Semiconductor Equipment . . . Taiwan Semi plans to double capacity in China. The firm is the world's largest contract microchip maker, said on Monday it plans to crank up production at a China plant opening next year to double its previously announced target.
Semiconductors . . . Wedbush Morgan downgraded Broadcom to Hold from Buy based on valuation as well as problems with the start-ups the co acquired in 1999/2000, which could mean that corporate performance could be uneven and much weaker in 2nd half 2004 than estimated.
PMC-Sierra processor selected for the new HP Color LaserJet 9500 printer series.
Semi Review . . . Industry margin analysis; a summary of IDF (including DDR-2, PCI- Express, servers and WiFi) plus regular channel checks.
I had a monumental idea this morning, but I didn't like it. -- Samuel Goldwyn
Although he passed away in 1974, maybe the idea that movie mogul Goldwyn had was to examine the potential margin leverage for chip companies during this cycle to see how things compare to past cycles. It's easy to see why he didn't like the idea since there is not a tremendous amount of margin leverage left on the table for most companies.
As a capital-intensive cyclical industry, semiconductors exhibit significant positive margin leverage during upturns (negatively during downturns). One reason why consensus estimates are traditionally too low during chip industry recoveries is that there is a natural bias to be conservative with estimates and model only incremental improvements in margins. In actuality, margin improvements are often sharper than expected (as Intel and National Semiconductor have recently demonstrated).
Over the last several weeks, analysts have consciously made many models more aggressive to try to better capture the potential margin leverage possible in the current recovery. The question is how do our estimates of leverage compare to recent cycles?
The average and median gross margins for the semi group over the last fifteen years and projected forward by three years. As of 2nd quarter 2003, gross margins for the group have recovered to roughly 45%. Over the next two and a half years, analysts are estimating that gross margins will recover to slightly less than 55% -- just shy of the 1999-2000 upturn but well above the levels reached in the 1988, 1994-1995 and 1997 upturns. From this perspective, average and median gross margins could be considered a bit aggressive, but believe that the apparently tighter constraints on capital spending during this cycle may help boost gross margins above pre-bubble cyclical levels.
Actual and estimated gross margins for our coverage group. Aside from the absolute level of gross margins, another consideration is the rate of improvement. As is evident in the graph, the slope of anticipated gross margin improvement during the presumed 2004-2005 upcycle is steeper than that exhibited during 1994-1995 but not as steep as 1999-2000. To a large degree, the pace of gross margin improvement is governed by the Quarter over Quarter rate of sales growth. Unless the slope of this recovery is as steep or steeper than 1999-2000, our average and median gross margin improvement forecasts do not appear overly conservative.
The operating margin picture is a tale of three cities. Overall, operating margins have recovered sharply from the cyclical bottom set in 3rd quarter 2001, and expect further improvement until late 2005 (when anticipate that the cycle will roll over.) However, the anticipated peak in median operating margins for the entire group is not expected to match the levels seen in 1994-1995 and 1999-2000, although they are higher than 1988 and 1997.
At first glance, the operating margin forecast appears conservative -- but it is necessary to disaggregate the 22 companies we follow into three groups. Group 1 is comprised of eleven companies that are expected to re- attain or come close to re-attaining past peak operating margins, including Altera, Analog Devices, Agere, Fairchild, International Rectifier, Linear Tech, Maxim, National, Pixelworks, Texas Instruments and Xilinx. Group 2 consists of companies arguably could have greater margin leverage than analysts are modeling, but for a variety of reasons (including conservatism and/or skepticism) analysts are not driving the margin models back to peaks. Cypress, Intel, LSI Logic, Marvell and Micron fit under the Group 2 umbrella. Group 3 are companies that would need either wildly spectacular sales growth or significant restructuring to regain past peak operating margins, including AMCC, AMD, Broadcom, IDT and PMC-Sierra.
When examined at this level of granularity, a slightly different picture emerges. Group 1's operating margins are expected to come close to past peaks, and are expected to do so at a clip faster than 1994-1995 but not as fast as 1999-2000. Operating margin improvement for Group 2 is expected to come well shy of prior peaks, again for reasons of conservatism and/or skepticism. Margin expansion for Group 3 is expected to stop far short of prior peaks for reasons we stated earlier.
According to analysis, it is fair to say that models for at least half of the semi coverage universe reflect the strong margin leverage traditionally associated with a semiconductor upturn. Unfortunately, current valuations for most of these stocks (ADI, AGRA, ALTR, IRF, LLTC, MXIM, NSM, TXN and XLNX) already appear to fully discount 2005 earnings. While high valuations won't necessarily keep stocks from going up, owners of these stocks are either counting upon further multiple expansion (from already pre-bubble record levels) or margin expansion greater than past cycles -- both of which seem to be dangerously optimistic ideas.
