"If something sounds too good to be true, then it probably is."
I want to start out by saying that I deeply respect and admire fellow contributor, Valuable Insights. He consistently lives up to his name and has nailed it out of the park with some of the very best calls on small cap technology stocks out there. So, of course, when he put out a counter argument to my bearish piece on Applied Micro Circuits (AMCC), I studied it carefully and, as always, took it very seriously. While I strongly encourage the reader to read his piece before continuing, the main summary points for his long thesis are the following:
- With the help of ARM (ARMH), Applied Micro will be able to successfully challenge Intel (INTC), the incumbent in the server space.
- Clever/optimal usage of an older process node helps to significantly mitigate the disadvantage of being two process nodes behind Intel in the low power server/embedded space.
- Fairly aggressive insider buying is a signal that the insiders truly believe the story that they are selling to the investment community.
- Applied Micro's solutions claim to offer a lower total cost of ownership compared to Intel's solutions, providing a very powerful selling point.
- The risk/reward favors Applied Micro: the idea is, even if AMCC takes even a small bite out of the server market, it could be a big deal relative to Applied Micro's size.
- Applied Micro could be an interesting buyout target.
- Excluding the X-Gene business, Applied Micro is cheap.
In this article, I look to tie recent developments (most notably, Applied Micro's product presentation at the Facebook (FB) OpenCompute Summit) to show that the excessive optimism that the Street has shown for shares of the stock over the last several months is still largely unfounded.
We've Seen This Before
It seems that this is the second year in a row that investors have seen this same pattern play out:
We see the same thing again...
However, there's a lot more to the story. I believe that the investment community is, at large, being completely misled by the management team (again), and I will now expose some of the very serious contradictions and half-truths that the company keeps hyping up.
Where Are You, X-Gene?
The 40nm X-Gene taped out last quarter, and working silicon seems to be back from the fab now. The company expects to be "sampling silicon to key customers" by the end of the quarter. Investors and the general public do not yet know any of the following:
- How much power does it use?
- What clock speeds does the actual silicon run at?
- How much does it cost to make? How much is Applied Micro charging for it? (i.e. how big is the "no compromises" silicon)
- Who is actually on board to buy it?
The official party line is that the chips will be shipping for revenue in late 2013/early 2014. Keep in mind that the last time Applied Micro hyped the chip up, it was slated for a late 2012 launch.
So, we won't see the first generation, 40nm chip until late this year/early next year. Keep in mind that Intel has already announced that it will be selling its own 22nm "Avoton" system-on-chip during 2H 2013 (and it has been sampling since at least October 2012).
That brings me to my next point.
Process Technology Matters - Applied Micro Says So!
It's amazing how investors fail to understand that process technology is absolutely one of the largest enablers in the microprocessor world. With smaller and more power efficient transistors, you can do the following:
- Consume less power at a given performance level or scale up performance at a given power level.
- Use less die area at a given performance level or scale up performance at the same die area.
Valuable Insights, in his article, made the following statement:
While a smaller die-size unquestionably gives Intel a cost advantage, and the ability to pack in more transistors should be an advantage, it's simply an oversimplification to say just because it's a smaller process node, it wins. Without getting into too much technical detail (read through the previously mentioned Linley report), we believe the X-Gene provides a more highly integrated solution and because of Intel's decades long heritage it has to be compatible with less efficient CISC code, vs. ARM's use of more efficient RISC code. In summary, process node is simply one of multiple dimensions impacting chip performance.
I agree. You need a good micro-architecture as well as a good system-on-chip architecture in order to unlock performance-per-watt. However, I would say that the "RISC vs. CISC" debate is moot. Modern X86 processors do not actually execute "CISC" code. The X86 instructions are decoded into "micro-ops," which are RISC-like operations that the CPU backend executes natively. There is certainly a minor power penalty as the decoding into micro-ops for X86 is a bit more power hungry than the decoding of an ARM or MIPS (MIPS) instruction, but this is largely moot at the die size levels that we are talking about here. In addition, a nearly three generation gap in process technology is more than enough to wipe out a slight decoding penalty many times over.
Further, the smaller transistor dimensions allow the chip designer to pack in more transistors (i.e. integrate more things) at a given die size/cost level. The advancement of microprocessor technology is first and foremost dependent on advancing the transistor technology. Just ask Taiwan Semiconductor (TSM) how "unimportant" process technology is as it scrambles to try to get its 20nm planar node out by late 2014.
Finally, at the recent OpenCompute 2013 conference, CEO Paramesh Gopi gave a presentation about how in 2014 the foundries would start moving to FinFETs (i.e. "tri-gate" in Intel-speak):
And we're really excited, because we're not just gonna do this with conventional CMOS processes. This will be the year that we will bring FinFET. 2014 will be the year when FinFET starts to converge with software defined X-Gene based data center components on a chip. What will that do to TCO? Well, from our perspective, we will geometrically reduce TCO with now the addition of the key process technology barriers that take us past conventional CMOS.
