Another recurring theme in my articles has been the tremendous overcapacity of Intel and the possibility that the capacity could be harnessed to manufacture the NAND memory chips that will be used in the $40-billion Solid State Drive (SSDs) revolution.
Of course these articles are some speculation, but informed speculation. As is the nature of speculation, much of it turns out to be dead wrong, some turns out to be partially right, and occasionally it turns out to be dead on. When dead on, informed speculation makes for great investment opportunities.
So far, Apple has not given Intel the Ax business and Intel has not gone into the memory business. In the same sense, Apple has not given Ax business to TSMC (TSM) or Global Foundries.
Apparently, the market analysts at IC Insights are beginning to think along the same lines as I am. IC Insights is not just another rumor mill; these guys have great credentials, and actually get paid for their research reports and speculation. So, here's an article of theirs for us to discuss.
IC Insights seems to agree with me and many others that an Intel/Apple deal to build Ax chips would be a win-win deal for both companies. It would be a good use for the Intel Fab 42 in Chandler, Arizona, that is currently in limbo. IC Insights points out all the standard benefits: Intel needs to fire up Fab 42, Apple would benefit from the Intel technology, etc., etc., you can read the article for the details.
A major difference with this article, again written by credible guys, is the suggestion that Apple might be interested in buying NAND memory from Intel directly or through the Intel/Micron (MU) joint venture.
What the article also DOESN'T mention is the desirability of getting mDRAM from the same source as the foundry for the Ax chips. In the iPhone and most other smartphones, the Application Processor (AP) and the mDRAM are produced as a Package on Package (PoP) assembly for size and performance reasons.
There are only two companies that can support both the AP foundry work and the memory necessary to make the PoP assembly required. It is not TSMC, and it is not Global Foundries. It is either Samsung (OTC:SSNLF), or some arrangement including Intel and Micron, or the Intel/Micron joint venture. We know that Apple is trying very hard to make Samsung a non-supplier, so the single company left that can supply everything for Apple is Intel.
The other thing that the IC Insights article DOESN'T mention is the need by Apple for a discrete LTE solution. So far, Apple has not designed an integrated Ax chip with on-board LTE. Every single Apple mobile device that connects to the cellular network needs a discrete, not integrated, LTE chip.
Now, I am not privy to what is coming up at the Mobile World Congress later this month, but if Intel doesn't formally announce the XMM 7260 LTE chip, I will probably join the "Dark side" with Seeking Alpha author Ashraf Essa.
Here we are: AP, LTE, NAND, mDRAM, all being available directly or indirectly from Intel. No foundry can provide all of these pieces, not even Samsung. Samsung would be missing the LTE and, of course, even if it had an LTE chip, getting Apple to use it would take a very good sales job.
Let's see what all of this is worth to Intel and Apple. I'm going to leave the mDRAM out, since Micron (Elpida) already has the lion's share of that business.
|Apple Prod||Note||Volume mil||Price||Market B$||Wafers req'd|
|1||NAND 128 GB||270||$9.40||2.54||56K/mo|
Note 1: Average of 24 GB of storage.
Note 2: Average of 64 GB of storage.
Note 3: 30% LTE attach rate.
Note 4: Average SSD size of 192 GB
So, with a total marriage of Intel and Apple, the business upside for Intel is about $16.8 billion. Realistically, it is about $9.19 billion in APs and LTE chips. The NAND is about $7.62 billion, and would have to be supplied by some arrangement with the Intel/Apple joint venture.
I think the interesting thing here is the approximate number of wafers required to service the Apple business. To service the NAND part of the business for Apple would require about 165,000 wafers per month.
Where would they come from?
Speculation: There is 40,000 wafers per month of capacity left to be built out in the Micron Singapore fab. There is an Intel fab in China that is a candidate for shutdown that could be re-fit for NAND production at about 80-100,000 wafers per month. NAND is not the problem at the moment for Apple, since it seems to be getting all the NAND it needs, and probably quite a bit of it from Samsung. So the NAND transition could be over time.
The Logic (APs and LTEs) wafers get very interesting. Using 28nm node calculations, we come up with the need for 68,000 wafers per month. That is about $11,000 revenue per wafer and exceeds the capacity attributed to Fab 42.
But Intel would produce these parts at a 14nm node or better. At 14nm, the number of chips (of the same number of transistors) would only require 17,000 wafers per month. That would be less than half of the estimated 40,000 wafers per month that Fab 42 is said to be capable of. For Intel, that would be about $44,000 per wafer. With a $5,000 wafer cost, for Intel the Apple business, under the right circumstances, could be an amazing 88% gross margin business.
For this, Apple would get all the benefits of Intel's 14nm process and pay no more money than it is paying to Samsung.
Now, what about the more advanced nodes for Samsung and TSMC?
First, Samsung. According to this piece, Samsung can't yield 20nm anything and is now out of the running for the Apple A8 chip. I can't vouch for the source on this.
If the above is true, it would mean that TSMC is the only one left to support Apple on a process that Nvidia (NVDA), at least, thinks is "worthless". This piece is nearly two years old, and therefore, could be suspect, however, I have not seen a retraction of any kind or kind words from Nvidia for TSMC. I have maintained for a long time that I feel TSMC will never ship anything of commercial value at 20nm or below. That doesn't mean that it won't ship 20nm products; it does mean that if it does, they will be leaky, no performance gain over 28nm, and will be very low-yield and expensive.
Is Intel not only the best choice for Apple, but could it now be the only choice?
The only missing piece is the LTE chip. If the XMM 7260 appears at the Mobile World Congress, the Intel package for Apple will be complete.
Is Intel holding 14nm production, waiting for this drama to play out with Apple? We know that Intel has shipped 25,000 design samples of the 14nm Broadwell chip, so 14 nm must be working.
I think Intel is a buy and 2014 is the year.