GS US Semi Asia Tour Notebook Days 2 and
3 - Capex datapoints extremely bifurcated
but it's clear to us that consensus 2006 SPE
estimates need to come way down
Bifurcated semi capex datapoints stood out on Days 2 and 3 of our tour. There were a
number of positive datapoints including likely capex increases from Toshiba (for FY05)
and UMC (FY06), strong CQ3 orders from UMC and Winbond and a bullish outlook from
TEL. On the other hand, there were some very negative datapoints, including likely very
significant capex cuts in 2006 from Elpida, Winbond and Toshiba (for FY06) and
indications from suppliers that Intel?s ?06 capex will be down y-o-y. Also, the lithography
companies which have the longest lead times and therefore best visibility are cautious.
Netting it all out, we feel the capex environment simply isn?t strong enough to support
current consensus ?06 SPE estimates, bullish Street sentiment, and rich valuations so we
remain very negative on the SPE group recognizing that intermittent positive datapoints
will periodically get traders excited.
During the second and third days of our Asia tour, we met with Winbond, UMC, Gemtek,
and Realtek in Taiwan. In Japan, we met with Toshiba, Tokyo Electron, Advantest,
Renesas, Marubun, Nikon, Canon, and Elpida.
While Day 1 of our Asia trip was focused mostly on end-demand related datapoints in the
handset, PC, and digital TV space, days 2 and 3 had more of a semi capex focus, as we
met with a number of the largest capital spenders as well as some of the largest and most
important semi capital equipment suppliers.
The bottom-line is that we remain very negative on the SPE sector as we believe the
negative datapoints outweighed the positive datapoints and we come away with increased
conviction that the 2006 semi capex environment will not be robust enough to support
current Street consensus EPS estimates for the SPE group (which are unreasonably high in
our opinion), bullish Street sentiment (we believe mostly from investors looking to add
beta for a potential CQ 4 rally), and extremely rich valuations on normalized earnings
(which we think will matter a lot if we are correct that Street estimates are too high by an
order of magnitude for 2006). On the last point, we have spent a lot of this trip refining
our 2006 bottom-up semi capex model and it now stands at -5% y-o-y. We find it very
surprising that the Street is modeling approximately 50% y-o-y earnings growth for
Applied Materials and then using that number to make a valuation argument for the stock,
when our bottom-up capex model for 2006 is so clearly flat to down. To that end, a fact
that we find extremely enlightening and one that we believe shows that the Street can
continuously overestimate the earnings power of the SPE companies is that Applied
Materials has guided below the Street consensus for out-quarter EPS for 4 out of the last 5
quarters. As a result, we have no idea why anyone would want to own the stocks until
either the company or the analysts or both capitulate on the real earnings power of the
company/industry.
We have heard the argument that equipment spending as a percentage of total capex
budgets will be higher in 2006 (an argument that gets made every downturn) thereby
driving revenue growth for the SPE companies even in a flat capex environment. We
make two points on this: 1) even if WFE/total capex increases dramatically and drives
15% topline growth despite flat to down capex, we think it is very unlikely that AMAT's
earnings can grow 50%, and 2) We asked the companies we have met with here if their WFE/capex
would be higher and no one said that would be the case. While we believe some U.S. companies
(i.e. Intel and TI) are likely to have higher WFE/capex in 2006, we strongly believe they are the
exception, not the norm and using these companies to try and justify unreasonably high estimates is
incorrect.
Relative to the specific datapoints that we learned on our trip, first on the positive side:
1) UMC is likely to significantly increase its 2006 capex budget from $1.0BN in 2005 to $2.0BN in
2006. Moreover, UMC is significantly increasing orders to the SPE suppliers in CQ3 and will
increase orders sequentially, again, in CQ4. Also, UMC would need to make additional orders in
H1'06 to support a $2.0BN budget. UMC's capex plan was the most bullish datapoint we learned on
the trip and we have a lot more detail on this issue in the body of our note.
2) Toshiba is very likely to raise its FY 2005 capex budget from about $1.55BN USD to about
$1.8BN USD to pull forward previously announced capacity expansion plans for NAND flash.
Tempering this datapoint, however, is the fact that Toshiba has already ordered all of this
incremental capacity from the equipment suppliers and that the higher FY 2005 capex means lower
FY 2006 capex according to the company, with FY 2006 capex likely down about 15% y-o-y.
3) Winbond is making significant orders to the equipment suppliers in CQ3 (we estimate about
$300m USD). Similar to Toshiba, this positive point is balanced by the very negative datapoint that
Winbond is going to reduce its 2006 capex budget by about 40% y-o-y to about $400mn in CY 2006
from about $700mn in CY 2005.
4) The fact that both UMC and Winbond are making significant orders to the equipment suppliers in
CQ3 supports positive datapoint #4, which is that TEL is quite upbeat about its order patterns for
CQ3 and CQ4. While the company didn't provide an official update to its flattish original guidance,
it seems as if the 5-10% sequential order improvement, which Applied Materials suggested for CQ3,
is a reasonable estimate for TEL's CQ3 order book as well. TEL is also hopeful that 2006 capex will
be higher but they admitted that this is not based on analysis, just "gut." As we suggested above, we
believe 2006 capex will be flat to down y-o-y.
While there were a number of very legitimately positive datapoints for the SPE space, we believe
they were more than balanced out by the negative ones:
1) Before we get to the really negative points, we would categorize what we learned about TSMC's
order patterns as an "in-between" datapoint. On the negative side, it seems like TSM won't be
ordering any new equipment in CQ4, so the robust recovery that many were yet again calling for in
CQ4 is not likely to materialize. That said, we do believe that TSMC will begin to increase their
order patterns in 2006 based on what we learned, both about their business and the strategy that they
are communicating to their equipment suppliers.
2) On the definitively negative side, there are a number of very significant capex reductions for
2006 vs. 2005 that we learned of including Elpida (2006 capex likely down 50% y-o-y to about
$650 million), Winbond (mentioned above), and Toshiba (also mentioned above).
