(c) copyright View from Silicon
Valley, 2005. All rights reserved.
On the back of Dell's numbers, the SOX went back up over 400 last
week (up 11+ points on Friday). There is another round of bloviating over the near-term future of the semiconductor
business. Various Wall Street and industry research analysts are touting an alleged decline in inventory as an “insider”
indicator that end-product sales will soon steepen their upward trek.
Unspoken in these breathless claims is the trade-off
between unit sales and average selling prices. For those of who think profits are somehow useful in determining the
advisability of herding back into tech stocks, this would be useful information. Sadly, few of the touts include such
details in their so-called "analysis."
So let's do a little digging on our own and take a tour through a few categories
of semiconductors.
CPU: Intel famously is touting margin expansion and sales growth in their
numbers. Reported by Intel but missing from the reports about Intel is their inventory. Despite reporting top-line
sales growth and the margin expansion, which accountants are allowed to claim when fabs are full, Intel also grew inventory 7%. They are touting high-end Centrino sales growth but have yet to explain how China's $350 or India's $250 sweet spot
for PCs is driving consumption of $600+ Centrinos.
Foundry: Capacity utilization fell at TSMC (78% down from 84%) and UMC (63% down from 72%) in 1Q05. Such low rates leave lots of room for growth but
generate little profit (if any).
One dynamic recently in play is chip companies are placing wafer orders "now"
for the upcoming Christmas 2005 /Chinese New Year 2006 season. The timing is wafer-in during April/May becomes wafer-out
July/ August, which is end-product assembly August/September in time for ~October /November shipments into US retail outlets.
This
might cause a 2Q05 "pop" in the foundry business. (Even with this "pop" the utilization rates may still be quite low.)
The hidden story here is chip makers more-or-less feel they have to place these orders, regardless of how strong they really
believe Christmas might turn out. Managers can get fired for not being ready for a market spike but having the
same over-inventory as everybody else when the spike doesn’t appear, gives them a better chance of keeping their
job. (Does this scenario sound familiar to anybody?)
DRAM: Prices are off
40%(!) within the last 90 days yet no alarm has been sounded. Even with DRAM manufacturers converting DRAM wafer
starts to NAND left and right, a 32Mx8 DDR400 DRAM is down from $4.00 to sub-$2.35 since February. 90 days is
not just a blip. It's long enough for Tier-I DRAM users to lower their contract prices to these levels. With prices down 40%+, units have to go up 70% just to keep top-line revenue flat. Is anybody
assuming 70% DRAM bit growth in 2005 Semiconductor growth stories? Gartner is openly predicting a 2006 crash in
the DRAM market.
One prominent rationale for ignoring DRAM softness is the forthcoming replacement of DDR with
DDR2 (aka DDR-II). DDR2 will focus on 512Mb chips and a 533MHz 64Mx8 is currently running ~$6.00. Clearly, $6.00
per chip is a much better ROI than $2.35 for most fabs. Unfortunately, this price premium is exactly the flaw in
the conversion proponents’ argument. While 512Mb of DDR400 is ($2.35 x2=) $4.70 there has to be another incentive
to convert to a $6.00 solution.
Performance benchmarks tend to show DDR400 is only slightly outperformed by DDR2-533.
DDR2 does save power vs. DDR, which is why it is showing up in some notebooks. DDR2 doesn’t much outperform DDR
until you get to 666MHz. The problem is the DDR2-666MHz premium over DDR400 will be even higher than +27% premium
for DDR2-533MHz.
Anticipating the next argument, also note there is a ton of DDR capacity in Asia. Even wholesale
conversion to DDR2-only by Samsung /Micron /Hynix /Infineon /Elpida may not be enough choke off global DDR supply and increase
prices or force migration to DDR2.
NAND: The other prominent rationale for ignoring a 40%+
DRAM price decline is NAND. Everybody loves NAND & there will not be enough to go around any time soon.
(Say it three times while clicking your heels.) Everybody seems OK with 1Q05 NAND prices being down
~25% year-over-year since it was more than offset by +66% unit volume increase. The issue I want to point out here
is NAND was only $2.074B in 1Q05. Given SIA/WSTS reports of $55.3B 1Q05 total semi sales, NAND is only 3.75% of the 1Q05 market (up from 3.71% in 1Q04). Read that again, a 66%
unit volume increase achieved only a .04% (0.0004) increase in market share. Granted an $8B to $10B market "is better
than a sharp stick in the eye," as my grandmother used to say, but how much is NAND actually growing as a share of the semi
business?
Gartner forecasts CY05 NAND sales will see a further unit increase of +141% against “only” a
-51% decline in per-bit prices. This nets out to ~+18% NAND dollar sales growth. (Gartner’s 2006 forecast is +105%
units accompanied by another(!) -45% drop in prices, which nets out to +13% dollar growth.) If the 2005 unit increase
is underachieved by even the slightest degree, prices will drop further than -51%. Even if volume growth hits +141%
in 2005, net NAND profits may be harder to find.
In fairness, compared to a “crash” in DRAM prices, double-digit NAND dollar sales
growth, even at just break-even, may not be such a bad deal. Sadly, break-even may be a stretch as additional
suppliers and capacity come on-stream...
Keep in mind, Gartner counts sales which SIA /WSTS misses and is among the more optimistic
research houses overall. Their top-line forecast is +3.5% growth from 2004's $220B to $227.5B in 2005.
However, nobody can precisely predict the NAND volume soon to come pouring out of Hynix, ST Micro and Micron as
they each ramp into volume production on top of die shrinks and additional 300mm capacity coming on-stream from Samsung and Toshiba /SanDisk. The NAND market-share wars are just getting started.
Conclusion:
At the end of the day (or at least at the end of this rant), it’s a question of trusting the forecasts, ours included.
“The trouble with forecasts” is they are often self-serving. SIA, Gartner, Wall Street, industry publications such as EETimes and many, many
others have expressed opinions. Don’t read just one. Figure out how the people generating the forecast
data get paid. If something doesn't seem to make sense, trust your instincts and plan accordingly...
