Applied Micro (NASDAQ:AMCC) gave an update on the status on its very hyped 64-bit "X-Gene" server-on-chip processor, which confirms my thesis that the company continues to mislead investors with the "numbers." While I have significantly detailed my short thesis here, and I followed up the piece here, I believe that certain statements made on the conference call and in the earnings release continue to mislead (but wording is carefully chosen so as to keep all statements "factual") investors, and more precisely, the sell-side analysts, both of whom are likely to see this earnings call as a source of renewed optimism for a company that, by all accounts, has given investors no reason to believe them.
In this article, I call into question a number of the statements made on the call and in the earnings release and to reiterate the need to stay skeptical.
X-Gene On 40nm: 36 Hours Fresh?
In the earnings call, the company noted that the company just received first samples of X-Gene over the last 36 hours. This means that every design that was shown off at the OpenCompute conference (held on 1/17/2013) was not in fact actual X-Gene silicon, but merely the chip in FPGA form:
While it was painfully obvious to anybody with basic knowledge of the semiconductor industry that a tape-out in late December was not likely to lead to fully functional silicon "out of the chute" by January 17th, the press, investors, and analysts seemed to believe that this was the case. This was apparent given that the stock has had a gargantuan run from the $4.50 mark to >$9.00 over the last several months.
So keep in mind that if Applied Micro is just now getting back the very first stepping on its silicon, it is likely not bug-free and fully functional yet. There is significant post-silicon validation that needs to be done in order for the chip to be ready to go into data-centers. We do not know how many additional steppings that the company needs in order to release fully debugged silicon (each respin takes 6-8 weeks to get back from the fab), nor do we know anything about the performance and much more importantly the performance-per-watt metric that the company has so eagerly touted over the last year.
Trying To Mislead With "Gigahertz"
On the call, CEO Paramesh Gopi was keen to emphasize the clock frequency at which the 40nm X-Gene runs at. In particular, he cited the value "2.4GHz" and then further went on to make comparisons to other ARM (NASDAQ:ARMH) core designs which run at "1.5GHz - 1.8GHz". This is absolutely, positively, and almost ludicrously misleading, as clock frequency comparisons are only relevant when discussing the same micro-architecture. Comparing a Qualcomm (NASDAQ:QCOM) Krait at 1.7GHz to an Intel (NASDAQ:INTC) Celeron at 1.1GHz solely by clock frequency and then declaring the former the winner is not correct; a common metric, known as "instructions per clock" (this actually varies depending on the mix of instructions in a given program), is a more meaningful metric of the performance of a micro-architecture.
For example, if the X-Gene is able to run at 2.4GHz but can only successfully execute two instructions per clock, it will be an inferior design to one that runs at a mere 1.8GHz but can successfully execute three instructions per clock. The problem is that without actual performance numbers (either from Applied Micro or a third party), the "2.4GHz Performance" phrasing that Dr. Gopi used frequently on the call is utterly meaningless.
Case in point: The Intel Pentium 4 ran at 1.5GHz at launch, whereas AMD's (NASDAQ:AMD) Athlon ran at 1.2GHz, but the Athlon was much, much faster. The fact that Applied Micro is even trying to pull the wool over professional semiconductor analysts' eyes like this is somewhat shocking.
Power Consumption: Conspicuously Omitted From The Discussion
A major omission from the usual "pump" of X-Gene in this conference call was any sort of boasting with respect to power consumption. Assuming that the X-Gene is substantially faster than the existing ARM designs (which I find highly suspicious at this time), at what power levels does it run? Is the performance-per-watt of an X-Gene core better than a stock Cortex A9, Cortex A15, or ARM's own 64-bit Cortex A57 (I'll talk more about this show-stopper soon)?
In previous presentations, Applied Micro has stated that the target power consumption per core is ~2W. This does NOT include many of the components on a modern server micro-processor, especially one with such highly integrated features. So for the hypothetical 4 core configuration, we are looking for at least 8W running the actual CPU cores, another 6-8W (at least) for the multitude of peripherals and interfaces that are integrated onto the chip, and then finally a couple of more watts (at least) for the L3 cache (cache is a power-guzzler) that Applied Micro has touted in previous promotional materials.
To top it all off, it is built on TSMC's (NYSE:TSM) 40nm process, which is now a last generation process. I continue to remind my readers that Intel will have a highly integrated competitor to X-Gene codenamed "Avoton" shipping for revenue in 2H 2013, which will be built on the highly power efficient 22nm tri-gate transistors. I believe that Intel will not only be first-to-market with a shipping product (this part has been sampling since October 2012), but it will have a superior (from a performance/watt perspective), more compatible product on all counts.
In short, there is something that Applied Micro is omitting; engineering is the art and science of trade-offs, but the company - according to Paramesh Gopi - has made precisely no trade-offs in the design of X-Gene. This smells suspicious.