3) Maybe most importantly, Elpida's 2006 capex commentary is the first sign that DRAM
companies' capex is actually dependent on the DRAM environment (which as we highlighted in our
Day 1 note, is not very good). The Elpida CEO explicitly stated that the company now needs to
prioritize balance sheet improvement after 3 years of spending capex to build 300mm capacity and
gain market share. The CEO was extremely clear that 2006 capex would be equal to the company's
operating free cash flow, which the company estimates at $600mn USD to $700mn. This datapoint
is extremely significant given the vast number of analysts who have insisted for the last two years
that DRAM companies would spend "no matter what" because "they have to build 300mm." Elpida
further suggested that they won't be making any orders at all to the equipment suppliers "anytime
soon." Elpida's 50% capex reduction in 2006, along with a roughly similar % cut at Winbond should
refocus investors on the risks to 2006 Street estimates based on DRAM capex.
4) Several of the SPE suppliers to Intel suggested the Intel's 2006 capex budget will be down y-o-y.
While this is consistent with our bottom-up model, this is quite different from Street consensus. Not
only would Intel's y-o-y capex declining likely be a blow to estimates, perhaps more importantly, it
would be a blow to Street sentiment, as it was in 2002 when they decreased their capex after
spending "countercyclically" in the first year of that downturn. Recall that the 4 year rolling average
capex/COGS slide that the company showed at its spring analyst meeting and that it has used over
the past 5 years to communicate their preliminary thoughts on its out-year capex suggested that
2006 capex would be down 10-15% y-o-y. While we hesitate to overemphasize that slide (and the
company has suggested that it shouldn't be overemphasized), that slide has been the best predictor
of out-year capex that we know of and that, in conjunction with what we learned from some of the
SPE suppliers on this trip suggests that investors should likely be taking a more cautious approach
to Intel's capital spending.
5) The lithography companies, which have the longest lead times and therefore, we
believe, the best visibility, are cautious. Nikon and Canon suggested that 2005 and/or 2006
estimates for global litho units are too high and both believe that 2006 capex will be flat or down. If
the companies with the longest leadtimes are still cautious, why should we overemphasize the
commentary of many of the U.S. semi equipment companies who are bullish just beyond their stated
visibility. For example, AMAT has about 3 month lead times and Tuesday night, they suggested at a
competitor conference that the recovery they called for on their last earnings call would be pushed
out to late H2'05 or early H1'06. With only approximately 3 month lead times, we question how
clear any outlook that extends beyond that period is?
We will be meeting with companies in Korea on Thursday and Friday, including Samsung SDI,
Samsung Electronics, Pantech, Shinsung, Samsung Electro-Mechanics, LG Electronics, and
MagnaChip. Below we provide detailed overviews of our individual company meetings in Taiwan
and Japan.
UMC - POSITIVE NEAR-TERM TRENDS AND STRATEGY FOR 90NM SHARE GAIN
LIKELY DRIVE SIGNIFICANT INCREASE IN 2006 CAPEX. We had a very productive meeting
with UMC during which we reviewed current business conditions and potential capital spending
scenarios for 2006. While the company won't make a decision on its formal preliminary 2006 capex
plan until November, we came away from the meeting believing that UMC will significantly
increase its capex in 2006 to about $2.0B USD from about $1.0B USD in 2005. We expect the
capacity will be added at the 300mm/90-65nm technology nodes. While current capacity utilization
at the company's 300mm fabs is only an estimated 60% (the company will only say that 300mm
utilization is well below the corporate average of 75%), management believes that the company
needs to add more advanced capacity in order to attract new customers in the graphics, baseband,
and PC networking space. We believe UMC could be trying to court ATI, TI, and Marvell
specifically with this strategy. While adding significant capacity when current utilization on 300mm
is only about 60% could be viewed as extremely risky, if the capacity additions don't attract these
incremental customers, UMC feels that this is a risk worth taking.
One of the reasons we suspect that UMC maybe willing to take such a bold approach to capex is
that near-term business conditions continue to improve with particular strength noted in handsets,
LCD drivers, and PC chipsets with increases in graphics also expected in CQ4. We believe that a
mid- to high- teens % Q/Q shipment increase in CQ3 could be followed by an approximate 5% Q/Q
increase in CQ4, although there is no official estimate from the company. UMC's visibility remains
at about 2 months, as it has been for most of the year, so we would expect that the company will
have a much firmer view of the market for CQ4 when it reports in October.
UMC is currently making orders (which we believe to be significant) to the SPE suppliers in CQ3
and is expected to increase those orders again in CQ4. These orders will result in capacity increases
Goldman Sachs Global Investment Research 3
Analyst Comment September 21, 2005
in CQ1 and CQ2 '06 with the more significant increase coming in CQ2. While the company is
already making significant orders toward its CY2006 capex budget, we believe that if capex
spending comes in at or above the $2.0B USD we are estimating additional orders will still need to
be placed in H1'06 in order to satisfy that budget.
WINBOND - FLATTISH OUTLOOK FOR PSRAM BUSINESS CURRENTLY AT VERY
HEALTHY LEVELS, 2006 CAPEX EXPECTED TO BE DOWN ABOUT 40% Y/Y AS
COMPANY ALREADY RAMPED ITS 300MM FAB IN 2005. We met with Winbond
management in Taiwan in what could be described as a reasonably upbeat meeting. Winbond has
two major revenue segments, with DRAM accounting for 55% of current revenues and logic
accounting for 40%. Flash accounts for the other 5%. Within DRAM, the company has three
products, PSRAM, SRAM, and commodity DRAM. Within the logic segment, the majority of
revenues come from the PC I/O business (this segment was recently augmented through the
acquisition of National Semi's PC I/O business) and Speech ICs for toys. Within the DRAM
segment, PSRAM accounts for 50% of the revenues, with Sharp the company's major customer.
Specialty DRAM (23%) and commodity DRAM (27%) account for the rest.
Relative to market trends, the PSRAM market was described as very healthy with shipments flat
Q/Q in CQ3 and CQ4, off of very high levels in CQ2 at about 51-52K wafers. The company has the
ability to slightly increase capacity for PSRAM to about 60K per quarter, but is choosing not to do
that at this time. Despite the healthy market, pricing was described as flat to down, with Winbond
expressing a willingness to price aggressively to attract customers. Winbond's total DRAM segment
is expected to show flattish revenues in CQ4 vs. CQ3, with PSRAM flat and commodity DRAM
and SRAM down a bit.
