Its been a while since I’ve looked at Intel’s NAND business and how that slow motion train wreck is proceeding. I figure this week’s as good as any and besides my notes need distilling.
Last month, Intel announced a two-year, $7 billion investment plan for its U.S.-based manufacturing facilities.
As far as I can tell, not a dollar of these billions is going to NAND. No surprise there. No need to throw good money after bad.
A year ago March, Intel’s NAND problems became public when it disclosed that its money-losing NAND business would knock two points off its gross margins. Analysts were stunned given that Intel’s flash business generated such a small portion of Intel’s overall revenue.
Three classic articles from that time. The titles say it all:
10 March 2008 Why Intel Should Dump Its Flash-Memory Business
05 March 2008 Sell More NAND, Watch Margins Drop
Consensus- Intel would be far better off without this “rotten” business.
Intel’s CEO, Paul Otellini, vowed to “fix” its NAND business. He promised that NAND wouldn’t “be a drag on Intel Corp.”
Easier said than done. Otellini’s solution to Intel’s low margin commodity conundrum? – SSDs. I wasn’t convinced then and even less now.
Given all this, many concluded that Intel’s flash business was up for sale last year. The question was and still is, who would want to buy it?
That line of speculation, is best left for another post- too many angles.
So how did Intel back itself into this corner?
Kicking off in 2006, Intel was late to the NAND game. Their IM Flash Technologies (IMFT) JV with Micron caught the tail end of the party. To my mind Intel could have picked a less squirrelly partner, but that’s my bias showing.
In any case, to form IMFT, Intel kicked in its NVM IP. Micron transferred it’s NAND designs and ownership of its Lehi Utah fab and the vacant half of its Virginia Manassas fab to Intel. Intel also seems to have taken ownership of Micron’s 200mm Boise fab.
Last but not least both Intel and Micron contributed funding.
This contribution balancing seemingly has given IMFT cover for patent protection through Intel’s cross licensing with SanDisk, but that’s another story. Suffice it to say that I’m not convinced and don’t think that SanDisk is either.
So how much has IMFT cost Intel? A lot.
According to SEC documents, Intel’s portion of IMFT costs was approximately $300 million in 2006, $790 million in 2007 and $1.1 billion during 2008. Looking at it another way, as of December 27, 2008, Intel’s maximum exposure to loss approximated its investment balances- which were $1.7 billion in IMFT (number from Intel).
On the technical front, give IMFT credit. Starting far behind, they’ve moved from 72nm to 50nm to 34nm. 34nm is leading edge stuff, but if Lazard is right, IMFT is currently having “quality issues.” My guess is that these quality issues revolve around MLC. The timeframe fits and multilevel issues increase with each shrink.
MLC hasn’t been IMFT’s strong suit. IMFT didn’t even start ramping MLC until early 2008.
IMFT rolled confidently through 2006 and 2007. NAND demand looked like it would go up forever.
While its hard to believe today, apparently in 2007, Intel and Micron were planning on opening a 300mm wafer fab every year. Next up after Lehi was to be Singapore.
IM Flash Singapore, LLP (IMFS) was formed in February 2007- again a JV between Intel and Micron. So far at least, this has not been a happy story for Intel which appears to have dropped another few 100 million dollars with only a fab shell to show for it.
Ground was broken in Singapore in 2007 on schedule for an opening in 2008. Today the fab facility structure is substantially complete, but the tooling and production ramp are on hold.
Intel and Micron keep saying that “the timing” of the fab has simply been pushed out, but it is a bit more serious than that, as this 31 October 2008 article from Singapore indicates:
The fab’s 800 employees have been laid off and at least from a local perspective, plans for the fab “scrapped.”
This IMFS fab was to have been roughly the same size as the Lehi Utah fab- running 60K starts/mo of 300mm 34nm wafers. To get there, apparently all it will take is another $3 to $4 billion. Small wonder Intel has cooled to the idea.
In the fall of 2008 IMFT announced that it would shutting down its Boise 35K starts/mo 200mm NAND fab, which is no longer competitive. Intel took it on the chin once again, booking a $250M asset impairment charge which included a cash payment to Micron of $24 million, and other cash payments of $7 million.
So where does this leave IMFT today in terms of NAND fab capacity?
Probably at about 100K starts/mo – 70K starts/mo at Lehi and 30K+ starts at Manassas. And both are close to fully utilized.
Curiously there is no mention of Manassas on the IMFT web site. IMFT is described as Lehi only.
