GS: Asia Tour Notebook ? Cycle very likely to
get worse before it gets better
We attended meetings with numerous supply chain, semi and semi equipment companies
in Japan and Taiwan this week. Our key takeaways from the first part of our Asia tour
include: (1) We believe the cycle is poised to get worse before it gets better and the
Japanese equipment companies agree, as they are more cautious on the ?05 outlook than
US equipment companies, (2) We are slightly reducing both our bottom-up and top-down
?05 capex forecast based on discussions during the trip, (3) DRAM makers continue to
invest based on the assumption that DRAM pricing will rebound sharply in CQ2, which
we believe is a very faulty assumption, (4) Q1 back-end orders appear likely to be quite
weak, and (5) NAND and LCD related orders are robust. Our global semiconductor team
is hosting a conference call for clients today at 10AM EST to discuss our findings
(domestic dial-in: 877- 208-2954; code: 1411).
We completed the first leg of our tour of Japan and Taiwan this week. We met with
numerous supply chain, semi and semi equipment companies in both regions, including
Canon, Tokyo Electron, Nikon, Elpida, Renesas, Ibiden, Toshiba, Fujitsu, Lite-on,
Compal Communication, BenQ, World Peace Industrial, and ASE. We will provide
takeaways from the next leg of our tour (during which we will meet with TSMC, UMC,
Nanya, Powerchip, Gemtek, RealTek, MediaTek, Sunplus, and Amtran) in our weekly
note out on Monday.
Several members of our global semiconductor team (Andrew Root, Jim Covello, Seogju
Lee, and Matthew Gehl) will be hosting a conference call for clients on Thursday, March
31st at 10:00 AM EST with more detail on our checks in Asia this week. The dial-in
information for the conference call is:
Domestic dial-in: 877-208-2954 International dial-in: 973-528-0056 Conference Entry
Code: 1411
Domestic replay: 800-332-6854 International replay: 973-528-0005 Replay Code: 1411
We provide our key takeaways from the first leg of our Asia tour below.
WE BELIEVE THAT THE CYCLE IS POISED TO GET WORSE BEFORE IT GETS
BETTER AND THE JAPANESE EQUIPMENT COMPANIES SEEM TO AGREE, AS
THEY ARE SIGNIFICANTLY MORE CAUTIOUS ON THE 2005 OUTLOOK THAN
THE US SEMI EQUIPMENT COMPANIES. We come away from the first leg of our
Asia tour more confident in our view that the semi equipment cycle is poised to get worse
before it gets better for four main reasons: (1) Most of the customers that are currently
being aggressive with their 2005 capex plans have front-half loaded budgets and,
consequently, are set to have declining orders in H2'05.
According to our checks on the
trip, Intel, TSMC, UMC, Toshiba and most of the DRAM makers all have front-half
loaded budgets. (2) While DRAM orders appear to have remained quite robust in CQ1,
Japanese companies are skeptical about continued orders from the DRAM makers (with
the exception of Samsung), as DRAM makers are currently investing based on an
assumption that DRAM prices will rebound aggressively in CQ2/H2'05, which seems
quite unlikely given the current supply-demand dynamics in the DRAM industry. (3) Japanese
company managements do not seem optimistic that logic foundries are likely to begin ordering in
the near term due to already low (and declining) utilization rates. Several Japanese companies
commented that the logic foundries request quotes from the equipment suppliers every month, but
no orders have materialized yet and hence the equipment companies are now becoming more
skeptical that the orders will materialize anytime soon. And (4) Another issue that the Japanese
equipment company managements raised during our meetings that causes us to maintain our
cautious view is the fact that several Japanese chipmakers have built 300mm fabs for which they
don't seem to have a long-term need. For example, Fujitsu is ramping up its new 300mm fab to 10K
wafer starts per month. The problem that appears to be emerging is that the main use of this fab
seems to be foundry capacity for Lattice Semiconductor and Transmeta, which we don't view as a
viable long-term alternative for this capacity. Another example is the 300mm fab at Matsushita,
which we understand will be used primarily to manufacture LCD drivers. We believe that LCD
driver demand for one company doesn't support the need for an entire 300mm fab. In short, a bit of
a hangover seems to be emerging from the last upturn when customers added capacity that they
really didn't need over the long term and the Japanese company managements seem to be
recognizing this as a problem that may take several years to fix.