Regarding the company's capital spending plans, Winbond still intends to spend about $700M USD
in 2005, of which approximately $500M has already been spent. This capital is being used to ramp
Winbond's first 300mm fab, which is expected to get to 16K WSPM by December 2006. For 2006,
the company is planning on reducing its capex to about $400M USD with the majority of that
capital (about $300M USD) dedicated to ramping the next 8K WSPM in the new 300mm fab. The
company is making orders for this capacity to the equipment suppliers in the current September
quarter with delivery expected in mid-2006 and they do not expect any additional orders beyond
that for the next several quarters. Management did comment that the total cost of the capacity ramp
is about 10% cheaper than they originally expected because of attractive terms being offered by the
equipment suppliers.
TOSHIBA - ROBUST NAND ENVIRONMENT; MIXED DATAPOINTS ON CAPEX. Our four
main takeaways from our meeting with Toshiba are: (1) Toshiba's 2005 capex budget is likely to be
above original expectations of 170 billion yen (~$1.5B USD), although 2006 capex is likely to
decline to 150 billion yen (~$1.3B USD). (2) NAND memory demand remains robust, possibly
outstripping supply to some degree despite rapid bit growth, as price declines and hot applications
like MP3 players continue to drive demand. (3) Management sees a positive trend and outlook for
HDDs. And 4) Notebook and TV markets remain challenging with significant price declines
offsetting strong unit growth.
Regarding the company's latest outlook (updated in Aug-05), the company expects overall revenues
to grow at a 4% CAGR for FY05 (Mar-06) through FY07 (Mar-08). The company hopes to grow
semiconductor revenues at a 9% CAGR, which appears to us to be fairly aggressive given our view
of long-term semiconductor industry revenue growth of 6%, and given the company's commentary
around choppy end market trends and pricing declines, as we discuss below. Toshiba indicated that
it plans to continue to invest in capex to support its growth plans. The company expects to spend
1,100B yen (~$10B USD) from FY05 to FY07 in capex, with ~500B yen (~$4.5B USD) allocated
to the semiconductor business. The company had originally expected its FY05 semiconductor capex
budget to be about 170B yen (~$1.5B USD), but management indicated that capex is likely to come
in a bit higher (perhaps about 200B yen or ~$1.8B USD) as the company is pulling forward capex
from 2006 into 2005 in order to increase its NAND capacity. To that end, Toshiba currently has
about 107.5k WSPM in 8-inch NAND flash capacity and at its 300mm facility (Fab 3) it currently
has 10k WSPM in NAND flash capacity. The company's capex is being used to increase 300mm
NAND capacity at Fab 3 to 21.5K WSPM by the end of 2005. Toshiba also intends to increase its
300mm capacity at Fab 3 to 30k WSPM by CQ2/CQ3 of 2006. Importantly, management indicated
that it has already placed all of the equipment orders to reach its 30k WSPM capacity plan. Given
the company's aggres
sive spending, Toshiba expects NAND bit growth of ~130% Y/Y in FY06 (FYE Mar-07). ASPs are
expected to decline about 30-40% Y/Y in FY2006. Capex associated with the semiconductor
business is expected to be about 150B yen in FY06 (~$1.3B USD), which given the company's
500B yen three-year semiconductor capex budget, would imply a FY07 capex budget which is flat
at about 150B yen (assuming that FY2005 gets raised to 200B yen as we expect).
NAND remains a large and accelerating area of semiconductor revenues for Toshiba. Toshiba had
105B Yen (~$1.0B USD) in memory sales in the June quarter, of which ~80% is NAND. Toshiba
believes overall NAND industry demand is outstripping supply and the company can only supply
about 70-80% of requested demand. We believe supply is tight, but not outrageously so, as spot
prices are stable, not rising, and our checks suggest customers are generally able to access supply.
Price declines in the market remain significant, with an approximate 60% price drop initially
expected in CY'05. However, contrary to Samsung's Jul-05 expectation for a price decline of 15%
Q/Q in Q3 and 30% Q/Q in Q4, given current solid demand Toshiba expects the industry price
decline to likely fall short of 15% Q/Q in 3Q and possibly be limited to a high single to low double
digit Q/Q decline in 4Q.
In regards to the company's HDD business, Toshiba noted continued positive trends and a positive
outlook for its HDD business. The company focuses on the 2.5-inch, 1.8-inch, and 0.85-inch HDD
markets, which are all expected to grow well. Toshiba's positive shipment and increased production
plans benefits Marvell, which we believe continues to be the primary HDD IC supplier at Toshiba.
Toshiba expects to ship over 8.0M HDDs in FQ2 (Sep), including over 3.5M 1.8-inch units and
over 4.5M 2.5-inch units, which is up from 6.5M units (2.5M 1.8-inch units and 4.0M 2.5-inch
units) shipped in FQ1 (Jun). The company continues to be optimistic regarding prospects for
0.85-inch HDDs, noting strong interest from handset manufacturers, which are expected to be
commercialized early next calendar year. Recall that Nokia announced the N91, which features a
4GB 0.85-inch HDD, and we believe other designs at other unannounced handset manufacturers
continue to progress towards production in 2006. The company expects to ramp perpendicular
record technology on the 0.85-inch around FQ2 of next year, which will take capacity points to 6GB
and 8GB. Toshiba continues to ramp production capacity towards ~300K/month, which is
consistent with the plan the company had highlighted during our Mar-05 visit.
Toshiba sees notebook revenue growth as challenging, with revenue growth of ~2-3% Y/Y despite
strong unit growth (~10% Y/Y), as prices continue to decline, while moderating cost reductions in
components such as HDDs and DRAM are not tracking price declines.
Regarding the company's relationship with Xilinx, Toshiba has taped out a couple of initial 65nm
parts for Xilinx, indicating that it hasn't experienced any big manufacturing problems and believes it
is ready to ramp to production depending on Xilinx's schedule. Toshiba also concurred with
commentary from UMC during our meeting regarding the expected production split between the
two manufacturing partners of 70% UMC and 30% Toshiba. Toshiba does not intend to increase its
share beyond 30%.
