Thank you to those that have enjoyed and commented on my recent Advanced Micro Devices (AMD) articles. This will be the last in what is meant to be a three-part series.
In the first article, I discussed my thoughts on potential negatives of AMD and the positive earnings surprise stemming from the better-than-expected sales in the Computing Solutions group. In the second article I discussed why I felt the Graphics segment of AMD's business had sequential revenue decline, along with potential positive catalysts for the next half of the year.
In this article I will discuss my thoughts on the business model surrounding AMD's semi-custom business and go into what I feel the real potential of consoles are for AMD (hint - it's not revenue directly related to console sales).
Gross Margins vs. Operating Margins
Throughout the earnings call, Mr. Read and Mr. Kumar both referred to the "power" of this new semi-custom business model. To quote Mr. Read:
What's powerful about this business is you can see a significant increase in our revenue quarter-to-quarter as we begin to ramp this at the same time you see our expense level decline by almost $30 million to the 450 level showing the benefit of doing that NRE (PH) customer funded development and how it flows through to the bottom-line at that double-digit level.
To illustrate an example, look at this quarter's revenues and gross margins. AMD generated $1.16B with a cost of sales of $702M, leaving AMD with $459M to run the business. After R&D and SG&A expenses totaling $479M, it was operating at a non-GAAP operating loss of $20M.
Had AMD already reached their goal of $450M OPEX, they would've generated a break even operating income.
The entire tone of the call regarding console sales from management was to ensure that it was understood these revenues would have roughly low double-digit operating margins. NRE stands for non-recurring engineering. Mr. Read's comment shows that the customer helps fund the research and development for the custom silicon AMD is designing. Pair this with the following statement from Mr. Kumar:
When you look at the semi-custom model from a game console standpoint, it is a very different model. It's a model where the costs are upfront funded jointly between the customer and us...It's lower gross margin than the corporate average, but the majority of the gross margin dollar's fall to the operating margin...As far as the details of the business and how you manage it between the sales marketing, G&A, and R&D, our leads are coming in the second on the R&D and G&A, but from a marketing standpoint, I can tell you that there is not much, because essentially we are selling products to captive customers...
AMD jointly working with the customer and sharing cost gives it the advantage of being able to deliver tailor made silicon fitting a customer's needs without breaking the bank. And because the operational expenses after production begins are low, the spread between gross and operating income is very tight, so "it flows through to the bottom-line at that double-digit level."
Minimizing Cost With Repeatable, Scalable IP
A few weeks ago I penned an article regarding AMD's Scalability in its IP. Recently, SA contributor William Meyers published an article echoing similar sentiments (Mr. Meyers's article touches on AMD's Embedded Systems business, which I will not be discussing in this article). AMD's reusable, scalable IP consists of low-power Jaguar x86 cores and Graphics Core Next IP blocks for the GPU portions of SoCs and GPUs.
By focusing their R&D efforts on repeatable blocks of IP, AMD is able to use these blocks in a wider array of applications. AMD's recently launched Kabini and Temash SoCs, which attributed to the earnings beat this quarter, are built using four Jaguar cores and 2 GCN blocks integrated on the same die along with in/out functions and a memory controller. Kabini and Temash are essentially the same chip, configured with different clock speeds in the CPU, GPU, and memory controller to achieve specific power consumption levels. The PS4's APU uses 2 Quad Core Jaguar blocks and 18 GCN cores to power a next generation console; these IP blocks can be scaled to various usages. AMD's GPUs are comprised of GCN cores.
AMD diverts the majority of its R&D into these repeatable IP blocks, which gives it greater return on investment due to being able to leverage these blocks to design the custom silicon I have mentioned previously, as well as support its own SKUs.
As for some numbers, AMD generated $1.16B in revenue this quarter with GM at 39%. The guidance for next quarter is a revenue increase of 22 +/- 3% with a gross margin of 36%. Therefore, based on guidance, revenue should come in at $1.38B to $1.45B.
Here I will suggest listening to the earnings call at roughly the 23 minute mark. When answering a question from Mr. Vivek Arya, Mr. Kumar uses single-digit operating margins at first, but sounds like he corrects himself to double-digit operating margins at the end. Mr. Read's "double-digit level" quote is my second data point to pin my operating margin at low double digit levels. I assume 10-20% is low double-digit levels, so here I will split the difference and go with 15% as an operating margin.
Assuming revenue with the exception of consoles remains flat at $1.16B, then the increase is derived explicitly from console APU sales, meaning that $220M to $290M in revenue is generated from console sales next quarter. Using the midpoint of my estimation, ($220M+$290M)/2 = $255M.
To verify this makes sense, assume the $1.16B revenue GM remains static at 39%. AMD expects revenue at the midpoint of the guidance to be $1.41B with GM of 36%. Therefore, $1.16B*39% + $255M*X = $1.41B*.36. Solving for X yields a gross margin for consoles of around 28%, so the 15% operating margin given the semi-custom business is jointly funded seems plausible.
