Being the time of year for all things dying and mysteriously coming back to life I pondered if this week to rant about the likes of Apple’s now very well documented near death in the late 90s, or the possibly more impressive though now looking quite temporary IBM renaissance that Lou Gerstner executed. Neither seemed overly interesting, so instead I thought about my longstanding thought experiment which is if you were given the job of being the CEO of Intel tomorrow, what exactly would you do?
This may end up looking like a hatchet job, however, let’s preface it by saying I have no skin in the game, and nor do I consider ARM or anyone else a plucky upstart pretender to the crown. ARM are just the company that looks in the medium term like Intel’s largest headache, but their true problem is systematic: they’re one company against a whole ecosystem.
If you focus on the x86 based PC market, which let’s face it is kind of important for enabling the world to tick over, the notion that Intel are heading straight for trouble might seem ridiculous, but it really isn’t, and ironically they were instrumental in proving this with the remarkably smooth PowerPC to Intel transition Apple effected in 2006. This proved that software translation, so that code written for one sort of processor can run on another, had reached a point of practicality that very few people had anticipated. Therefore given the right push either Microsoft or Apple could be tempted to do this again, only to move away from Intel on to something else. The reason they’re not is right now that something else is theoretical, but everyone can see it coming.
The most obvious statement of intent comes from Apple, and that is the 64 bit A7 in the iPhone 5S which is noteworthy for being the first 64 bit ARM based processor in mass production. While the extra processing capacity is useful for phones it’s hardly a game changer, and these days you’ll get far more improvement with faster graphics processors or faster memory (see the fuss around the PlayStation 4 having 8GB of GDDR5) for typical mobile usage. Apple also designed this chip largely themselves, so this is quite a deliberate activity, and not something they’re doing for fun. But if it’s kind of moving into a void above mobile but not quite there for laptops, what is it for?
The educated guess here is it’s a stepping stone to replacing or even co-existing with x86 based chips in low end Mac laptops. The reason for doing this would be you could push for much longer battery life with the same batteries, or just use smaller batteries and make the laptops smaller. As a bonus the heat produced would be much lower enabling fanless designs.
So in the laptop market there is pressure to improve “performance per watt” but the real use of this is in the server space, where energy costs are such a high proportion of total running cost that it is a real priority when purchasing hardware. Intel have made massive strides at improving their performance here in the last decade, however, they’ve now reached a sort of nasty void where the really large customers, with thousands of computers in a single location, are running workloads where the processor spends a large proportion of the time waiting for information from the other computers, and so it’s not simply a question of how much power you use while you’re adding two numbers together, but also how much you use when doing nothing, and this adaptability to power usage is exactly what the likes of ARM have been attacking in the mobile market for getting on for 20 years. Intel’s chips perform reasonably well when fully saturated, but when they’re not they are a lot less competitive.
Intel had seen the mobile threat a while ago, and had they succeeded in tackling it then ARM wouldn’t be in a position to move up to threaten them in other areas, but it has at this point basically failed to make any dent in that business. Their results this week demonstrate what little part of mobile they were in is actually shrinking in revenue quite dramatically year-on-year: 61%. We could debate the merits of the x86 architecture (or lack thereof) for mobile until the cows come home, and while I happen to think it’s unsuitable for that purpose the real problem Intel have is that they’re inflexible.
When you want an ARM processor, ARM have merely sold the design blueprint to a chip manufacturer to actually make it, except there is often another integrator in there that took the barebones processor design and added all sorts of peripherals, and got that made instead. This results in what is known as a System On a Chip. In the mobile market this means cramming as much of the stuff your phone needs to do on to a single piece of silicon as possible so that the manufacturing of everything else in it becomes much simpler, and thus cheaper. In servers you could do the same thing, but for networking or high performance I/O and security systems.
Intel don’t really get this, and are stuck in the PC model of the 80s and 90s where most integration took place beyond the silicon. They stick to this line not because they’re blind to it, but that a significant proportion of their advantage in the modern market is their admittedly superb fabrication technology. They’re simply about 18-24 months ahead of anyone else when it comes to how to physically make chips. This means they can make clunky designs but get decent performance out of them, and charge a higher price because people want backwards compatibility and the top end performance. As a consequence Intel have much higher margins than any of the other chip makers, and they plough this straight back into fabrication R&D to maintain this edge.
The problem here is those other people keep getting slightly closer every year. In the chip making world you have Intel who design and make their chips, and then you have everyone else, where designers and suppliers are separated. Nvidia, for example, might contract TSMC to make their newest GPU, but if Samsung came out with a better manufacturing process maybe they’d use them instead. As a consequence there is a full competitive market of chip manufacturing occurring outside of Intel driving progress independently.
But where Intel really tie themselves in knots: they’re stuck with x86. It’s now too late to create an alternative for the world to transition to (see things Intel should have been doing 15 years ago) – ARM are big enough that everyone just wants Intel fabrication technology to be used to start making ARM compatible chips. Intel know such a move would eventually be the death of the x86 business, and thus those high margins which are in turn used to fund their R&D which keeps them at the front of the pack, and so they steadfastly stick to the current path, to the point of leaving a brand new factory in Chandler, Arizona completely empty.
This leaves Intel in a nasty spot. If they start building non-x86 chips the performance gap will close and the prices won’t be high enough to be worth it. While they persist in not doing so the rest of the industry gets closer to attacking them outright every year. As a result I think within five years Intel will end up “doing an AMD” and splitting the design and manufacturing groups apart. More precisely, I think they’ll split the x86 business and the manufacturing apart.
In manufacturing they have clear advantages, with leading technology that others would love to build with, and a large US presence which could be emphasised for markets where supply chain integrity is a growing concern. (i.e. do the chips you’re getting made contain just what you specified, and there aren’t any sneaky bonus parts phoning home somewhere?)
For the x86 business things are different, but given the freedom to pick other manufacturing partners they could design all manner of lower cost x86 devices suitable for the pending Internet of Things, which really doesn’t need everything to be made on the latest and greatest silicon. Given the right leadership they may even be able to create a credible threat to ARM in the processor IP market, but a willingness to break from the past is going to be needed, and they will need to start sooner rather than later.
The problem is welding the design and manufacturing businesses together is now holding them both back. If they continue to wait for ARM to invade much more of their territory it will be game over and they won’t be able to make money in either design or manufacturing, with the likely endgame there being acquisitions of IP by interested parties, such as, for example, Apple. So don’t envy the Intel CEO. While they will have to deny it until it happens, their job has become picking the best time to make this decision to split to get the most value out of it for the shareholders. They have simply run out of other options.
Nigel Birkenshaw runs atomirex. He can be reached at firstname.lastname@example.org