Regarding the projection/FPD TV business, Toshiba indicated that it is trying to shift production
towards flat-panel displays (FPDs), away from projection TVs, as it believes that the projection TV
market is shrinking (especially in the US) due to continued price encroachment from FPDs. We
believe that the shrinking projection TV market and Toshiba's transition of production away from
this area are negative for TI given DLPs strong position in the projection TV market, including
Toshiba's use of DLP in much of its projection TV product line. Toshiba expects to increase its
position in the FPD market towards a target 13-15% global market share in FY'07 (from its current
share of 6-7%). The company believes global market share of at least 15% will be required to enjoy
healthy profitability in this segment. Toshiba also indicated that Sony, Panasonic, and Matsushita
plan to compete aggressively in this market.
RENESAS - BUSINESS APPEARS TO BE GENERALLY WEAK. Renesas indicated that overall
business conditions are difficult. The company had entered the year expecting a decline in revenues
Y/Y in FY2005 of approximately 3% from FY2004 (ended March 2005) sales of 1 trillion yen
(about $9B USD). The company currently expects revenues in the first-half of the fiscal year to be
approximately 440 billion yen (about $4B USD). While management would not provide an explicit
updated forecast for FY2005 revenues, the company did indicate that sales are likely to come in
below original expectations. Management indicated that weakness is broad based across
geographies and segments with the exception of the automotive segment, where the company's
performance is OK.
The company also commented that all consumer products have been slow, particularly digital still
cameras. In regards to the company's PC business, management indicated that the PC market is not a
significant business for the company. The company did indicate that Renesas has tended to be
strong on the high end in the mobile PC segment, but the overall market has tended to be stronger
on the low-end for PCs.
Management commented on the environment in the MCU segment by indicating that while trends in
flash-on-chip MCUs and smart card MCUs are improving, the company's MCU business remains
below the peak levels it reached in FY2004. Renesas shipped about 77 million MCU units per
month in FQ1 and expects to ship about 84 million MCU units per month in FQ2, which is below
the 90 million MCU units per month the company was shipping in FY2004. Renesas' flash-on-chip
MCU units are increasing to 15 million units per month in FQ2 from 14 million units per month in
FQ1. Renesas' smart card MCU's are increasing to 20 million units per month in FQ2 from 13
million units per month in FQ1. While the company believes that the overall market is weak,
management thinks it is possible that Renesas is losing a bit of market share in the MCU segment
and the company is primarily concerned from a competitive standpoint with Freescale and NEC
(particularly in the Japanese market).
Renesas indicated that it is also struggling in the RF business. The company has historically had a
good position in GSM, but has had difficulty shifting to a new GaAs process and was late to market.
Renesas produced 7.2 million RF units per month in the June-quarter of 2004. However, in the
June-quarter of 2005, Renesas only produced 3.2 million RF units per month.
Regarding the company's 3G chipset venture with NTT Docomo, Renesas is now shipping
evaluation samples of its dual mode mobile handset chipset, supporting both W-CDMA and
GSM/GPRS. Renesas indicated that it currently wasn't trying to go after 3G business outside of
NTT DoComo, but that it would expect to do so later in 2006 and into 2007, as it has now gained
3G expertise. The chipset is using Renesas' SH based application processor, which according to the
company, has gained significant design wins in Asia. However, these design wins have been mainly
with small Chinese players and the company currently does not appear to be aiming for the US and
European markets. Renesas believes that the strongest competitor in the applications processor
market by a wide margin is Texas Instruments' OMAP solution.
In-line with recent commentary from Infineon, Renesas also commented that the chipcard business
environment remains tough. Aggressive pricing from NEC as well as sluggish demand has put
pressure on profitability.
In terms of Renesas' capital spending plans, the company is investing about 80 billion yen in capex
in FY2005, which is down from 90 billion yen in FY2004 (Renesas also
spent 120 billion yen in FY2003). The company expects to spend about 100 billion yen in capex in
FY2006. However, management indicated that it may become more or less aggressive on its capital
spending plans depending on whether or not it sees signs of a recovery as the year progresses
6 Goldman Sachs Global Investment Research
September 21, 2005 Analyst Comment
(particularly in the System LSI business), which thus far has not happened and the company's
confidence in such a recovery continues to decrease. Management further indicated that it has no
plans to place any significant front-end equipment orders in the near-term.
Renesas currently has 15k wafers per month in 300mm capacity and the company's current
utilization rate is at 94% (flat from 94% in the June-quarter and up from 88% in the March quarter).
The company's capacity has historically been divided between about 50% flash and 50% system
LSI, although its current budget is skewed towards preparing for copper production and shrinking
system LSI geometries. Management indicated that if demand were to improve meaningfully, the
company could build out the first floor of its 300mm fab (the second floor is currently fully
equipped), which would require a total investment of about 150-160 billion yen. Note that the
company is able to gradually build out the floor.
ELPIDA - MASSIVE BIT GROWTH PLANS A PROBLEM FOR THE DRAM INDUSTRY; 50%
CAPEX REDUCTION IN 2006 A PROBLEM FOR THE SPE INDUSTRY. Elpida's long-term goal
is to become a top-3 DRAM competitor and the company believes that it has competitive
advantages including: (1) a high proportion of 12" capacity, allowing for large bit capacity per
wafer, (2) a rapid ramp of 90nm 12" product and a significant increase in foundry wafers, and (3) a
focus on memory products for mobile & digital consumer and server end markets. However, as we
discuss below, we have a number of concerns that limit our optimism about Elpida's opportunities
in the still fiercely competitive DRAM market.
We believe Elpida may be underestimating the amount of capex its competitors are spending this
year, and so may misestimate its competitors' bit supply growth. We see the company's estimate of
CY05 industry bit growth of 50% Y/Y as too low, particularly if it achieves its own bit growth
target of 200% Y/Y and given its major competitors' capacity expansion plans. Also, while we
believe Elpida does have good market share in mobile and consumer DRAM, we also expect other
companies to compete aggressively to produce product, given higher margins and premium ASPs,
and we expect higher margins will inevitably decline as a result. Further, we expect DRAM
oversupply in 2006 to drive ASPs down significantly, which would apply, both to commodity and
specialty DRAM products. Finally, while Elpida stated that it has experienced no business,
IP-related, or R&D impact due to NEC and Hitachi's recent sale of their 14% stake in Elpida, and
the two former parent companies had hinted at lowering their ownership stakes, we believe this
action could be indicative of a lack of confidence at the two companies.
