market, that is still growing (at least for now), and threatens one of their core markets.
I also think it is important to understand some of the basic economics of the semiconductor business. It turns out, that semiconductor costs and semiconductor prices are not really connected. Costs are determined as a function of geometry. The main cost of a semiconductor is the fabrication of a ‘wafer’ of chips. These are big silicon discs a little bit bigger than vinyl records. The more chips you can fit on those discs, the lower the cost of each chip. By contrast, prices are set by the market. So in the data center market, where Intel is practically the only supplier, their chips start around $300 and go into the four digits. By contrast, a leading mobile baseband from Qualcomm probably costs around $40, with low-end products in China priced sub $10. This is the trick to semis. The mobile chip and the data center chip cost roughly the same amount to manufacture, as you can fit roughly the same number on a wafer, but the prices are radically different. (Note for the experts: I use the word ‘roughly’ the same size, I recognize there is considerable variation here.)
This leads into the question are Intel chips more expensive than other mobile products. The argument I heard on Twitter was that Intel has struggled in mobile because their products have to include a mark-up to cover all the costs of Intel’s internal fabs. I think this is wrong. Regardless of who manufacturers a chip, the price of fabrication has to cover someone’s cost of building that fab. Whether these are allocated internally at Intel, or bundled into the cost of TSMC’s foundry, those costs are there.
Instead, Intel has struggled in mobile because their CPU processors are seen as being too power hungry and because their 4G LTE chips are really only launching now. I do not think they are losing sockets because of price, except in super price-sensitive markets like China. And to be precise, if they really do win an iPhone socket next year, then they are gaining share.
However, there is a separate concern that Intel is fighting hard to get into a market where prices are much lower than they currently enjoy for other products. And there is no easy remedy for this. At this point, they do not offer any features or speeds that could justify a price premium to other products on the market. But this goes back to the question of whether Intel should be in mobile or not. And there, I think the answer is that they have to make an attempt.
There is also the question of whether Intel should split up into two companies – one selling their processors the other selling foundry services. As I mention above, I do not think the allocated foundry costs really hamper Intel’s competitiveness. To the contrary, the one key advantage they have is that their fab processes are better than everyone else. It is clear that Moore’s Law is slowing down as everyone takes longer to get to the next manufacturing node. This matters because the smaller the chip the more can be fit on a wafer and thus the lower the unit costs. If Intel is the first to shrink, that means they will have a longer period at each future node where they would have a meaningful advantage over the competition. This topic is very debatable, but it seems to me that having an internal fab has the potential to be an advantage at this stage of the cycle.
There is now a big push on Wall Street to split things up. The many “activist investors” out there have deployed this strategy successfully at many companies. So there is a tendency to stick with that pattern. It is also something of Corporate Finance 101 to argue for simpler, stand-alone companies that focus on one thing. I understand, and generally agree with, that theory, but it should not be a hard and fast rule. There are some cases when splitting up a company does not work. I think the idea of splitting Intel or Qualcomm falls in the latter category. Both companies generate tremendous synergies from their two sides – Intel’s design and fabs, and Qualcomm’s design and licensing businesses.
The big question hovering around Intel is whether they should sell their foundry services to outsiders, that is, should they manufacture other companies’ designs in their fabs. The prime directive of owning a fab is that you keep the fab full. It costs $5 billion to build a plant that lasts four years, which means you start each year with the business in a cost-hole of $1.25 billion. Intel has struggled with this in recent years amidst declining PC units. I drove by their fab in Arizona recently and it certainly looks pretty empty from the outside. So many argue that Intel should take outside orders. Again, this is an idea that looks good on paper, but executing it will prove very hard. They would have to build a sales organization and more importantly, a support organization. That is a massive shift in culture for Intel, and also something that cannot be done overnight.
Which oddly enough leads back to Apple. I mentioned in my last post that one of the incentives Intel could offer to Apple is the ability to use Intel’s fab to manufacture Apple’s A and M series chips for the iPhone. If Intel could pull that off, they would fill a fab with a customer that does not need as much support as most other customers. (The only other customer that would be as ‘easy’ to serve would be Qualcomm, an idea I half-jokingly suggested back in March.) However, the flip side of this is that Apple is now already the biggest customer for Intel’s PC business. Everyone seems to think that someday Apple will stop using Intel’s chips in their Macs. I am not sure that will happen, but if that came to pass it would be a big blow to Intel, and as I seem to be saying a lot lately, this is just the state of the electronics business today.
And let me repeat, this is just my speculation. I have no inside knowledge here.