For comparison’s sake, before idling capacity, SanDisk/Toshiba was running at 260 300mm wafer starts/mo- fab 3 was running at 150K starts/mo and fab 4 was running at 110K. Fab 4 has only been ramped up a bit over 50%. When fully utilized Fab 4 alone will be capable of 210 wafer starts/mo.
Eli summarized IMFT’s challenges well a while back: It is tough to be number 4 in a rough neighborhood.
Back in 2005, before IMFT even officially launched, Apple cut a deal with Intel and Micron for NAND supply. Apparently Apple prepaid $500M to secure NAND, no doubt at best pricing.
The idea was that Apple would secure the supply it needed for its flash-based iPods, while Intel and Micron would secure a long-term contract which purportedly would smooth fluctuations in the NAND market.
It appears the deal worked wonderfully for Apple. As NAND prices collapsed, Apple emerged the big winner- profiting all the more each leg down. On the other hand, the NAND fabs, including IMFT, have suffered and then suffered some more.
Apparently Intel has finally said enough is enough. According to SEC filings, in Q4 2008 Intel cancelled its supply deal with Apple and returned the remaining $167M to Apple (out of its $250M share).
Whether this was Intel panicking at the darkest hour, or simply needing supply for its own purposes, or something else entirely, remains to be seen. In any case NAND prices appear to have bottomed in December.
According to Dan Amir of Lazard (31 March notes), Apple has two new iPhones on the way:
“Our checks confirm that two versions of iPhones will be introduced this year. Production of the new phones should start at the beginning of April and will ramp in May. We believe the two new versions will be announced in June — a low-end and a high-end version. The low-end version may lack WiFi and a few other features that the high-end version will have, such as video capability, a better camera and 32GB of NAND, but it will still be 3G. We believe it is possible that the two phones will be aimed at different regions. The high-end version is expected in North America and Europe, and the low-end version may be for the BRIC countries or China only.”
Amir believes that:
“NAND prices should move up further off the Apple orders. We believe that with the new iPhone design which could include 32GB NAND, Apple is setting the stage for another potentially big flash order starting in April. With the potential for 3M units to be shipped in 2Q, we believe the magnitude of such an order could enable prices of NAND components to move up an additional 20% in the 1H of 2Q. We expect the price for the 16Gb (the main component for the iPhone) to hit $4.50 in 2Q
If Apple is not longer getting NAND from Intel, this puts all the more pressure on other suppliers, which in turn should help keep NAND prices strong. Good news for SanDisk, and in a round about way, good news for Intel, even if it won’t be supplying any of those Apple orders.
End Game Strategy?
Last year after Intel announced its Q1 fiscal 2008 earnings, CEO Paul Otelllini hinted that the its NAND flash business was up for sale. Here we are a year later and Intel is still in the NAND business, maybe reluctantly, but in the business nonetheless.
From my perspective, not much has changed, other than Intel positioning itself to limit the downside. Expansion has been put on hold. Unprofitable production halted. And capital infusions minimized.
In its conference calls, Micron has noted that Intel is locked into operating the IMFT/IMFS JVs for the long term. Looking at Intel’s SEC filings, this appears to mean 2016- but subject to prior termination under certain terms and conditions.
With Singapore on indefinite hold and Lehi and Manassas close to full utilization, Intel’s additional NAND capital commitments appear manageable. Micron is in no position to force an aggressive ramp, given its rapidly deteriorating balance sheet.
No buyers for Intel’s NAND business seem to have materialized.
If they don’t, I suspect Intel’s exit strategy is just to hunker down in Lehi for a few years and then take the write downs and call it a day while exploring other options for NAND supply.
In February of this year, Intel’s CEO, dropped new hints about where its NAND strategy might be going, sans IMFT:
“It may not be essential for us to have our own NAND factories to build (flash memory). We could probably specify the product that we want and buy it from third parties”
At the time most assumed that this would mean that Intel would pass along its NAND designs to a contract fab such as Taiwan Semiconductor. While this may be true, it could also mean that Intel simply could pick up supply from existing NAND players.
In any case, its time to close this post. This one has gone on long enough.
Given the challenges of 2x NAND and the expensive next gen fab equipment required, I suspect that 3xnm NAND is going to be with us longer than anticipated. If so, multiple bit technology beyond MLC (2 bits/cell) is going to be critical.
With X3 and X4 SanDisk/Toshiba is wonderfully positioned. Intel has a problem. It’s struggling with MLC, while SanDisk/Toshiba is transitioning to X3.
Rather remarkably SanDisk’s 32nm X3 chip is smaller than Intel’s 34nm X2. More bits for fewer dollars.
SNDK X3 32nm die size- 113 sq mm. Intel’s 34nm MLC X2- 172 sq mm.