Elpida plans to expand its 12" wafer capacity quickly, transition quickly towards 90nm, and ramp
foundry wafer starts. The company looks to drive bit output up 200% from current levels by Jun-06.
Elpida expects to move from 60K WSPM capacity today (comprised of 30K 12" WSPM, 24 8"
WSPM, and 5-6K foundry 12" WSPM) to 106K WSPM capacity by June-06 (comprised of 54K 12"
WSPM, 20K 8" WSPM, and 32K 12" foundry WSPM, of which 25K is expected from Powerchip
and 7K from SMIC). The company noted that it has already placed the orders for the capital
equipment for this expansion. Additionally, Elpida is looking to drive bit growth through its 90nm
transition and noted that it plans to have 50% of its 12" capacity on 90nm by Dec-05. Longer-term,
the company has a 4-stage capacity expansion plan: (1) current ramp to 54K on 100nm and 90nm
process technologies, (2) 16K wafer expansion on 80nm, (3) 15K expansion on 80nm and 70nm,
and (4) 15K expansion on 70nm and 55nm. Next year, the company expects to begin step #2,
ramping only part of the 16K wafer expansion depending on business conditions. Elpida currently
forecasts capex at $600-$700M in FY'06 (down from $1,300M in FY'05), as the company's ability
to spend on capital investments is dependent on its operating cash flow generation as it is limited by
its weak balance sheet.
Elpida believes that it has a unique product portfolio, given its emphasis on server, mobile, and
digital consumer DRAM products, rather than the high-volume but lower margin PC products. The
company believes it can achieve higher margins in these areas, which represented 63% of revenues
in the Jun-05 quarter including 42% from mobile & digital consumer and 21% from server products.
Longer-term, the company hopes to keep PC-related DRAM sales limited to ~30% of sales. Elpida
Goldman Sachs Global Investment Research 7
Analyst Comment September 21, 2005
also believes it holds significant market share in these areas, and estimates its market share at 70%
for mobile DRAM and believes it supplies "most" DTV DRAM products.
CANON AND NIKON - DOWNSIDE TO EXPECTATIONS LIKELY IN THE GLOBAL
STEPPER MARKET. Our meetings with Canon and Nikon suggest downside to expectations in the
global stepper market. Our checks suggest that the global lithography stepper market (in unit terms)
is expected to decline at a faster pace in 2006 than previously thought and 2005 units are also
expected to be lower than originally anticipated. We continue to believe that ASML's 2006
consensus forecasts (calling for EPS of around EUR0.80) have considerable downside bias, at least
towards our estimate of EUR0.57.
Canon stated during our meeting that it expects orders to be flattish Q/Q in the September-quarter
and to increase slightly Q/Q in the December-quarter. While order patterns have slowed over the
past few quarters from its US, Korean, and European customers, each month since April, the
company has seen successive order improvements from its Taiwan based DRAM customers.
Management indicated that Canon continues to enjoy strong Taiwan DRAM orders through the end
of 2005, but the company is concerned that these orders may decrease in early 2006 and noted that
initial indications for orders in early 2006 are not robust. The company is also concerned that
increases in 300mm yields could drive excess supply and hence lower demand in 2006, specifically
in the DRAM segment. As such, Canon expects worldwide capex to decline slightly Y/Y in 2006,
consistent with our bottom-up capex model, which suggests an approximately -5% Y/Y decline.
The company continues to expect 2005 lithography units to be around 450-480 vs. Nikon's estimate
of 430-440 units. Canon also currently expects global 2006 lithography unit sales to be around 400
vs. Nikon's estimate of 400-420, although the company will host a meeting in October during which
it will update its unit forecast. Canon noted that its current orders are being driven by both i-line and
KrF tools. The company expects the industry to ship 200 i-line units in 2005, of which Canon
expects to ship close to 100 units. Management also noted that lead times are currently around 6-8
months.
In regards to laser sources, Canon indicated that the decision as to whether to use GigaPhoton or
Cymer lasers is made by its customers. In the domestic market, Canon noted that customers tend to
prefer to use GigaPhoton, while Cymer has a stronger worldwide market share. Customers that
prefer GigaPhoton lasers tend to do so because GigaPhoton's lasers are less expensive and have
lower maintenance costs. From a technical perspective, Canon believes that the quality of
GigaPhoton and Cymer lasers is fairly similar for both KrF and ArF. The majority of Nikon's ArF
tools (except the 306 machine) are utilizing a laser from Cymer given GigaPhoton's product
limitations for tools with a NA of less than 0.85. Our conversations suggest that ASPs to Cymer
have recently remained relatively stable.
Canon noted positive trends in the LCD business, with tot
al LCD stepper units expected to increase Y/Y in 2005. At the beginning of the year, industry LCD
units were initially expected to decline about 20-30% Y/Y in 2005 and are now expected to increase
10% Y/Y, with Canon expecting to grow more than the industry. Canon's 2005 FPD unit forecast is
about 115 units vs. 85 units in 2004. The company does not yet have a 2006 unit forecast for the
LCD business.
Management highlighted that the lithography market has shrunk from peak units of about 1200 in
2000 to about 600 units at the peak of the most recent cycle in 2004. At the same time, the average
ASP of lithography units has increased from about $4.8 million in 2000 to about $8.8 million in
2004. Canon still expects the lithography industry to show no Y/Y revenue decline in 2006 based on
ASP increases, but admits that this assumption partly depends on how many (higher priced)
immersion tools are shipped in 2006.
The company currently expects a global immersion lithography unit market of around 30 tools in
2006. Canon is sticking by its previously given expectation to enter the immersion lithography
8 Goldman Sachs Global Investment Research
September 21, 2005 Analyst Comment
market with a 1.3NA tool introduction in early 2007, while Nikon expects to have a 1.3NA tool in
Q3/Q4 of 2006. ASML expects to launch its 1.3+NA tool in Q2/Q3 2007. Nikon appears to be on
track to deliver its 1.07NA immersion tool offering to customers in Q4 2005, and expects to ship a
total of 10 immersion tools in 2006.
