View from Silicon Valley- MEMORY INDUSTRY UPDATE, February, 2005
Denali published very insightful, but sparsely reported, facts about the semi industry.
They very diplomatically the various +/-5% growth forecasts could be exceeded, in either direction, but
the actual data points the report all point the same way...
Most-relevant seem to, "both NAND and NOR flash will have to ship more than 70 percent more bits
in 2005 to match 2004’s revenue numbers," and "Foundries mostly reported... declines sequentially and (in) outlooks
(in wafer starts, pricing and profit margins)" from already-reduced run-rates.
Bit growth and unit growth must never stop if they hope to offset the
irresistible price pressure from all the new capacity coming on-stream. A collision between an unstoppable force and
an irresistible object is always "interesting."
* * * * MEMORY INDUSTRY UPDATE Semiconductor
Market Outlook
We won’t dwell long on discussing the high degree of uncertainty and variations in market forecasts
for 2005. Suffice it to say that the outlook is “concentrated” between “up 5 percent” and “down
5 percent”, for the total chip industry in 2005 vs. 2004’s ~$213B revenue.
Such a forecast is surely
“by gosh and by golly”, as no one can really put their arms around the driving forces of demand, which is complicated
by a weakening US dollar (the metric in which the chip industry measures itself ), and the dark forbidding specter of price
competition, especially in memories, which can conceivably undo any revenue growth in all other markets combined. Cell phone
growth is said to be strong or weak; PCs are said to be just the opposite—weak or strong. Inventories, as they usually
do, have hung around for longer than thought.
This is a problem common to all downturns and softening markets: no one
can really count inventories, and count them accurately, at every stage of production and consumption, nor can they agree
on what the proper amount is, and, understandably, since the flywheel of chip production has significant inertia, manufacturer’s
have a strong reluctance to slow down production and (thereby) increase unit cost. Few chip-makers cut production unless there
are clear signs (1) that there is excess in the line and in the channel, and (2) that it will be there indefinitely. Such
is the situation today—inventories did not come into balance nearly as fast as some early forecasts indicated they would,
e.g. Winbond’s just-released January results noted a sequential decline in sales of about 10 percent, attributed it
in part to inventory hangover from 2004.
Within the past two weeks, Toshiba, NEC and Hitachi all reduced their profit
outlook for their fiscal year ending 31 March, for their entire corporation, noting the weakness in their chip sector’s
business.
Profit margins suffered in 4Q04 (see memory company financials in the article below), leading to adjustments
in labor force, business withdrawals and a significant sobering up after a 6-8 quarters of good growth out of the trough of
2001-02. Of all companies reporting this past two or three weeks, only ARM Holdings was optimistic about its outlook. Second
place in the “positive outlook race” was “flat in 1Q, hopeful for 2Q”, but not really believing in
the strengthening in 2Q05, since these companies could provide no reason for optimism except guessing that the inventories
would be worked off by then—despite a weak order book for 1Q05 today. Foundries
mostly reported acceptable and profitable results for 4Q04, but with declines sequentially and outlooks (in wafer starts,
pricing and profit margins), which were decidedly lackluster out into 2005.
Remember that a steady and
flat revenue run rate from 4Q04 compared to 3Q04, because of the strong ramp during 2004, means that 2005 would be up about
3.5 percent with stable sales, so any forecast for 2005 that is less than 3.5 percent implies an inflection point downward
at some point in the year—if it has not already happened, since for the who industry 4Q04 came in at $55.1B,
already down from 3Q04’s $55.6B. Whether this turn down in 4Q04 is a brief respite or the onset of another downturn
is anyone’s guess. Pundits have already trimmed their 1Q05 estimates, setting expectations for another sequential
quarterly decline. But for the whole of 2005, smart forecasters line both sides of the road, and the landscape changes daily.
Though
“official” 2005 total semiconductor forecasts are blocked between about +/- 5 percent, compared to 2004, Denali
believes the real spread is much wider, both on the upside and downside, perhaps as large as +/-20 percent—it all depends
on price stability and demand strength. Though we enter 2005 with the vectors pointed the wrong way, we’ve been surprised
before, and there are several cards that remain to be played that have important consequences: reversal of the dollar weakness,
stronger economies in Europe, more difficulty ramping 0.11µ line geometries, leaving supply growth constrained, further restraint
in vendor CapEx for 2005, etc.
On the memory side, bit demand continues to be pretty good, but there’s still
downward price inertia from the DRAM price declines over the past 3-4 months, and, with major ASP declines in 2H04 now the
new “pricing baseline”, both NAND and NOR flash will have to ship
more than 70 percent more bits in 2005 to match 2004’s revenue numbers—even
if prices stay put, which is also unlikely.
There is something new afoot in 2005, as well. Several DRAM makers can
swap DRAM capacity for NAND flash capacity (Samsung, foremost, but also Infineon, Micron, and Hynix), which can take pressure
off DRAM supply growth and prices and move it to NAND flash. There is a lot of common tooling between the two products, and
the larger constraint appears to be established market position (how to get rid of the NAND you make). Though Samsung has
by far the most developed NAND flash business among DRAM makers (and Elpida has none at all today), clearly trading off
increasingly unprofitable DRAM capacity for still-profitable NAND capacity is on every DRAM maker’s mind. Micron
had some interesting analysis of the NAND-DRAM capacity tradeoff in its 28 January Analyst Meeting, which offered their estimates
of the relative 2Gb NAND and 256M/512M DRAM price tradeoffs that would dictate maximum profitability. Fortunately for
NAND, the market is expanding very rapidly with new, even lower price points and exploding applications.
NOR may suffer
a different fate, with its future tied largely to cell phones, in which the flash portion will get increasing competition
from NAND. Though a large and robust market, NOR is not growing nearly as fast as the sum of all of the NAND “bulk
bits for audio (MP3) and video” applications. For sure, NOR and NAND flash have priced many companies out of the
profit zone for most companies (NOR more so than NAND, where Toshiba, Renesas and Samsung still make money) over the past
six months; DRAMs are teetering on the edge of more red ink for most vendors, even as they add capacity to make
still more bits.
Woe to all those NAND flash makers who joined the hunt in 2004—Infineon,
ST Micro, Hynix, Micron — and ran right into price per MB declines on the order of 50-60 percent just since last spring.
* * * * *
Planning for the irresistible force seems prudent.
