I quite eagerly awaited the announcement from Advanced Micro Devices (AMD) regarding its "ambidextrous strategy." After the firm's disastrous Q3 earnings report coupled with tepid guidance, investors were eager to understand just what lies ahead for the company.
At the earnings report, CEO Rory Read made it clear that the company was downsizing in order to be able to be cash flow positive at a lower-than-traditional revenue rate. This, of course, is necessary as the firm's cash reserve of $1.48B will not last indefinitely against a barrage of $100M - $200M per quarter net losses. The company further continued to stress that going forward, its focus would be on developing solutions that were made up of reusable IP.
On the 29th, AMD finally made the first "big" announcement regarding its "ambidextrous" methodology - that it would be using CPU cores from ARM Holdings (ARMH) in its own system-on-chip to be deployed in micro-servers that utilize the proprietary interconnect fabric gained from its SeaMicro acquisition.
What Does This Mean?
The last paragraph was a mouthful, so the question on the minds of most investors, then, is "what does this mean?"
Well, this basically means that AMD is going to take off-the-shelf ARM-designed cores (it will not be designing its own custom cores that just happen to speak the ARM language), build a system-on-chip around it, and then plug a bunch of these chips into a board that allows these chips to interact (i.e. pass data between each other) in an efficient manner.
This isn't too different from what AMD has been doing with its SeaMicro systems, except current designs ship with Intel's (INTC) Atom processors.
So, what's ARM got to do with it?
ARM Is Short For "Outsourcing" - Cost Cutting At Its Finest
As I watched the AMD/ARM announcement conference, the company was quite eager to hammer home how "great" a partner ARM is and how wonderfully efficient its processors are, and so forth.
That's actually true. ARM is a wonderful partner for companies that do not want to spend the money to develop their own low power processor cores. The designs are quite good, and the license fee is quite modest - roughly 1.8% of the sale cost of the chip.
This actually makes perfect sense for AMD, as it significantly reduces operating expenses. It would have cost significant time and money to develop a competitive x86 low power micro-server oriented chip to compete with the upcoming Intel "Avoton" as well as the solutions from the ARM camp such as Applied Micro's (AMCC) X-Gene (using a custom ARM compatible core).
However nice this might seem in the short term, especially as "ARM in servers" is in vogue, this is a losing long-term strategy.
Very Little System-On-Chip Experience
While AMD is cutting corners by licensing the ARM "Cortex A57" core, it still needs to take these designs and build a viable, low power system-on-chip. Contrary to popular belief, the actual CPU core is actually a fairly small part of a modern mobile system-on-chip, and a "server-on-chip" would likely even be less about the CPU as it would need to integrate many server-level peripherals.
Granted, AMD's experience in designing server CPUs is likely to be of help here. It's hard to remember, but AMD did actually bring quite a lot to the table with its initial iterations of the "Opteron" processor. AMD took advantage of Intel's complacency at the time and produced superior CPU cores, platforms, and interconnects. AMD will likely leverage this technology and expertise in building a next generation ARM-based server chip.
Unfortunately, the big selling point for ARM "servers-on-chip" are the fact that they are extremely low power and utilize a number of very clever power-gating techniques to keep these chips using power only when necessary. AMD's system-on-chip design is NOT anywhere near the level of sophistication of its larger, better capitalized competitors. Given that AMD plans to intro these chips in 2014 - and further given that this "ARM Epiphany" was likely made in the last year or so (near the time of the SeaMicro purchase) - it is very suspect that the firm will ship a particularly competitive part at launch.
Intel - Vicious Competition Even In This New World
AMD is very likely backing out of the higher performance server chip race because it simply cannot compete. Intel's relentless drive to bring power efficiency up and more tightly integrate more components in its "Xeon E5" series of chips has rendered the current "Bulldozer" based Opterons almost worthless. According to the Anandtech review of the Xeon E5,
Whether you want high performance per dollar or performance per watt, the Xeon E5-2660 is simply a home run. End of story.
The review goes on to note,
For those who are price sensitive, the Xeon E5-2630 costs less than the Opteron 6276 and performs (very likely) better in every real world situation we could test.