Nikon commented during our meeting that it expects the global lithography unit market to be around
430-440 units in 2005. This is slightly lower than the company's forecast of 450 units earlier in the
year. While Nikon is sticking to its own shipment forecast of 185 units (35 ArF) in 2005, we believe
there is a downside bias to that estimate based on less aggressive patterns at DRAM customers and
the foundries. We also believe that the company has recently stopped competing for business for
some unprofitable tool lines at the low-end, given Canon's recently aggressive pricing behavior.
Nikon expects semiconductor capex to be flattish Y/Y in 2006. As we noted earlier, Nikon expects
the industry to ship 400-420 lithography units in 2006 and the company aims to gain considerable
market share from ASML in 2006. Like Canon, the company noted positive trends in the LCD
business. We estimate that that this part of the business continues to deliver above average margins
for Nikon.
Nikon management noted that it has recently seen some push-outs for i-line tools from second-tier
logic customers, which appear to be somewhat mitigated by pull-ins for i-line tools by a blue chip
customer. The company's NAND customers remain relatively aggressive in their order patterns,
while business with MPU customers has remained relatively stable. Nikon noted that DRAM
activity is relatively subdued based on limited customer penetration in this end-market.
TOKYO ELECTRON - OVERALL STABLE TO IMPROVING SPE ENVIRONMENT, WITH
THE DIRECTION OF 2006 CAPEX DEPENDENT ON DRAM INVESTMENT. Tokyo Electron
(TEL) indicated during our meeting that the overall SPE market remains stable. Specifically, the
company noted that the CPU market is good, DRAM companies continue to invest, NAND flash is
strong, and the foundries are still cautious although the company expects Taiwanese foundry
purchasing to increase gradually in Q4'05 and Q1'06 as the foundries continue to invest in
technology-driven capacity additions. TEL also indicated that it expects the order environment to
improve gradually over the next few quarters.
In regards to 2006 capex, TEL expects capex to be up Y/Y in 2006, although management believes
DRAM investment stability will be a driving factor in the Y/Y direction of capex. Specifically
regarding DRAM investment, the company indicated that Taiwanese DRAM customers remain
optimistic that DRAM ASPs will not go any lower through mid-2006. The company believes that
DRAM customers continue to monitor DRAM prices, and if prices go lower (which is consistent
with our view that DRAM prices will be weak through 2006), DRAM customers may have
difficulty funding incremental capital investments, despite their desire to add more 300mm capacity.
We would highlight that our current bottom-up 2006 capex model implies about a 5% Y/Y decline
in capex.
TEL noted that DRAM bookings tend to be lumpy on a quarterly basis as DRAM customers are
making large investments (an order from a DRAM customer for one fab on a quarterly basis tends
to be about $100M). Management also highlighted that DRAM customers continue to aggressively
move capacity over to NAND flash. To that end, TEL believes that NAND flash demand remains
quite robust.
In regards to the foundries, management indicated that TSMC is getting more positive, although it is
not making significant new investments as it is focused on managing profitability and cash flow by
keeping utilization rates at higher levels. TEL management also indicated that UMC is expected to
place more meaningful orders in the near-term, which is consistent with UMC's commentary during
our meeting. Interestingly, Tokyo Electron believes that foundries are currently investing differently
in capex than they have in previous cycles, with an increased focus on providing customers with
strong technology as opposed to simply supplying significant amounts of capacity.
Recall that Tokyo Electron, like Applied Materials, has a flat panel display (FPD) equipment
business (which accounted for about 25% of TEL's CQ2'05 total orders). The company had
expected the FPD business to decline about 40-50% Y/Y in 2005, but the business hasn't been as
weak as expected and the company now expects the FPD business to decline about 20% Y/Y in
2005. Management noted that this business continues to be better than expected in the near-term.
On the competitive front, Tokyo Electron acknowledged that it lost a few points of market share in
the etch segment, we believe driven by Lam Research penetrating portions of a few new accounts.
Management indicated that since TEL introduced a new etch process chamber, the company has
taken back some market share in etch leading to an essentially unchanged overall competitive
environment in this segment, with TEL having lost some share and then gained some back. The
company additionally mentioned that it continues to evaluate the single-wafer cleaning market and
is considering a more aggressive market entry, initially for Front-end of the Line (FEOL)
applications. TEL's commentary underscores our view that SEZ's (U/N - covered by Matt Gehl)
early market share lead in the single wafer clean market is assailable, especially as the technology
gains more acceptance in FEOL applications.
TEL also indicated that consolidation across the SPE segment would be a positive, as it would allow
companies to provide a common service platform and gain greater critical mass (the company
believes that the SPE market is difficult for companies that have less than $1 billion in annual
sales). TEL management, however, stated that it does not believe in simply acquiring companies for
greater scale without providing shareholders with greater profitability.
The company also provided some insight on the wafer probe card market, as Tokyo Electron has
competed in this business for many years, although it remains a tiny business for the company. TEL
indicated that FormFactor is the leader in the wafer probe card market, although many companies
are trying to develop solutions in this segment given its high margins and greater overall stability
relative to the broader SPE segment.
Management touched on the Chinese semiconductor market and indicated that it expects Chinese
semiconductor companies to continue to invest primarily in used wafer fab equipment, which TEL
believes can be a profitable business for the company (TEL would benefit from refurbishment and
field services sales).
ADVANTEST - DRAM AND FLASH MEMORY TESTER SALES LIKELY TO REMAIN
ROBUST IN 2006. Adv
antest highlighted several drivers of intermediate-term growth during our meeting. Specifically
regarding the outlook for 2006, Advantest expects the demand for DDR2 to be strong in 2006, as
DDR2 becomes mass produced (management believes that about 30% of DRAM will be DDR2 in
2005, which will increase to about 60% in 2006). On the SOC side of the business, the company
expects the SOC business to be flattish Y/Y in 2006 (we believe driven by a decline in business at
Intel offset by increased business at new customers). The flash business is expected to remain
robust in 2006. Our view is that the significant increase in DRAM and NAND flash capacity in
2004 and 2005 that has driven and is likely to continue to drive very robust bit growth, is likely to
lead to continued memory tester sales for Advantest into 2006.
