In this article, I'd like to take a look at Advanced Micro Devices' (AMD) ARM (ARMH) server strategy. The company announced back in late 2012 that its "ambidextrous" strategy was one where it would flexibly utilize IP from multiple different vendors in order to continue its transformation into a broader "system-on-chip" company rather than simply the "other PC chip company". I think AMD has made some interesting decisions with its server strategy going forward that I believe could actually be substantially more profitable than what it is currently doing in the server space today.
The Positives - Development Costs Substantially Reduced, Focus On SoC
The primary positives for AMD's new ARM server initiative actually have nothing to do with the ARM cores - the notion that ARM's designs are somehow inherently better than competing designs simply by virtue of being compatible with the ARM instruction set are ludicrous. However, the great thing about them in this particular instance - and for a company like AMD which is trying to shave off R&D dollars - is that it provides a "good enough" ready-to-go processor core which can then be designed into a custom system-on-chip that can integrate various functionality specific to the particular target market and workloads.
For example, AMD/SeaMicro highly tout their "Freedom Fabric" interconnect which, according to the firm, allows for elegant stitching together of a very large "sea" of processor cores for dense-serving applications. These applications typically don't require too much grunt on a per-core basis, but given the volume of the requests such a machine needs to power, having a lot of cores available to the software is helpful.
Also, Mr. Feldman - head of AMD's server division (and former CEO of SeaMicro) - pointed out that development of ARM based server processors is substantially cheaper than X86 ones. This isn't due to the ARM instruction set versus the X86 one, but instead due to the fact that AMD no longer needs to design and validate its own processor cores (this is the hardest part), and instead can focus on developing the peripherals around the core.
The Negative - The Loss Of Control Trade-Off
Nothing comes free, and I believe that the major trade-off that AMD is making here is that it will now be at the mercy of ARM's core development schedule. See, it would make no sense for AMD to take an ARM architectural license and design its own core for this segment; if it were to do that, it would probably be a better idea to design a custom X86 core for this job, since it actually has an x86 license and can try to build a selling point by claiming x86 compatibility.
The bet, it seems, is that the core doesn't matter all that much in this space - you're supposed to be looking for good enough on a performance/watt basis and then differentiate on system features. Only time will tell if this turns out to be the case, but it really depends on how the workloads look in this space, and whether ARM has really been optimizing its core for these workloads this whole time.
Another problem here is that anybody can license the same core and build an SoC around it. The rapid time to market and lower development costs that make this market attractive also attracts competition, which could strangle profitability in this space for all but a few. AMD's reputation as a server chip supplier may be helpful in gaining a foothold here and holding back the hordes of competitors.
I hope AMD simply stops development of its money-sink Opterons (and big cores in general) and bets the company on these sorts of initiatives. With less than 5% market share in the slower growing traditional server space dominated by Intel (INTC), and with R&D costs for processor cores skyrocketing especially as the industry moves towards new nodes, AMD needs to position itself as an integration and customization company rather than a bleeding edge CPU core design company. I think the former strategy is not only likely to prove much more profitable, but with AMD's willingness to use ARM cores to attack new markets and with "Jaguar" looking like a solid performing X86 core, the company should simply bet on these two initiatives (in addition to graphics IP) and continue its diversification away from the PC space.
The question that still lingers though, is whether AMD will be able to break the curse and finally achieve consistent, profitable growth? Previous management nearly wrecked the company, but Rory Read seems more capable than his predecessors. Only time will tell, though.
Additional disclosure: I am short ARMH