Intel (INTC) is the largest manufacturer of semiconductor in the world and has spent most of this week at Computex in Taiwan telling the technology world about their new Haswell processor and future mobile processors.
When I was a kid I enjoyed assembling jigsaw puzzles to the point that I would turn the pieces over to the blank side and assemble them with only the piece outline as a guide. The Intel, Micron (MU), Apple (AAPL), and Samsung (OTC:SSNGY) story is a first degree jigsaw puzzle.
Let's try to assemble this puzzle. The number of the puzzle piece will be noted in the article.
1. Unexplained massive over-capacity at Intel
2. Apple needs to move all semiconductor manufacturing away from Samsung
3. Apple needs to cut off Samsung from NAND memory
4. Possible Samsung IP problem in A series chips
5. Apple needs a secure source of NAND
6. Apple needs to move the "A" chip business in order to maintain their technology edge
7. Apple helped save Elpida as a source for mobile DRAM
8. The industry will need a few billion more 128Gb NAND chips for the solid state drive revolution
9. The memory industry might not be able to keep up with future demand
10. High performance microprocessors have hit the "memory wall".
11. Intel is dependent on the memory manufacturers to supply enough product to support sales of at least 300 million CPU processor chips plus 4-500 million mobile application processors.
12. Virtually every DRAM memory chip connects to either and X86 device or an ARM based mobile application processor.
13. Virtually every NAND memory chip connects to either an X86 chip or an ARM based mobile application processor.
Apple would like very much to not be dependent on Samsung for any critical component (2, 3) for all the iProducts. Since Samsung has something of a stranglehold on Apple in the form of NAND memory (5), apple is not able to move their "A" chip production away from Samsung. A significant complicating factor might have to do with some essential Samsung Intellectual Property (4) intertwined in the "A" chips that Samsung might not be willing to transfer to another chip manufacturer. A long shot possibility might be that Apple has to abandon in-house application processor design due to the Samsung/IP issue discussed in the link above.
Since, by the time that Bay Trail becomes available, it might be the best performance/power choice for the Apple iPad products. By the end of the year the Intel Merrifield might demonstrate superiority over the latest "A" chip for iPhones, leading Apple to choose to use the Merrifield for the iPhones on merit (6) as well as the IP problem (5) with Samsung. Suddenly, the debate over Intel building ARM based "A" chips would be over. Just as suddenly, a great deal of the Intel excess capacity would be utilized (1).
The PC and tablet trend toward thin, light, and long battery life virtually requires the use of solid state drives in place of hard disc drives in all notebook and tablet PCs. Intel, in a miserable year, will ship at least 300 million PC CPU chips. Since the memory industry will slide into a shortage position on the NAND parts that are essential to the continued success of Intel, it would be reasonable to expect Intel to take matters into their own hands (8, 9, 11). This would involve assigning one of the Intel ground-up huge new fabs to the production of NAND memory (1). To ensure that the NAND memory business never slid into a commodity business model, Intel, in the name, and fact, of performance, could include the SSD in the CPU package (12, 13). Thus, the SSD business associated with the Intel CPU would be captive to Intel. These SSD would have to be tiny to fit into the CPU package. By tiny, I mean eight 128Gb NAND chips stacked on top of controller chip for total footprint of about 150 sq. mm, just about the size of the Haswell chip sitting next to it. For those who needed more than a 128GB SSD, Intel could mount two NAND stacks in the CPU package. NAND SSDs built by Intel would use up the rest of the excess capacity. Since the SSD would cost about $1 per GB, the drive would cost $128, or about the same average selling price of the Intel CPU chips of today. To finish these "Compute Modules", the DRAM in 4GB stacked assemblies could be added to the CPU package (10), thus overcoming the memory wall (10).
With a Compute Module as above, Intel's content per PC would increase from approximately $100 to approximately $300++. That would quickly soak up the excess capacity that I have previously estimated at some multiple of that required for the CPU business alone. This capability could also apply to the mobile segment, which would more than double the Intel dollar content in mobile appliances.
The impact on the PC manufacturers would be to allow resources to be spent on product differentiation rather than running around cutting deals on NAND and DRAM that does precisely nothing to enhance the computing experience.
Primarily because of the almost critical need for solid state drives to mate up with the power efficient Haswell chips, the required volume of new capacity for NAND memory is mind boggling and the time frame for the capacity to become available is unbelievably short. In fact the capacity would have to be in place now in order to solve the problem (1). The volume of NAND parts, assuming and average SSD size of 192GB would be 3.6 billion 128Gb chips. Assuming Intel would build chips on a 14nm process (OTCPK:LINK), a 128Gb chip would be about 120 sq. mm, about 400 good parts per wafer, or about 9 million wafers. A $5 billion dollar, 100,000 wafer start per month giga-fab can produce 1.2 million wafers per year. My personal opinion is that Intel will have at least two excess fabs of this scale by the end of this year. Even this amount of capacity is less than 30% of what would be needed for a complete conversion to solid state drives in PCs.
To be reasonable, not all PCs will convert to SSDs immediately and there is some additional NAND capacity that could become available from IMFT, the Intel/Micron memory joint venture. Micron has about 40,000 wafers left to get to full capacity at the Singapore plant and Micron has alluded to the possibility of converting at least one 100,000 wafer start per month factory from DRAM to NAND. Even with this, the NAND capacity will only be half what will ultimately be needed.
Conclusion: truly light, thin, and low power Ultrabooks are not going to use HDDs, there is no extra NAND capacity available from the traditional memory manufacturers. The only massive unused capacity in the entire industry resides at Intel. Intel must go into the captive memory business as a matter of self- preservation. Paul McWilliams, the remarkably accurate technology writer of Nextinning thinks an Intel acquisition of Micron is an outside possibility, he says, "A wide-eyed speculation would be a tighter coupling between MU and Intel, with the potential of INTC actually buying MU, and with that, re-entering the memory markets."
When we turn the finished puzzle over we see a picture of an Intel dominant in not only in the PC world but also dominant in the mobile business as well. The picture includes an Intel in charge of their processor generated memory requirements through direct manufacturing or an outright purchase of Micron/Elpida. We see an Intel with $20 billion worth of memory and $30 billion worth of mobile business stacked on top of the existing $53 billion of PC and server business. At the scale of $100 billion in revenue, gross margins can decline to 50% and still produce a net margin of 25% or $5/per share. The amazing thing is rate at which that this change must occur.
After a three year, $35 bullion CapEx spending spree, the capacity is in place at Intel. Haswell and Silvermont are in place, Apple might not be able to continue with in-house designed processors due to intermingling of Apple and Samsung intellectual property. Micron is waiting to close on the Elpida acquisition and would then make a sensible Intel buy at $20 per share.
The next 12 months will witness unbelievable change in the semiconductor industry.