This week, we have been surprised by the stark difference between the bullish views currently being
offered by the US equipment company managements compared to the much more conservative view
being offered by the Japanese company managements who universally believe that orders will get
worse before they get better in this cycle. In general, it seems as if the biggest difference between
the US equipment company managements (who remain extremely bullish and are insistent that the
cycle will ramp in H2'05) and the Japanese equipment company managements (who are calling for
at least several more quarters of deteriorating orders) is that Japanese semi equipment company
managements are more skeptical about what their chip customers are telling them while the US semi
equipment company managements are taking what their chip customers are telling them at face
value. For example, as opposed to the US companies, which are taking what their DRAM customers
are telling them and using that as the basis for their bullish view on the market, the Japanese
companies are doing their own analyses on the likelihood of continued orders from these customers.
It seems as if the Japanese companies' own analyses are leading them to a significantly more muted
conclusion regarding the outlook for the remainder of this year and into next year.
WE ARE REDUCING OUR BOTTOM-UP 2005 GLOBAL SEMICONDUCTOR CAPEX
FORECAST TO -2.5% YEAR OVER YEAR FROM ESSENTIALLY FLAT YEAR OVER YEAR
PRIOR TO OUR TRIP. WE ARE ALSO LOWERING OUR TOP DOWN CAPEX FORECAST TO
-10% YEAR OVER YEAR FROM FLAT TO -5% YEAR OVER YEAR. We are lowering our
bottom-up 2005 global capex forecast to -2.5% year over year from our essentially flat
year-over-year capex forecast prior to our trip due to lower than expected CY2005 capex estimates
from Toshiba and Renesas. The specific capex data points related to the front-end that we have
learned thus far on our trip are noted in the table below (back-end capex commentary is in a later
section):
Capex 3/05 Capex 3/06 Comments Billion Yen
Toshiba 203 140-150 70% in H1'05; 75% of Yokkaichi fab already ordered Elpida 130 100
Ordering aggressively now Fujitsu 50 90 Ordered most of'05 budget but still increasing orders
Renesas 94 94 Ramping 300mm fab to 10K WSPM by 3/06 Source:
In addition, we are now more cautious on our top-down forecast, which we are lowing to -10% year
over year from flat to -5% year over year prior to our trip. We are lowering our forecast as, based on
conversations this week, we expect that capex announcements from the other Japanese chipmakers
during the April reporting season (recall that most Japanese companies have a March fiscal year end
and will therefore likely provide official FY2005 capex forecasts in April) as well as potential cuts
from some of the second-tier logic foundries and some of the second and third-tier DRAM makers
(although we view this as less likely than the Japanese and logic foundry cuts) create downside risk
to our current bottom-up model. Our 2005 earnings estimates are already significantly below the
Street consensus but we believe this likely downward bias to capex estimates creates tremendous
risk to Street consensus earnings estimates, which are modeling a significant ramp in H2'05
earnings.
Our more cautious stance on 2005 capex is supported by estimates we received during the trip from
the Japanese lithography makers Canon and Nikon, both of which are now assuming that 2005
lithography units decline about 20% year over year to about 450 units. Note that Nikon formally
reduced its 2005 unit estimate to 450 units from its previous 500-unit estimate earlier this week in a
press release. Canon further believes that 2006 litho units will decline significantly again to 400
units (a view we share) while Nikon believes 2006 litho units will be flattish compared to 2005.
Again, we highlight that even this more optimistic view offered by Nikon is significantly worse than
Street consensus estimates, which are calling for a massive ramp in 2006 earnings estimates after
the already strong ramp projected in H2'05. We further note that cautious indications by the
companies that sell the longest lead time equipment do not bode well for the US equipment
companies that generally sell shorter-lead-time equipment and are therefore l
ikely to get more cautious indications from their customers well after the litho makers receive more
cautious indications (which they are clearly now getting as evidenced by Nikon's recent unit
estimate cut).
DRAM MAKERS APPEAR TO CONTINUE TO INVEST BASED ON THE ASSUMPTION
THAT DRAM PRICING WILL REBOUND SHARPLY IN CQ2/H2'05. Checks during our trip
indicate that Elpida, Hynix, Powerchip, and Promos all continue to be extremely aggressive adding
capacity under the assumption that the DRAM price declines are over. We think this is a critical
point as, to the extent that this sharp hoped-for reacceleration in pricing is instead a continued drop
in DRAM pricing in CQ2, we believe (and most of the Japanese equipment companies seem to
agree as we highlighted above) that there is significant risk to the DRAM orders in CQ2/H2'05.