Well, now that's interesting. I assume that Dr. Gopi did not mean to imply that FinFETs add barriers, but in fact remove them. But the point still remains: better transistors significantly expand what a processor/system-on-chip designer can do in a particular generation at a given cost/gross margin profile. Also, it's kind of interesting that Applied Micro is misleading investors by claiming FinFETs will be coming out in 2014 from the foundries. I remind everybody that this is likely a 2016 event, based on the fact that Taiwan Semiconductor believes that its 20nm planar node (i.e. non-FinFET) is coming in 2014 (probably late if the timing of the 20nm node is similar to that of the 28nm node):
And enough discussions have taken place with enough customers with large requirements to lead us to believe that in both its first and second year of production, in both the first and second year production of 20 SoC and that first year will be next year, 2014. The second year will be a year after that, 2015.
It is extremely saddening to see Mr. Gopi mislead investors in this way.
Finally, it is foolish to assume that Applied Micro, with its very limited experience at actually designing general purpose micro-architectures, can out-design Intel. Advanced Micro Devices (AMD) hasn't been able to come close in nearly 10 years (after briefly catching Intel off guard in 2003), so it stands to reason that a threatened Intel is unlikely to cede the design prowess to Applied Micro, a company whose "Titan" CPU design (which also had suspiciously amazing specifications and even got to tape-out) went down in flames.
Where Are The Design Wins?
It was widely speculated that Applied Micro would announce a design win with Facebook at the OpenCompute conference, and that Facebook "wants to go all ARM." Well, no, that didn't happen. Applied Micro contributed a few reference designs to the OpenCompute project, and Frank Frankovsky from Facebook congratulated Applied Micro for getting some silicon out. However, that was it. No design wins, just a PR in which Mr. Frankovsky said that Facebook was excited to be able to test the X-Gene (among other SoCs).
While I'm sure that Applied Micro will get some design wins, I remain skeptical that the product is actually any good and will be the panacea that so many who believe in the "ARM In Servers" party line expect.
What To Do About The Stock
Applied Micro has run up too far, too fast. Unfortunately, the hype around the ARM server push is so incredibly strong, that a short position might not work for a few quarters. The timing will be difficult, since short sellers will need a very real downside catalyst. Should such a catalyst actually hit, then the stock becomes a short, but what could those catalysts be?
- If by the end of 1H 2013, there aren't any significant X-Gene design wins announced, then investors are likely to become suspicious and start to short the stock again, especially if Intel decides to roll-out its "Avoton" and "Rangeley" 22nm Atom SoCs for server/networking/communications far before Applied gets its 40nm X-Gene shipping for revenue.
- A less-than-stellar earnings report on January 30th (although it is unlikely as the analysts seem to be quite enthralled with the X-Gene story).
- A less-than-stellar forecast from ARM Holdings at its earnings report on February 5th (also unlikely).
- The company decides to do a nice equity offering at these prices to raise cash.
That's the problem with going short here. As long as Wall Street in general keeps buying the hype from Applied Micro's management, the stock could continue on to a $1B+ valuation (1/5th of the entire cloud server market opportunity by 2014) before the reality sets in. What I personally will be doing is buy puts ahead of the earnings report to minimize my risk. If the "bubble" actually pops and the stock goes back to $6 in a large gap down (or a very steady sell-off as the stock has exhibited in the past after the euphoria of X-Gene hype wears off), then the $7.50 May puts for $0.25/contract could have the potential to be worth $1.50+/contract, or 500%+ upside. I also have a few months to let it play out.
Be skeptical, everybody. When a company that has almost no experience in designing high performance processors goes and makes bold claims such as this:
Further, the company is claiming to integrate absolutely everything under the sun onto a chip on an out-dated process node, while keeping power consumption low, while also keeping costs down. It seems downright suspicious, especially since we have seen no official power figures (the slide above was pulled), no third-party verifications that the thing isn't a slow dog, and no real reason to believe that Intel will not put out a superior product.
Right now, people are drinking the Kool-Aid of anything with the word "ARM" in it and making extrapolations out to infinity. The wave might be fun to ride for now, but ultimately, if Applied Micro so much as hints that their fantasy-land of a perfect, "no compromises" chip on an out-dated process node from a team of 600 may not be as great as originally expected, expect a very swift and painful gap-down. And that's what I'm betting is ultimately going to happen. After all, if it seems too good to be true, it probably is. Nothing in life comes without cost, and in the competition's case, that cost is $13B in capex and $11B in R&D.
Additional disclosure: I may buy some AMCC puts, and I am long ARMH puts.