The other half of the coverage generally reflects strong gross margin leverage but does not get the fall through to the bottom line that has been seen in the past. In some of these cases (particularly Cypress and Intel), analysts are hoping that this view proves overly conservative. In the others, structural issues, conservatism, skepticism and/or risk aversion causes us to feel that we are 'optimistic enough.'
IDF Roundup
Although there was not anything immediately actionable at the Intel Developers Forum last week, analysts did identify a number of themes that should gain more traction as we move into 2004, namely DDR2, PCI Express, Serial ATA, server chipsets and Intel's WiFi efforts.
125 million transistors can't be wrong
On the desktop side, there were relatively few new announcements Intel made that it had not alluded to previously. With the 90nm Prescott scheduled to launch within the next 6 to 8 weeks, it was the focus of a number of presentations at the IDF. Of course there were a few rumors swirling about of a delay in the launch of the Prescott, but based upon checks, continue to believe that the 3.4GHz and 3.2GHz Prescott processors will officially be launched sometime in early November. Since by then, we would already be towards the end of the Christmas build, we continue to expect that Prescott volumes in 4th quarter will remain relatively modest.
The Prescott will have over 2x the number of transistors on the current P4 (125 million vs. 55 million), in part due to a level-2 cache that is twice the size of the 512KB cache on current P4s, and also due to additional features (including the Lagrande security features). However, building it on the 90nm process allows the die size to shrink about 18% to about 110mm2. According to some reports, the Prescott is expected to consume about 100 Watts of power, up from about 80 Watts for the current P4, although Intel did not confirm this.
In addition to the Prescott, Intel also announced that it will begin shipping a version of the P4 targeted at the gaming/enthusiast market. The main feature of this processor is that it will run at 3.2GHz and have an additional 2MB level-3 cache, over and above the 512KB level-2 cache that is part of the P4 core. Intel is currently shipping a version of the Xeon processor based on the P4 core with a 2MB L3 cache priced at $3700. Do not believe that Intel will be pricing its enthusiast P4 at anywhere near those levels, but suspect that Intel's finance folks would probably not be too happy about a sub-$1000 price point for this part-which it would need to get any traction with the gamers.
Grantsdale, the next generation desktop chipset remains on-track for a 2nd quarter 2004 launch, and by 3rd quarter 2004 there will be four versions shipping. The big announcement on the chipset side was related to the mobile platform. At the IDF, the company launched the 855GME chipset for the Centrino platform that includes an enhanced graphics core, and support for DDR333 memory. In addition, Intel announced its Sonoma platform that will succeed today's Centrino. Sonoma will include a new chipset, Alviso, which will bring serial- ATA, PCI Express and DDR-2 support to the mobile platform.
DDR-2
As expected, all five DRAM suppliers claimed that they have working DDR2 silicon, and that they will be ready for the DDR ramp in 1st half 20'04. From a cost perspective DDR2 is about 20% more expensive than DDR. The die size of DDR2 is slightly larger than that of DDR, driving a 5% die cost adder. DDR2 also requires organic packaging of some type, most likely a fine-pitch BGA (ball grid array) packaging, or FCPGA (flip-chip pin grid array) packaging. BGA packaging essentially doubles the cost of assembly from $0.25 per die to $0.50 per die, representing another major portion of the cost adder. Additionally, new testers are required for DDR2, also increasing costs.
Most manufacturers expect DDR2 to remain a relatively small portion of overall DRAM volume in 2004, though the ramp is second half centric. Consensus seems to be that DDR2 will constitute about 10-20% of total DRAM volume in 2004, and will exit the year at 25-40% of total DRAM volume in 4th quarter 2004 (note the large range, which illustrates the uncertainty at this point. Server products will ramp first and steepest, followed by desktops, with small form factor mobile DIMMs ramping later in 2nd half 2004.
Pricing will pace the adoption rate of DDR-2. As with all DRAM products, pricing is determined by supply and demand market characteristics. DDR2 will be priced at a premium to DDR for all of 2004, though it will generally be a shrinking premium over time. Given that the Grantsdale chipset, due to launch in Q2'04, supports both DDR and DDR2, expect that OEMs will switch back and forth between DDR2 and DDR offerings as is required to keep DRAM at or below 10% of total BOM costs.