We were truly shocked to learn during our meetings that at least some of the DRAM companies are
still assuming supply bit growth of only 45% for full year 2005 when most of the DRAM companies
achieved almost 30% bit supply growth in CQ1 alone. We seem to be in a situation in the DRAM
industry where every company is assuming they will grow their own bit supply by 60%-80% but
none of the companies believe that the other DRAM companies will be able to achieve the same
level of bit growth. We believe this is an extremely nanve assumption on the part of the DRAM
makers and one that will lead to an even worse pricing environment as we move throughout the
year, as most of the DRAM makers will likely be surprised by the level of bit supply growth at their
competitors.
Q1 BACK-END TESTER ORDERS ARE LIKELY TO BE WEAK. Our checks this week indicate
that the back-end package and test houses ordered very little to no capacity in the first quarter. We
also believe that the IDMs continue to order at very low levels given current low utilization rates.
To that end, we believe that Q1 tester orders are likely to be weak. While the Street view has been
that tester order rates are unlikely to continue to decline and are likely to pick up in H2'05, our
checks indicate otherwise as utilization rates are likely to remain at low levels in 2005 and the
outsourced test houses are focused on being disciplined about adding capacity. ASE, for example, is
cutting its capex significantly in 2005 to $350 million (about $100 million for test) from $780
million in 2004. The company did not order any capacity in Q1 and expects to order either no
capacity again or a very small amount of capacity in Q2. The vast majority of the company's capex
budget is left to be spent in H2'05 and management indicated that it is very unlikely to spend the
official budget if customer indications are not better than expected in H2'05. Short lead times on
testers allow the company to react quickly to changing customer demand and hence ASE intends to
wait to add capacity until/unless business conditions are improving significantly. ASE's tester
utilization rate is in the 70% range, with utilization rates tighter on the leading edge, and
management expects its utilization rate to improve about 5% in Q2.
In terms of back-end equipment company specific data points, we learned that Agilent's 93k tester is
regarded as the best at testing PCI Express chips. We believe this has negative implications for
non-memory test equipment makers like Teradyne and LTX, which are relying on the PCI Express
transition to drive new tester sales. Additionally, we learned that a package and test house has
received six of Credence System's Sapphire testers but has only "unpackaged" two of the systems
(in other words, it has only installed two of the six machines). This has two important implications
in our view. First, back-end test equipment makers continue to ship tools to customers that aren't
even being utilized (and likely won't even be paid for until/if they are utilized). Customers already
have excess capacity on hand from their installed base and didn't want/need the "free" (or
"consigned") tool in the first place, which if/when eventually utilized will be additional capacity that
will reduce the need for new incremental tester orders. Second, our checks indicate that the
Sapphire platform (which Credence Systems acquired via its purchase of NPTest in early 2004) is
considered an unproven platform and has yet to gain significant traction with customers. We view
the slow adoption of the Sapphire platform as a negative for Credence, as we believe that the
company is relying on the platform to help it maintain a position in the SOC test market.
NAND RELATED ORDERS REMAIN EXTREMELY ROBUST AND ARE EVEN
ACCELERATING DUE TO CONTINUED TIGHTNESS IN THE NAND SEGMENT; LCD
ORDERS ARE ALSO STRONGER THAN EXPECTED AS TAIWANESE FLAT PANEL
MAKERS' SPENDING PLANS APPEAR MORE AGGRESSIVE THAN ORIGINALLY
EXPECTED. Not all of the data points we learned on the trip were negative as several NAND
customers (including Samsung) appear to have been pulling in shipments at the end of the quarter to
try to alleviate tight supply-demand dynamics in the NAND flash segment. While we expect
NAND-related orders to remain extremely robust, we do believe there is some risk that everything
NAND-related (including the end-products driving the NAND tightness in the first place) is being
double or tripled ordered right now due to supply chain constraints. To the extent that end-demand
for the I-Pod shuffle and other flash-based MP3 players is anything but extremely robust, there is a
risk that NAND orders to the equipment suppliers would decline in H2'05, as demand for other
NAND-driven devices (most notably digital cameras) appears to be quite choppy according to
checks during our trip. However, in the mean time, we expect NAND-related equipment orders to
remain extremely robust and we expect that the US equipment company managements will highlight
this area of strength as the main driver behind their bullishness on the upcoming earnings calls.
In addition to NAND related equipment orders, the other notable area of strength that was
consistent thus far on our trip is LCD equipment. LCD equipment orders are stronger than had been
expected as the Taiwanese flat panel makers are being more aggressive than original expectations.
Expectations had been that flat panel capex would be down by as much as 40% in 2005, but this
number now looks to be closer to -20% to -25% and could seemingly move even higher than that.
I, Jim Covello, hereby ce |
