AMD has reportedly hired a bank to help it “explore options.” This can mean different things, depending on what sort of options the company has in mind, but it’s rarely a good sign for the company in question. This development means AMD has brought in outside help to value its various assets and attempt to ascertain how to best leverage those assets to keep the company afloat.
According to Reuters, who initially broke the story, an outright sale is unlikely at this point in time. Instead, the company may look to sell various patents or other intellectual property. That’s a great idea – in theory. However, it’s unlikely to stand up in practice, for several reasons.
Firstly, there’s the fact that AMD already went through this process a few years back. AMD sold its handheld graphics business (Imageon) and its 32-bit MIPS SoC (Xilleon) to Broadcom and Qualcomm back in 2008. It divested itself of some 200mm fab equipment at the same time, and later spun GlobalFoundries off altogether.
True, AMD might be sitting on a number of interesting patents, but we’ve got a feeling that the most valuable ones are already licensed to the company most likely to be interested in acquiring them – Intel. Ditto for Nvidia. Cross-patent agreements between all three are extensive and have been updated on a fairly regular basis. Any attempt AMD might make to sell those patents to a third party would likely be constrained by the prior agreements it has in place. A potential customer, like Samsung, isn’t going to be interested in paying top dollar for patent rights it can’t exert against its chief competitors.
That leaves AMD’s two big businesses: x86 and discrete graphics. Even if AMD’s x86 patent rights are broader than they used to be, no one is going to be able to acquire that patent without first going through Intel, and taking on Intel in its core market is a losing strategy. If it wasn’t, AMD wouldn’t be in the position it’s in now.
AMD’s discrete business is valuable, but the GPU market is on a long-term trend towards integrated hardware. Discrete GPUs will never die – not so long as gaming and workstation applications benefit from them – but the market isn’t a hot growth opportunity, either. A company that wanted to take Nvidia on in the professional and HPC spheres could conceivably buy Radeon and explicitly target those areas – but that’s a pretty dubious use-case.
AMD can’t afford to sell its APU patents; they represent the future of the company. It can’t afford to scrap Radeon wholesale (it needs assets for APU development) and its ability to sell x86 assets is going to be substantially constrained.
Let’s talk about Kaveri
Then there’s the added problem of Kaveri. We’ve stated this as a side point in articles before now, but given the current situation, it’s worth bringing out into the light. Kaveri, which was originally slated for a 2013 launch, hasn’t even taped out yet. Given the 14-15 month window between tapeout and product launch, that puts the Piledriver follow-up debuting right around the same time Intel will launch Broadwell, its first-generation 14nm part.
That means that Steamroller/Kaveri will launch into the teeth of a CPU two full generations ahead of them. The reason the chip hasn’t taped out yet, according to our sources, is that AMD has spent time trying to improve its power and IPC to ensure it isn’t thoroughly hammered out of the gate.
As for Richland, there’s considerable confusion over what it really is. Estimates range from a “Piledriver 2.0” with slightly tweaked clock speeds, to a refreshed GPU with Piledriver CPU, to a mislabelled slide. All of these are possible, since AMD already used that codename to refer to budget A4 products, but we’re confident that regardless of what happens to Trinity in 2013, there won’t be a Steamroller refresh for either the APU or CPU families next year.
AMD may or may not be able to further improve Steamroller by spending more time on the microarchitecture, but 2014 isn’t the problem here, 2013 is. Sales for the next 12 months are going to depend heavily on Kabini (still on schedule) and the “Piledriver 2.0” refresh, codenamed Richland. That’s probably the biggest reason AMD has brought a bank in now, when it still has some capital in play.
AMD’s ability to survive through 2013 and into 2014, when new parts (and its ARM-based server chips) arrive is very much in question. It’s simply not clear how much more the company can cut its payroll or other expenses before it directly jeopardises its long-term ability to deliver a roadmap that investors trust to be competitive with other companies. There’s some solid reasons to consider Sunnyvale’s Trinity hardware, but that won’t hold true for long if OEMs stop believing AMD can deliver future products.