AMD announced its quarterly results for Q3 2015 yesterday, and the short-term news wasn’t good. While company revenue rose by 12% compared to Q2 2015, total Q3 sales were down 25% year-on-year. AMD reported that demand for both its semi-custom and PC hardware increased in Q3 thanks to Carrizo’s ramp and the approaching Christmas holiday. AMD announced several new projects to hopefully boost its revenue in 2016 and beyond, including the sale of its assembly, test, mark, and pack facilities to a new joint venture with Nantong Fujitsu Microelectronics (NFME).
As part of this deal, AMD will transfer 1700 employees to the new company, contribute its facilities in Penang, Malaysia, and Suzhou, and sell 85% of its share in these facilities and operations to NFME. AMD will receive $371 million, including $320M in cash. Earlier this year, we heard rumors that AMD had attracted a new source of capital that would help offset its cash burn-through rate, and this announcement confirms it.
AMD reported some modest ASP improvements in GPUs, but CPU/APU prices were down sequentially and year-over-year. The enterprise, embedded, and semicustom segment (EESC) grew 13% sequentially, thanks to seasonally higher sales of Xbox One / PS4 chips.
Last quarter, we noted that AMD’s APU sales had tanked year-on-year, and while Q3 improved those results, the gap is still critical. For the past 15 years (save for a brief period in 2004 – 2006), AMD has had one critical weakness compared to Intel — it lacked a low-volume, high-margin business that could shield it from the vagaries of the consumer space. One critical reason Intel is still clocking in margins north of 60% and recording solid revenue quarter after quarter is because Intel’s share of the still-growing enterprise market is propping up its financials. AMD has no such shield.
Even in the consumer space, AMD has very little presence in the high-end boutique systems that are driving mobile gaming or high-end consumer sales. Its Radeon Fury family, including the R9 Nano, will sell into these segments (Fury sales are credited with improving GPU ASPs), but it can’t offer a high-end AMD solution that’s competitive with in both CPU performance and power efficiency. The lower-end systems that anchored AMD’s APU sales are falling to tablet SKUs.
AMD bears full responsibility for its own mistakes and missteps, but the company would be in a vastly different position if the PC market was still growing at 2-3 percent a year, or even holding flat. The steep quarterly declines that we’re seeing are killing the company’s profits — AMD has been unable to trim its costs faster than the market is trimming its income. The company isn’t just trying to reinvent itself — it’s trying to reinvent itself in the worst, most prolonged slump in PC sales, ever.
One thing Lisa Su noted yesterday is that the company is exploring shifting from an “opportunistic” licensing strategy to a strategic one. Put simply, AMD wants to see if its patent portfolio can be shopped around for licensing opportunities. Su didn’t use any of the language patent trolls tend to prefer, like “Aggressively pursuing infringement cases” or “We have a new method of sucking the blood out of grandmothers and war orphans,” but I suspect there’s two motives behind the move. First, AMD needs the revenue patent licensing could bring. Second, AMD wants to show investors that it’s serious about leveraging all available assets.
I believe that AMD’s decision to reorganize and centralize its Radeon division is a good move for the company and the future of its graphics products, but this kind of restructuring also makes sense as a precursor to a spin-off, however ill-advised that spin-off might be. Even a casual glance at the company’s business segments shows that the embedded / semicustom division is the one making money. In the past, when AMD split its GPU and CPU reporting into separate divisions, the GPU division often earned a small quarterly profit.
These restructuring and alignment changes are meant, I think, to show potential investors that AMD is serious about improving its own finances by taking steps it had previously avoided. The Silver Lake deal may be on hold, but it wouldn’t surprise me if AMD is still shopping for an angel investor.
The last thing I want to talk about is K12 — or rather, the increasingly obvious lack thereof. During the conference call, AMD mentioned Zen repeatedly. When asked about Zen’s roadmap and prospects, Lisa Su had this to say: “In terms of long-term roadmap we are extremely committed to high performance x86 CPUs. And there should be no confusion on that point… it continues to be our number one priority for the company.”
Compare that statement against her remarks on the ARM business, when analysts asked when we should expect to see the Cortex-A57 in-market:
“I continue to believe ARM has a place in the datacenter both as you think about towards the convergence between networking storage and servers. I think it’s fair to say for all of us that it’s been slower to adopt in the server market just due to the some of the software and the infrastructure. Relative to Seattle we will be starting our first modest production shipments in the fourth quarter in this coming quarter this year. I view it as a longer-term bet, so no question that server market is attractive, datacenter is attractive, we’re very focused on it from an X86 standpoint and we’ll continue our ARM efforts in a complementary way.” (Emphasis added)
I’m increasingly convinced that K12 is either canceled or indefinitely delayed — and there may be good reasons for AMD to take this route. Ever since Keller left, AMD has fallen over itself to reassure people that Zen is finished, ready for prime time, and will launch in commercial volume in 2017. K12 hasn’t even gotten an official mention. With the PC market collapsing around its ears, AMD needs a product it can sell to a market that it knows exists. ARM servers may be a great play in the long-term, but there’s no short-term certainty in that space.
The fact that it takes 3-4 years to bring a new microprocessor design to market, even in the best-case situation, suggests that AMD made the decision to build K12 back in 2012, when the PC market was a very different place and its own revenues were much higher. At the time, the company obviously thought it could launch simultaneous new products in both the ARM and x86 space, capturing burgeoning demand for the new ISA and positioning itself to go toe-to-toe with Intel again in the high-end x86 market. With CPU revenue collapsing and the clock running out, AMD simply can’t afford to pour money into a server product for a market that won’t catch fire for several more years.
This is all supposition — I have no inside information on K12 or AMD’s long-term ARM plans — but I suspect the company wants to have Seattle available in case the ARM market starts to show signs of life, but has pulled resources away from the more-expensive custom efforts to focus on capturing market share in x86. For a company in AMD’s position, it’s a smart move.
AMD expects revenue to decrease 10% in Q4 2015, thanks to a seasonal decline in demand for semi-custom units. Compute and graphics revenue should increase sequentially, possibly offsetting some of that decline. The company made no remarks on its rumored R9 Fury X dual-GPU products, or on timetables for future GPU introductions in 2016.