Advantest highlighted that it has gained market share in both the memory and SOC test segments. In
memory test, Advantest had approximately 70% market share in 2004, up from about 61% share in
2003. In the SOC test segment where Advantest offers the T2000 platform, the company increased
its market share to about 17% in 2004, up from about 11% in 2003. Advantest is targeting about
20% market share in the SOC test space, which the company believes it can achieve partially by
gaining traction with US fabless companies. Recall that Advantest's offering in the SOC test space is
the T2000, which is a tester that is open, easily upgradeable, and easily transferable (i.e. can be
easily modified using different modules to test various kinds of devices).
As we highlighted in our note on Monday from the first leg of our Asia tour, ASE Test indicated
10 Goldman Sachs Global Investment Research
September 21, 2005 Analyst Comment
that while it is in favor of more open and modular test platforms (like Advantest's T2000), it had yet
to see any significant customer demand for the T2000 platform. Advantest noted that the platform
currently has two customers (Intel and Toshiba), and the company expects to gain additional two
customers by FY2006. These two customers are likely to be in the graphics and communications
space, which would imply that the testers would be purchased by package and test houses.
Last year Advantest shipped 200 units and expects to ship about 1.5x more units (primarily to Intel)
in 2005. The T2000 platform has had significant business at Intel, which has been the primary
customer of the platform, and we walked away from our meeting encouraged that management
expects to have continued traction at Intel as Intel swaps out its old testers for T2000 testers.
In regards to the memory test business, management stated that the transition to DDR2 has been
slower than expected. The company indicated that wafer prober sales tend to lead back-end final
test demand by about 3-6 months, and given that wafer prober demand has been robust, the
company believes that its memory tester orders are likely to increase in H2'05 vs. H1'05. On NAND
flash specifically, Advantest noted that Hynix has aggressive plans to buy NAND flash testers.
Samsung, however, is able to use its old DRAM testers to test NAND flash and hence is not
investing aggressively in new NAND flash testers.
Advantest derives about 40% of its revenues from flash (NOR and NAND) and about 60% of its
revenues from DRAM. The company had originally expected flash revenues to decline as a
percentage of revenues to 20% in FY2005, but now expects that flash may be stronger than
expected and hence about 30% of revenues in FY2005. The Y/Y decline in flash as a percentage of
revenues in FY2005 is being driven by strong NOR flash sales (we believe at Intel) in FY2004.
While Credence has recently won NOR flash business at Intel, Advantest does not appear terribly
concerned with not having won that business, as the company does not believe that NOR flash is a
significant growth segment of the market.
On the wafer probe card front, the company has a high end wafer probe card that is being evaluated
at customer sites, although wafer probe cards currently do not contribute to the company's overall
sales. Management indicated that it believes FormFactor is the primary competition in the wafer
probe card segment. Advantest believes that its wafer probe cards will be able to offer customers
high accuracy and speed. While the company believes that some of final test may shift to the wafer
level, management believes that unit growth from 300mm is robust enough to drive increased final
tester sales that would offset any declines that could come from more of final test moving to the
wafer level.
In regards to consolidation in the back-end segment, particularly in light of Agilent's plan to spin-off
its semi test business, management indicated that Advantest is not focused on M&A solely to
increase its product offerings. Rather, the company is focused on adding technology or addressing
customer needs through M&A activities. Our sense from speaking with management is that
Advantest is therefore not likely to purchase Agilent's semi test business, as it would require
cannibalizing either Agilent's business or Advantest's T2000 platform and the company does not
need Agilent's technology.
Management also commented in regards to the significant increase (about 40% M/M) in back-end
orders reported in the US Semiconductor Equipment and Materials International (SEMI) August
book-to-bill data out on Tuesday evening, that it believes the strength in the data was likely driven
by wirebonders as opposed to testers.
REALTEK - BROAD BASED STRENGTH, INCLUDING IMPROVING GIGABIT ETHERNET
TRACTION. Our two main takeaways from our meeting with Realtek are: (1) Realtek has begun to
see traction for its gigabit Ethernet controller, though Broadcom and Marvell continue to be leaders
and Agere's presence remains limited, and (2) Realtek expects a strong Q3 on favorable seasonality
and broad-based strength across product lines with the company's Q4 outlook conservative given the
potential impact of weaker seasonality.
Goldman Sachs Global Investment Research 11
Analyst Comment September 21, 2005
Realtek is a fabless IC design company, with major product areas including Fast Ethernet
controllers (40% of H1'05 revenues); Ethernet switches, gateways, and routers (17%); gigabit
Ethernet and WLAN products (7%); audio codec ICs (22%); and LCD controllers (7%). Realtek
expects significant growth in Q3, driven by broad-based product strength, with revenues expected to
be up 20% Q/Q in Q3 and down 10% Q/Q in Q4 due to seasonality. The company also expects
margins to expand, with gross margins expected to be more than 45% in Q3 vs. Q2 gross margins of
42% and Q1 gross margins of 38%. Gross margin improvement is expected to be driven by better
yields, favorable wafer pricing, and product mix. Q4 gross margins are expected to decline slightly
from Q3. Realtek expects revenue growth to continue into 2006, with 10-15% Y/Y growth expected
to be driven by gigabit Ethernet, LCD controllers, and WLAN. Longer-term, Realtek expects to
realize revenues from its development efforts for DVR and DTV IC solutions. Realtek is currently
investing heavily in this area, with more than 250 engineers (out of a total of ~750) focusing on
DVR/DTV efforts.