So, unable to compete in the "real" server segment, AMD is now attempting to fight in the "micro-server" space. Unfortunately, Intel is not a company that likes to lose business, even at the low end.
According to both official and leaked information, the upcoming Intel "Avoton" micro-server chip will:
- Launch in 2013 (well ahead of the supposed 2014 launch of AMD's micro-server ARM chip)
- Feature 2-8 next generation "Silvermont" Atom cores
- Have everything a micro-server needs integrated in: SATA, Gigabit Ethernet, USB, DDR3/DDR3L support
- TDPs for the entire SoCs will range from 5W - 20W
Intel will be showing up to this fight, too, and given its very aggressive Atom roadmap and the investment community breathing down its neck about the ARM problem, it is likely that it will put its considerable resources to use to try to dominate this space.
AMD Is Between A Rock And A Hard Place
To be perfectly fair, I do not believe that AMD's current struggles are necessarily indicative of poor execution and/or planning on management's side. In the x86/high performance CPU space, it is simply outgunned by a major competitor with many times the financial and technological resources.
AMD was able to catch Intel off-guard earlier in the early 2000s mainly because Intel was complacent. After losing significant server and PC market share to AMD's superior products, Intel finally went into full force and used its muscle to focus on bringing the hurt where it counts - in R&D. Intel's chips are now unquestionably superior on a performance/watt basis on the CPU side, and its integrated graphics processors are improving dramatically (2x improvement) each generation.
The problem I see with AMD is that it's simply between a rock and a hard place. Against Intel in the high performance, high efficiency CPU spaces, it is a very limited competitor. In the low power smartphone/tablet space, it is nearly non-existent as it, once again, has no really competitive products to offer. While its latest "Z-60" APU is very fast, it is also a power guzzler compared to the Intel "Clovertrail" or the army of ARM-powered SoCs. In micro-servers, I just don't see this competitive positioning improving unless the SeaMicro fabric really turns out to be a "silver bullet" (of which I am still skeptical).
A "Buyout" - Don't Fall For These Obvious Traps
Every so often, a buyout rumor comes along. I can't even begin to describe how many times Nvidia's (NVDA) stock ran up over the last 10 years or so on rumors that Intel would acquire it. I expect that there will be similar "rumors" for AMD in light of the massive share price decline and the large institutional ownership. Reactions to such a rumor are likely to be quite violent in light of the high short interest:
The arguments people will make will likely include:
- Company has valuable graphics IP
- x86 design skills will be relevant to other companies
The truth is, while the graphics IP is valuable, it is also very specialized. Unless the potential acquirer wants to go head-to-head with Nvidia in producing gaming GPUs for laptops and desktops, the graphics IP is not particularly useful. One could also make the case for general purpose GPU/heterogeneous systems architecture being useful to a potential acquirer. This could be useful, but the caveats here are:
- The major SoC vendors that design their own graphics - Qualcomm (QCOM), Nvidia, and Intel - are already pursuing this quite aggressively in their low power, mobile designs.
- For the less aggressive vendors, ARM and Imagination Technologies offer very good off-the-shelf GPUs to license for mobile (that do or will support GPGPU).
In short, the technology that AMD has, people either don't want (x86, as Intel owns that space) or don't need (mobile graphics).
From a more financial perspective, AMD has $1.48B in cash and $2.04B in debt, so any potential acquirer is going to end up inheriting a net debt position, further making the proposition unattractive.
AMD is doing what it can to survive, and I believe that it will survive. However, don't let the "ARM partnership" and buzzwords fool you - licensing a core that anybody else can and building a system-on-chip for servers isn't anything special. The SeaMicro fabric may prove to be interesting for these applications, but it is unclear just how much of a competitive advantage this will be on a go-forward basis.
AMD will likely carve a nice niche as a semi-custom CPU/SoC developer. Its ability to leverage both x86 and ARM designs gives it a nice degree of flexibility, and its graphics expertise will come in handy (especially for things like game consoles). It will likely reset to its new baseline revenue rate of $1.3B (the new target for cash flow breakeven) and then from there start the slow and steady path to growth. However, this growth can't happen if the company is not focused, and I worry that AMD is simply throwing everything it can at the wall and seeing what sticks.