28nm "Test Chip": The Biggest Deception Of All
The earnings release contained the following nugget of hype, likely planted in to ensure that tomorrow's analyst reports are filled with upgrades and praise:
During the quarter, the Company taped out its ARM 64-bit X-Gene processor chip in 40nm. In mid-January 2013, the Company also taped-out a 28nm test chip version of the X-Gene chip which incorporated several other features in addition to the geometry shrink.
Notice the very careful wording and the distinction between the language used for the tape-out of the 28nm processor versus the 40nm process:
During the quarter, the Company taped out its ARM 64-bit X-Gene processor chip in 40nm
This is a very clear statement: the company finished the initial design of the 40nm X-Gene, which means that the masks were sent off to the fab. This is usually the final, fully featured product, and then from there, the chip will go through post-silicon validation to make sure everything works correctly (it very often does not on the first try).
But then I look at the second statement:
The Company also taped-out a 28nm test chip version of the X-Gene chip which incorporated several other features in addition to the geometry shrink.
While it is very likely that this is not a false statement, it does not mean what most people think it means. Before I interject my own explanation, I would like to include what Dr. Gopi had to say about it when directly asked about it by an analyst:
Analyst: And also just to clarify, the 28nm stuff you talked about earlier, is a test chip tape-out, so we're still expecting the final product tape out to happen a little bit later this year, right?
Paramesh Gopi: We've not announced when it is, but yeah, this is a test chip tape out, very similar to what happened last year with the first X-Gene test chips.
Anybody following the Applied Micro story will know that while we saw first test chip tapeouts of the 40nm X-Gene in late 2011/early 2012, we did not get the final production silicon tape-out until December 2012/January 2013. If the development with 28nm continues along a similar development trajectory, then we will not see a true first tape-out of the 28nm chip until late 2013/early 2014, and given that the 40nm X-Gene will not be shipping for revenue until late 2013/early 2014, we can reasonably conclude that - if everything goes according to plan - we will see first commercial shipments of the next generation 28nm X-Gene in late 2014/early 2015. This should be just about the time when Intel will begin selling its 14nm "Atom" chips for micro-servers/communications/networking:
In short, Applied Micro's nearest (and strongest) competitor will have a 2+ generation process technology lead, in addition to far greater R&D resources. It also has a more developed set of server customer relationships in addition to an outstanding track record as a supplier. Without a substantially better product (and all accounts suggest an inferior one), Applied Micro's solution will be considered very uncompetitive and will be unlikely to sell, unless it is offered at a significant discount to its peer (which may be difficult, given Intel's manufacturing/margin advantages).
It should also be noted that the 22nm Atom for servers has been sampling for months, suggesting a much, much earlier tape out than Applied Micro's first gen X-Gene. On a competitive side, things do not look good.
Can This Venture Even Be Profitable?
One thing that I find very interesting is that no analyst seems to question the profitability prospects of this venture. While Applied Micro will be first-to-market in the ARM space with a low-power SoC for micro-servers, investors seem to forget that ARM Holdings itself will be releasing its Cortex A57 64 bit core design that any company can choose to license. This means that while Applied Micro spent all of this money and time designing and validating its own processor core (which is touted as a main ingredient to X-Gene's "secret sauce"), this will quickly become commoditized.
Now any SoC player such as Qualcomm , Samsung (OTC:SSNLF), Broadcom (BRCM), and even AMD will be able to produce competition to the X-Gene (and have announced plans to do so). This means that there will be a whole deluge of players in this new "micro-server" space, in addition to Intel, to try to take some of the X-Gene pie on a longer-term basis. We are seeing the problem of profitability in the mobile SoC space hit many vendors hard, and I believe that with broad commoditized nature of the processor core, it will be very difficult for a company like Applied Micro to maintain profitability in this new space as well.
So with Intel, Broadcom, Qualcomm, Samsung, AMD, and others to compete with, can Applied Micro capture enough market share here to get the quarterly run rate of these products high enough to reach GAAP profitability? It does not seem likely, given Applied Micro's size and shoddy track record of executing on these grand initiatives (like the firm's "Titan" that was claimed to be this amazing, wonder chip but was quietly cancelled).
Applied Micro reeks of hype and fanfare, and it seems to me that we have seen this before. Early last year, the stock ran all the way up to $9.30+ on X-Gene hype before the stock came crashing back to reality in the $4.50 range. I believe that this is no different, and that as the year progresses, and as the enthusiasm for this product fades (especially as Intel begins the hype machine around its 22nm Atom), there will be a gradual, but eventual fade of the stock back to the $5-$6 level as the hype machine runs out of steam.
Disclosure: I am long INTC, AMD, QCOM. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Additional disclosure: I am long AMCC put options, and ARM put options