Realtek has been relatively late to market with its gigabit Ethernet solution. However, the company
is experiencing improved momentum and a strong unit ramp in 2005. Realtek shipped 3.4M gigabit
ethernet units in 2004, 3.5M units in 1H'05, and expects to ship 5.0-5.5M units in H2'05 with the
monthly run-rate in August at more than 1M units. We believe Realtek's traction has primarily been
with motherboard manufacturers, which continues to be led by Marvell and has seen increased
competitive pressures from Broadcom and Agere. We believe Marvell and Broadcom also continue
to experience solid trends related to gigabit Ethernet, as PCs continue to transition to the advanced
technology (Realtek estimates that gigabit penetration currently stands in the 30% range, though
market research reports have suggested crossover occurred earlier in the year). From a technology
perspective, we believe that Realtek remains somewhat behind with its PCI Express solution
expected to begin shipping in September. We expect gigabit Ethernet pricing to remain aggressive
(consistent with expectations), with Realtek's pricing currently in the $2.20-$2.30 range, vs. Marvell
and Broadcom, which we believe are in the $2.50-$3.00 range for cost-optimized (feature limited)
motherboard sockets. We continue to expect overall gigabit Ethernet pricing to continue to decline
in 2006 as volumes ramp, with the potential for pricing to trend towards $2.00 by year-end 2006.
GEMTEK - WIFI STRENGTH ON FAVORABLE SEASONALITY, CENTRINO PRODUCTION
RAMP; RALINK AND MIMO ALSO PROGRESSING. Our four main takeaways from our
meeting with Gemtek are: (1) Among WiFi chipset vendors, Broadcom maintains its solid position
with Gemtek, however the market continues to be competitive. (2) Intel Centrino dominates
traditional mini-PCI module (used in notebooks), which is a high volume but low-margin business
for module vendors. (3) Consumer applications are the focus of growth opportunities for WiFi. And
(4) WiFi chipset vendors continue to price products aggressively. MIMO technology is advancing,
and pricing is 1.5-3x current 11a/g and 11g, respectively, but 802.11n continues to face delays and
is likely not to be significant until mid-2006.
Gemtek is a major ODM for WiFi LOM modules, with Intel (with WiFi modules for Centrino) a
significant customer, as well as WiFi modules for the retail market, with major vendors including
Linksys and Belkin. Chipset suppliers to Gemtek include Intel (for the Centrino module), Broadcom
(75-80% of non-Centrino-based products), Atheros (15% of non-Centrino-based products),
Conexant (small and declining), and Ralink (small and increasing).
Gemtek expects its WiFi LOM module business to increase solidly in H2'05, driven by favorable
seasonality and a ramp in Centrino modules for Intel at its China factory following execution
hiccups in H1'05. The company expects revenues to increase 15% Q/Q in Q3 and 20% Q/Q in Q4,
as Centrino units ramp from very small volumes in H1'05 up to 500k-600k units/month currently,
and to ~1M units/month in the peak period in October/November. Gemtek's strength related to
Centrino is driven by share gains, as Intel is believed to be transitioning business with
manufacturing partners, in particular Solectron. Gemtek expects its overall unit volumes to grow
solidly in 2005 to 18M-20M, up from 12M in 2004. Margins remain challenged due to Gemtek's
operational issues in 2005 as well as continued price declines and Intel's pricing power. Gemtek
hopes to increase sales in consumer electronics and MIMO 802.11g, which are expected to have
12 Goldman Sachs Global Investment Research
September 21, 2005 Analyst Comment
higher ASPs and margins.
Broadcom maintains a solid position at Gemtek with its 802.11g product. The majority (~80%) of
Gemtek's unit volume in 2005 is expected to be 802.11g and Broadcom supplies about 90% of
Gemtek's 802.11g chipsets. However, emerging Taiwanese competitor Ralink has been increasing
its presence at Gemtek through aggressive pricing (40-50% discount) and advanced technology
(MIMO 802.11g). Broadcom maintains a strong position with OEMs. Going forward, we expect
WiFi vendors to partially offset ASP declines by ramping next-generation WiFi products, in
particular MIMO solutions, including 802.11n. We believe that Broadcom has an 802.11 "pre-n"
solution sampling now, with a volume ramp expected in H1'06. Pricing for the "pre-n" solution is
expected to be in the neighborhood of $30, which is about 3x the current 802.11g chipset price.
MARUBUN - SEASONAL HANDSET STRENGTH, FLATTISH INVENTORY TREND. Our
main takeaways from our meeting with Marubun are: (1) Marubun is seeing strength in 3Q in
Japanese 3G handsets, consistent with seasonal handset strength and TI's positive mid-quarter
update on broad-based strength. (2) The company is experiencing fairly broad-based strength,
though there are certain pockets of weakness, in particular driver ICs (due to an inventory
workdown) and digital still cameras (in-line with expectations). And (3) inventory is expected to be
fairly flattish Q/Q (vs. typical seasonal building), but the trend is masked due to weak pricing
(memory) and a certain inventory workdown (display drivers).
Marubun is the biggest distributor in Japan and the biggest distributor for TI, both in Japan and
worldwide (with TI products accounting for 45% of Marubun's 44.0B yen Electronic Devices
segment revenues in the Jun-05 quarter). Marubun is also the biggest distributor for Freescale in
Japan (with Freescale products accounting for 5% of Marubun's 44.0B yen Electronic Devices
segment revenues in the Jun-05 quarter). We believe Marubun is seeing continued positive trends
related to TI's products, with fairly broad-based momentum, while Freescale-related trends remain
fairly stable, which is pretty consistent with overall business expectations for both TI and Freescale.
Other major suppliers include Samsung (15% in the Jun-05 quarter), Seiko Epson (10%), and
Maxim, On Semiconductor, and Philips (3% each). Major customers include Sharp and Mitsubishi
(10% each), NEC (6%), and Kodak, Sony, and Canon (4% each).
We think Marubun's commentary suggesting that Japanese 3G handsets are experiencing seasonal
strength in 3Q is consistent with seasonal handset strength and with TI's positive mid-quarter
update. Marubun distributes TI's OMAP applications processors and general analog products for
cell phone applications, in particular to major customer Sharp, and we expect TI is benefiting from
the 3G strength.
Marubun is also experiencing strength in Bluetooth with a new (non-TI) supplier into handsets.
Additionally, the company is seeing strength this quarter in DSPs, "special-use ICs", and analog
ASSPs, particularly those incorporated into 3G handsets, and is also seeing incremental growth with
GE-PON (gigabit Ethernet passive optical network) telecom infrastructure products. Areas of
weakness include handset driver ICs (due to inventory workdown related to particular customers'
handsets) and embedded NAND flash into DSCs (as Japanese DSC unit growth is slow, in-line with
expectations from our Japanese research team).
Each of the analysts named below hereby certifies that... |
