I think TCO is a big deal for big clouds--they will save a lot of money on power consumption with a room full of SSDs vs a room full of HDs in addition to not needing as many drives and getting far better performance.
But put together what Lam said in their most recent CC about costs--
Martin B. Anstice - Chief Executive Officer, President and Director
No, I think having a conversation around NAND is just fine. I mean, I -- there's a decent component for the industry of conversions, right? I mean there's about 0.5 million wafer starts to conversion likely in NAND flash this year and that's a lot of spending, that's approximately $3 billion of spending.
Message 28905940
with what Luczo said about amount of storage needed--
Our industry shipped 100 exabytes of data five years ago, 400 exabytes in 2011, and we’ll probably ship a zettabyte sometime between 2015 and 2016. A zettabyte is equal to all the data that’s been digitized from 1957 through 2010. Everything, however you want to think of it, cards, tapes, PCs, mainframes, client/server, minicomputers – one zettabyte. And we’re going to ship that in one year.
....
Q: Someone said to me if you tried to replace all world’s drives with flash, today – your 400 exabytes – that there’s not even enough flash capacity in the world…
A: It would be like 25% of that. And that would mean not building any phones or iPads or anything else. And what’s the revenue opportunity for that? 25% of the drive market is $10 billion. So you’re going to turn over every fab in the world for $10 billion, even though they’re generating probably $150 billion doing the other stuff? Think of it the other way though – if I am trying to address 400 exabytes of data, and ok we admit the whole world’s not going to go that way – like we admit its only 20% of the notebook market – so the notebook market is 125 exabytes, and 20% of that is 25 exabytes; that means I need four of the most recent Samsung Fab 16s – the world’s largest memory fab – four. That’s $50 billion. And by the way – it has to be up and running today.
Q: What’s the total global capacity of magnetic media now?
A: Again, 400 exabytes, which is why units don’t matter. You can increase units, but they you’re taking a big exabyte hit. Here’s the real question – if you’re trying to grow exabytes, what’s the capital investment to grow exabytes. For flash, it is $10 billion at a chunk, which is what it takes to build a new chip fab. For drives, it’s a fraction of that. If the industry had to grow from 400 to 500 exabytes, we’d do that on $1 billion of capital.
Q: To get to a zettabyte, how much additional capital is that?
A: About $1 billion a year for us, and then about $1 billion a year for WD. Versus $50 billion. Here’s the more interesting question. I don’t think we can get to 10-20 zettabytes. I think around 2016, we start running into a sustained storage on meeting exabyte demand unless somehow people start getting aggressive on capital. And the only way people are going to get aggressive on capital is if there is some stability on margins.
Question--Lam says it costs $3b to convert half a million NAND wafer starts; how much storage does that half a million wafers represent? Assuming MLC chips.
I have no idea off-hand how to compute that, but I'm sure someone reading this must. Or at least know someone who would know (Shlomi? Dan?).
Whatever the answer may be, I doubt very much if Samsung's $2.3b investment in their new fab in China that is slated to begin production next year will be anywhere near enough to create a glut, even if it may be enough for them to increase their market share somewhat.
One more datapoint--my favorite reverse indicator, the estimable Mr. Berenbaum at MKM, just lowered his rating on Micron to Neutral from Buy, even as he increased his price target to $13 from $11. This makes me more bullish on Micron, especially after they close the Elpida deal. |