Stacy Rasgon of Sanford C. Bernstein & Co. has console royalties pegged at $100M to $150M a year. Mr. Kumar stated in his closing remarks prior to the Q&A session that:
...game console royalty revenue included in the semicustom business is expected to decrease moving forward although the decline will be more than offset by our growing semicustom...revenues.
Mr. Kumar also stated when answering Mr. Rasgon's question that the focus should be on operating margin.
I believe this quarter should be the beginning of volume production for the console silicon, and based on the estimated console revenues of $220M to $290M with operating margins of 15%, I estimate the actual operating income from consoles at roughly $33M to $43.5M this quarter. As yields improve and volume ramps higher, I expect this number should improve, which will then more than make up for the loss of royalties. The next two quarters will be the first chance to see if AMD's new semi-custom business model pays off.
Lastly, assuming that revenue stays flat at $1.16B excluding console sales, and that console sales will add $255M give or take, we arrive at $1.41B in revenue. $255M/$1.41B works out to 18%, which is roughly inline with AMD's guidance of 20% of revenues from semi-custom. However, my estimations could be wildly off, and we won't know until Q3 and beyond.
For another possible catalyst, China is potentially lifting the video game ban, but with many caveats. This could give Microsoft, Sony, and Nintendo all a boost, which I'm sure would make AMD happy. However, because of the caveats I am not personally banking on this.
But at the end of the day, management has beat expectations for three quarters in a row. Given 36% GM on $1.41B, and accounting for an anticipated OPEX of $450M, the thought of being cash flow positive seems plausible.
How I View Consoles for AMD
I view consoles as a stepping stone for AMD in two different ways.
AMD is known for lagging in single threaded performance, which penalizes it when compared against Intel (INTC). Many games, benchmarks, and applications benefit from single threaded performance. Microsoft and Sony both felt comfortable with putting AMD silicon in next-generation consoles, which will require coding programs to take advantage of more cores.
Regarding the graphics portion of the APUs, this will give AMD the chance to work more closely with developers, which could help create a halo affect with the PC APUs and GPUs. Electronic Arts (EA) will not be releasing sports games that use their new Ignite engine on PCs in the near future. FIFA 14 will be released on PC, but it will not make use of this new engine; it will be interesting to see how the visuals stack up when comparing PCs to consoles. When EA stated that consoles are more powerful than PCs, you have to keep what it is saying in context. On paper, PCs may be more powerful, but this article on bit-tech goes into some of the excess overhead associated with coding games for PCs.
So by having games developed for AMD hardware, there is a chance that performance could improve for AMD's CPUs and GPUs.
Unified Memory and the HSA Foundation
I have not seen any very definitive information regarding consoles being HSA compliant, and given that the first revisions of HSA standards are just now hitting the streets and that the first games will be developed using APIs, the consoles will most likely serve initially only to get programmers used to dealing with a unified memory address space.
The HSA foundation is the collaborative work of AMD, ARM Holdings (ARMH), Qualcomm (QCOM), and several other companies. The overall goal is "to make it dramatically easier to program heterogeneous parallel devices." This is accomplished through open source software support from low level compilers and royalty free open specifications.
Straight from AMD's 2012 annual report the company states:
...Intel has been able to control x86 microprocessor and computer system standards and benchmarks and to dictate the type of products the microprocessor market requires of us.
Is there anything wrong with this? No, it's just business on Intel's part. But HSA gives AMD the chance to jointly develop compilers that will translate high-level code into machine language that will run more efficiently on AMD processors.
HSAIL stands for Heterogeneous Systems Architecture Intermediate Language. How this works is a programmer writes code to the HSA open standards specification. It is then the device vendor's responsibility to ensure the HSA runtime infrastructure is in place. Other than the stated advantages of HSA from the PR slides, the HSA runtime infrastructure comment leads me to two major advantages.
First, if an app is coded to HSA standards, that means there is much less overhead in porting that program between different devices. Based on IDC numbers, 712M smartphones and 128M tablets were shipped in 2012, compared to PC shipments of 352M during the same period; given that ARM IP powers the majority of smartphones and tablets, ARM based SoCs outnumbered x86 devices by around 2:1 in consumer applications. By writing code to the HSA standards, all the app developers can reach a broader audience by having programs that don't care if they're running on AMD x86 hardware or a Snapdragon 600, meaning HSA programs are largely hardware agnostic.
The second advantage is having programs compiled for AMD hardware on a compiler not optimized for Intel architecture. Recall my quote from the annual 2012 report regarding Intel essentially dominating the x86 market. A computer consists of the hardware, software, and I/O devices we use to control it. By ensuring the code this hardware runs is optimized for AMD hardware, the efficiency of the device goes up. And AMD excels in both OpenCL and C++ AMP performance, which are both supported by HSA standards.
