One of my reader's recently gave me some advice that I would like to take to heart in this article, and that is that less is sometimes more. This article will be more focused than most.
In order for Advanced Micro Devices' (AMD) semi-custom strategy to work, the company only needs to break even in the traditional markets, and garner semi-custom design wins. In this article, I will use the console chips as a case study to help readers understand the semi-custom model.
Semi-Custom Business Creates "Sticky" Revenue and a Moat for AMD
To understand the earnings power of consoles, or future semi-custom designs, you must also consider the type of design wins these represent. Mr. Read referred to this as "sticky" revenue; it gives AMD a moat as they are contracted to make the chips, and these designs typically have longer product lifetimes.
The CEO seemed to talk about the console SoCs as almost having a business card status, showing the potential of semi-custom silicon to other customers, and stating the business model is generating customer interest on two separate occasions during the conference call. More confirmed semi-custom design wins for AMD would be a big win, based on the fact these designs provide steady revenue and would increase the semi-custom moat.
One of the design goals of HSA, along with semi-custom, is to manufacture one cohesive piece of silicon designed for a specific use case.
The above image is courtesy of SemiAccurate. The thing to notice is the blurb in the upper-lefthand portion of the diagram regarding the 15 special purpose processors. The other thing semi-custom brings to the table is the ability to interface DSPs with the CPU and GPU cores to create a highly-specialized chip, capable of performing its specific task very well.
For the non-technical, a DSP stands for "digital signal processor." Think of it as a chip that is designed to do one thing, but do it very well. Comparing a DSP to a traditional CPU core, since the DSP only performs a very specific function, it is typically able to do its job more efficiently, and with fewer transistors, meaning it is smaller and consumes less power.
As businesses look to differentiate from competitors, AMD is able to help achieve this by designing custom silicon tailored for specific use cases.
Illustrating the Earnings Power of Semi-Custom
Based on Q3 numbers, to achieve the ~$450M OPEX goal, R&D spending totaled ~$290M, with SG&A expenses at ~$160M. Note, these numbers are not the actual Q3 numbers, but rounded to sum to AMD's target $450M OPEX.
During Q3, we can see this reduction in OPEX greatly assisted AMD in returning to profitability. And it's this reduction in OPEX that gives the semi-custom business so much clout.
In order to explain, in Q2 AMD's CS (computing solutions) operating segment generated an operating income of $2M on $841M in revenues, whereas GVS (graphics and visual solutions) was break even on $320M. Overall results for Q2 was an operating loss of $29M, with a net loss of $74M (interest expenses of $42M). If, in Q2, AMD had met the company's goal of $450M OPEX, it would have had a break even quarter at the operating level, with losses stemming mainly from interest charges.
To make semi-custom work, all that has to happen is break even in traditional products so the R&D used to design that IP can be applied to semi-custom designs. This allows a consistent level of OPEX, with the customer funding above and beyond AMD's typical R&D level, such that OPEX does not go up unless there is the associated product in the pipeline.
AMD creates CPU cores for the company's CS segment, and creates GPU cores for the GVS segment. The IP created to create CPUs, APUs, and GPUs is then used to create custom silicon, with the customer funding a large portion of the NRE (non-recurring engineering expense). This allows AMD to keep OPEX minimal (little R&D required on AMD's part to design the chips), while growing revenues.
The concern was the focus on gross margin, but this is the wrong view of the semi-custom model. Gross margin should be the concern of traditional products as AMD needs to recoup the R&D that goes into building them. Because there is little overhead from expenses like marketing, SG&A expenses also come in lower in semi-custom. So while AMD may take a gross margin hit to supply custom chips, most gross margin dollars fall to the bottom line, unlike in the traditional business.
Looking At Revenues Per Console
TechInsights puts the ~price of a console APU at $125, which is higher than most of the initial estimates. Wedbush estimated console revenues at $401M during Q3, which if these numbers are close, leads to 3.2M APUs sold during Q3.
Both Sony (SNE) and Microsoft (MSFT) managed to sell over 1 million of the companies' respective consoles on launch day (each, not combined). Sony is anticipating 3 million units shipped by year's end, and analyst Colin Sebastian is expecting a similar number for Microsoft.
IHS (via Economist.com) is slightly less optimistic, expecting a combined shipment of 4.6M units.
However, Microsoft and Sony still must build the consoles if the companies expect to sell them, and both are prepping inventories for a worldwide roll out throughout 2014. Based on this data, I believe it is possible for AMD to ship around 4M console APUs during Q4, give or take.
4M console chips at ~$125 each generates revenues of $500M, which would be roughly in line with management's expected sequential increase of 2% to 8% for Q4. Based on CFO comments, along with those of Dr. Su, there should also be room to extract further profits from these APUs by doing things like boosting yield.
Given this data, a 100 bps increase from ~15% operating margin to 16%, and $500M in revenues, would total $80M. With IHS forecasting 4.6M units sold by the end of the year, I will consider anything exceeding this estimate as a bullish indicator for the potential of consoles.
One takeaway from this I would like to reiterate is the ~$290M runrate in R&D. AMD had been working on console chips for quite some time, and yet it was kept somewhat a secret until this year. But now that these chips are in production, we can approximate what AMD feels should be the average expenditure on R&D, based on the stated $450M OPEX target.
A potential indicator of a design win could be an increase in R&D spending significantly above the $290M level.
Whether this holds true or not remains to be seen, but it will be interesting to see how it plays out if we learn of another semi-custom design and can plot R&D spending along development of the project.
The semi-custom model is very powerful provided AMD is able to fund the R&D required to build the IP necessary to remain competitive against Intel (INTC) and Nvidia (NVDA), AMD's primary competitors. This allows AMD to focus R&D spending toward key IP blocks that can then be re-used to create custom silicon.
Console APUs exhibit AMD's ability to intermix various IP blocks to create a custom chip that lowers the BOM for hardware vendors by integrating more functions in a single piece of silicon. Because AMD pays for the R&D by the sales of products in the traditional markets, associated OPEX on semi-custom designs is kept to a minimal level. This allows AMD to generate operating margins and turn a profit while sacrificing gross margins.
Taking this a step further, semi-custom design wins give AMD a moat; something the company hasn't necessarily had the luxury of in recent times. Intel has great high end chips, data center, and foundries. Nvidia has CUDA and the HPC market. But unlike the competition, contracts with OEMs prevent AMD from having competition during the product lifecycle. And as AMD's market share in tablets is non-existent, and extremely small when compared against Intel in both PCs and the data center, any design win for AMD is accretive to earnings.
While semi-custom design wins carry lower gross margins, the impact to operations is still significant. Each semi custom design win will widen this moat, which should increase investor confidence by adding stable revenues, adding certainty to AMD by decreasing reliance on the PC market.