After the market close on Dec. 12, Bloomberg had the audacity to spread one of the sleaziest rumors that I've ever seen perpetuated about Intel (INTC) and Google (GOOG). The rumor is essentially that Google is looking to design its own chips and cut Intel out of the equation. Naturally, Intel shares are now down about 3% in the after-hours session on this pure fantasy. In this article, I would like to step through this rumor and show why it's probably complete nonsense.
The rumor itself is sketchy
Here's what Bloomberg had to "report" with respect to this,
Google Inc. is considering designing its own server processors using technology from ARM Holdings Plc, a move that could threaten Intel Corp.'s market dominance, said a person with knowledge of the matter.
Oh, really? Google is going to design chips based on ARM Holdings' (ARMH) IP? OK. So let's assume that Google indeed wants to take out an ARM processor or architectural license to design its own chips. This means that Google is going to need to be hiring some chip engineers en masse, right? After all, designing and validating a chip isn't cheap.
You'd think, then, that Google would have some job openings for chip design engineers right?
Well, let's search for jobs on Google - let's figure out if they're hiring chip engineers.
I searched for "SoC", and I got one job opening that had nothing to do with SoCs design. OK. Let's look for other common chip development type jobs. Well, after trying multiple searches including "CPU", "Validation", "SoC", "PHY", "CMOS", and many other searches, the only thing that I could find was the LONE listing that Bloomberg's "crack" team found:
See this job posting? The one that the folks at Bloomberg thought implied that Google was going to build its own chips? Well, Google may be building its own chips, but they sure aren't microprocessors/SoCs. An "ASIC" - which is short for "Application Specific Integrated Circuit" - is a chip that is designed to do a special task. In servers, networking hardware, and storage systems (found in the job listing) there are tons of ASICs that are used to do various functions completely unrelated to what the CPU/primary SoC is responsible. Just because Google may be designing a chip doesn't mean that it's designing anything that it expects to compete with Intel in its datacenters.
Oh, one more search. Let's search Google's job boards for "ARM" and see what we find,
Okay, so they're not really hiring all that many chip engineers. Or, really, any.
If Google Wants To Build Its Own Server Chips, It Will Probably Acquire A Chip Company
If Google were actually going to build its own server SoCs, then it would probably need to buy a pre-packaged, ready-made team. NVIDIA did it for its "Denver" CPU, Apple did it by buying PA-Semi and Intrinsity, Samsung lifted an entire team from AMD (AMD) about a year ago, and Broadcom (BRCM) did it by acquiring NetLogic.
So, if you see Google, say, buying Cavium Networks (CAVM), Applied Micro (AMCC), or another chip firm, then there would be some real credibility here. But it is naive to believe that Google can simply design a chip that offers significantly better total cost of ownership than what it can buy from Intel. I'll show you why in a minute.
Designing Chips Isn't Getting Any Cheaper
People seem to think that designing chips is cheap and easy because ARM licenses processor IP. This couldn't be further from the truth. Thanks to the increased complexity of these designs and the process nodes that they're built with, the costs to actually design and build chips is skyrocketing. I give the following slide from Broadcom's recent analyst day to illustrate this phenomenon:
In order for Google to design chips based on a leading edge process node (and they would need to do so in order to have a prayer of having anything close to performance/watt competitive with Intel), it will have to spend about $300M. Chump change for a company as profitable as Google, but as these nodes become more complex, those design costs are going to get even worse. Google possibly has the scale to amortize the design cost that goes into these chips given its scale, but keep in mind that Google uses many different chips from Intel (which, again, has the scale to develop them all profitably).
Will Google's Chip Be Any Good? Who Will Build It?
Two questions that Google will need to ask itself if it wants to try to build its own chip(s):
- Can it design a chip with better performance/power characteristics than what Intel's own processors can provide?
- Who's gonna build it?
Well, right off the bat, I would very seriously doubt that Google could - in any reasonable timeframe - build up a team with the ability to put together a meaningfully superior SoC for datacenter applications. But, even if it could - either via magical pixie dust or via acquisition of a credible processor vendor like Cavium - would such a chip be competitive with what Intel will be able to do?
Probably not. On a performance per watt basis, Intel's low power "Avoton" easily surpasses the offerings from Calxeda or Applied Micro, and Cavium's "Project Thunder" is still deep in development with no ETA given (and by the time it arrives it will probably be built on TSMC's 28nm process and will be competing with Intel's 14nm chips).
Does Google really want to trust its datacenter fate to a custom designed chip by a team far less experienced than what Intel brings to the table and, at the same time, with a process technology (which implies performance/power) disadvantage? Especially when Intel has made it quite clear that it will do custom chips for people?
I propose a more likely scenario
There is no reason for Google to do its own ground-up custom chip, but I can certainly see the case for Google wanting to, say, design a custom IP block (say an accelerator) that doesn't come with Intel's suite of processors. Intel has made it clear on numerous occasions that it would be happy to work with such a customer to design a custom SoC that integrates whatever special IP blocks that the customer wants. This is why Intel is building SoCs based around both its "Core" processors as well as its "Atom" processors so that it can be flexible to the demands of its customers.
This is the whole basis of Intel's "semi-custom" foundry initiative:
One More Thing
A final proof point to show how poorly thought out the Bloomberg piece was, the report cites Google joining the POWER Alliance (POWER is not ARM) as a reason that Google would want to design its own custom ARM chips. Huh? Wouldn't this be a sign that Google could potentially want to build chips around the POWER ISA and/or architectures from IBM?
Highly irresponsible journalism.
Intel's stock price just lost 3% in the after-hours session following this "rumor". I do not believe there is much truth to it (for the reasons outlined above), but even so, the rumor itself was so vaguely worded (i.e. to get page-views by shocking people but to cover Bloomberg's collective behinds with all of the wishy-washy disclaimers). I get that it's cool to pump ARM and bash Intel these days, but this story just smells too fishy and is too closely timed with today's upgrade from FBR of ARMH on "microserver potential" to really pass the smell test.