At the end of the day, games will be programmed on the next generation consoles to take advantage of the multi-threaded capabilities of an 8 Core CPU and a dozen or more GCN cores using a unified memory space. This will force programmers to gain exposure to optimize games for AMD silicon if they want to take advantage of the new hardware. If the consoles turn out to be full-fledged HSA devices, even better.
Speculative Best Case for AMD
Whether or not Microsoft (MSFT) is a silent partner of the HSA Foundation remains to be seen. But Microsoft nonetheless supports the HSA efforts. For the past couple of years, AMD has hosted what is known as the Annual Fusion Developer's Summit. Microsoft is a corporate sponsor of this event.
Last year during AFDS 2012 was when the formation of the HSA Foundation was announced. The year prior at an AFDS event, Microsoft introduced the C++ AMP (Accelerated Massive Parallelism) programming language to allow programmers to gain extra compute power by tapping into the GPU. The capabilities of this language were demonstrated on AMD hardware.
Microsoft's involvement with AMD would be a mutually beneficial relationship for both companies. Microsoft just incurred a decent sized inventory of their Surface RT tablet. An HSA compliant Surface RT would give RT users access to a wider array of apps and programs as HSA compliant programs are written. The same broader range of access could equally apply to the Xbox One as Microsoft aims to gain a more dominant role in the living room.
This again is highly speculative. As far as timelines for when we may receive news - there is the Hot Chips Symposium in August. The last event scheduled for the symposium is a discussion of the Xbox One APU led by Microsoft. Another possibility would be at this year's AFDS in November. Even if it turns out Microsoft has no official involvement, stock moving news could come from these events so they're still worth paying attention to.
Prior to the conference call, it seemed most of the buzz around AMD was generally positive. The stock held onto gains after hours, and it was the negative catalyst of downgrades due to console margins that caused share prices to tumble the next morning. After the stock tumbled, it seemed to me the negative press picked up causing the herd to sell, further driving down the price.
One of the things that caught my attention when writing this article was the fact that the Credit Suisse analyst covering the conference call stated something to the effect of being disappointed with management guidance during the call. Credit Suisse was one of the major firms issuing downgrades after the CC. I will look for similar sentiments and tones during the Q3 call.
But the kind of news that would make me personally want to liquidate my shares in AMD -- think big payments to GloFo, production issues, or cash dipping low -- did not materialize during the earnings call. I have spelled out what I consider to be potential negatives in prior articles, as well as link to other SA contributors that point out potential pitfalls in AMD.
I feel there is a potential for the operating/gross margin spread to tighten, pulling the operating margin closer toward 20% as the production process improves and volume ramps. This could also give full-year revenues from console APU sales closer to $1.2B. At $1.2B and 20% operating margins, the console operating income is closer to $240M income, which would add roughly $.32 positive EPS to total earnings for the full year (using 752M shares). Given Mr. Rasgon's estimation from earlier, this adds roughly another $90M-$140M to AMD's bottom line. I will adjust these estimations during the Q3 call accordingly, but given that $240M is higher than the current revenues consoles are likely generating for AMD, I still view the win as a positive. Especially given the minimal OPEX required for AMD to fund the development.
Other than a steady stream of revenue, I see consoles as an opportunity to make AMD hardware more mainstream, and the possibility of the improved performance discussed above.
Overall, I remain bullish on AMD. There have been three positive earnings surprises this year, and AMD expects to be cash flow positive in Q3. Currently my cost basis is in the low $3 range. I intend to add small lots of shares to my position at deflated prices with the overall goal of selling portions of my position if the stock moves higher to lower my cost basis for the remaining shares. At the same time, I have no problem liquidating my entire position if I feel things are going south, but I have not seen any of my trip wires materialize.
One of the large negative factors I look out for is payments made to GloFo. During the conference call Mr. Kumar stated that roughly half the required purchases from GloFo had been made for the year.
I do not consider a revenue increase of 22%, winning spots in the next generation consoles, and cash flow positive next quarter as "going south." AMD gained traction with its Richland APUs, and this week Kaveri motherboards have started rolling out, showing signs are building for the Kaveri launch. Mr. Read stated there are "strong mobile design wins" in the works. There are other positives aside from consoles.
Since I consider AMD a speculative play, I ensure that I maintain my cost basis as low as possible and set my limits to make sure I am not losing money that I am not okay with losing.
But to me, the turnaround seems firmly on track. AMD typically trades north of $6 when positive and I see several strong potential catalysts in the next half of the year and beyond. So for now, I remain in the smaller herd of bulls.
Additional disclosure: I actively trade my AMD position. I may add or liquidate shares at any time.