Do not anticipate that DDR2 will present a long-term or sustainable profit opportunity beyond that of DDR given the products higher cost to manufacture and given OEMs ability to switch back and forth between DDR2 and older DDR. As such, the DDR2 ramp, while perhaps providing some short term profit opportunities as happened with the SDRAM to DDR transition, does not change analysts long term view of the industry that DRAM companies will once again be profitable when only three or four major manufacturers remain competing in this market. Until that happens, the potential exists for companies to build more capacity than demand requires in an attempt to one up each other, thus driving ASPs below cost.
PCI Express
With the introduction of Grantsdale on the desktop side scheduled for 2Q04 and Alviso on the notebook side in 2H04, both PCI Express and Serial ATA appear to be positioned for greater adoption as we move through 2004. Analysts have heard that server OEMs are a little more reluctant to part with the specialize PCI-X bus used in today's hardware, so Intel is hedging its bets for PCI Express in the server market by offering a PCI Express to PCI-X bridge chip.
PCI Express is notable as it could drive a landscape change in market share positions in a number of areas, including graphics chips (which will drop the AGP interface) and Gigabit Ethernet chips. The Gig-E chip market is split between the Intel/Marvell combo chip, a pure Intel Gig-E chip, Marvell's 'Yukon' offering, and Broadcom. Currently the Intel/Marvell chip is used for Dell's desktops, while Broadcom also has a strong position with the Dell notebooks and Hewlett Packard's desktops. In the white box segment, the market is split between Marvell's Yukon product and Broadcom.
As PCI Express becomes widely adopted, believe a new round of PCI Express based Gigabit Ethernet products will be the standard by 2nd half 2004. As such expect a series of PCI Express Gigabit Ethernet products from all the market players to appear later this year and early 2004 for the spring refresh cycle. At IDF Intel did show prototype silicon of single and dual channel Gigabit Ethernet products for PCI Express. Intel declined to comment on those Gigabit Ethernet PHY (theirs or Marvell's) was used, but Intel has stated that they plan to utilize their internal silicon whenever possible.
While a small percentage of disk drives have already moved to SATA (Serial ATA) from parallel ATA, Alviso and Grantsdale appear to be driving the storage world into the SATA standard. SATA has long been discussed by Marvell as a driver for further SOC marketshare gains in drives and does appear to offer Marvell in roads to penetrate up the storage food chain with SATA hub products in disk arrays and RAID storage applications.
Server Chipsets
On the server side, a number of IC companies selling into the server and enterprise storage markets showed pre-production products based upon Intel's latest Lindenhurst server chipset. Lindenhurst is scheduled for introduction in 2nd quarter 2004 and is generally seen as the stake through the heart of Broadcom's ServerWorks franchise.
What is notable is that nobody at IDF or elsewhere seems to have seen even samples of a competing Broadcom product. In contrast, Lindenhurst samples have been in the market for a number of weeks now. This reinforces our view and the conventional wisdom that Broadcom may miss the server boat entirely for the 2004 hardware refresh cycle. Since corporate customers are likely to phase in new servers more gradually than product transitions in other PC segments, believe that Broadcom will be able to hit its stated $200 million sales goal for server chipset in 2004.
However, it is becoming increasingly evident that this market is shifting heavily into Intel's favor and Broadcom's revenue in 2005 could be nominal at best (are currently modeling 2005 server IC sales for Broadcom of $73M down from $208M in 2004).
WiFi
On the WiFi side, one knowledgeable source stated that Intel currently has an attach rate of between 50-65% for its 802.11b chipsets on its Pentium M processor sales (we believe that Broadcom has most of the remainder with its 802.11g chipset). Since Intel current lacks a 802.11g solution, Broadcom has run the table for 802.11g solutions at virtually every major notebook provider.
Intel recently began sampling a 802.11a/b solution and is targeting the end of the year to begin sampling a 802.11b/g solution in its Centrino offering. This is a significant risk to Broadcom from both a WLAN market share and pricing perspective. For reference, we estimate that Broadcom currently generates about 10-15% of its sales from the WLAN market.
Channel Surfing
Summary: Processor prices remained stable with the discount to list on Intel processors widening slightly from 4% to 5%, while AMD processor prices slipped slightly less than 1%. Following the Intel Developer's Forum last week, this week its AMD's turn in the limelight with the launch of its 64-bit desktop processor scheduled for Tuesday. DRAM spot prices were mixed with DDR slipping 2%-3% while SDRAM spot pricing remains firm. 2nd half September contract pricing was relatively compared to 1H'Sep. Over the next two months, generally expect contract prices to be flat to slightly up (with increased seasonal PC trends), and then decline in December as the holiday build winds down.
Processors: Pricing remains stable ahead of Computex and Athlon64 launch
Pricing in the processor gray markets continued to be generally stable, with the discount to list on Intel processors widening by just a point to 5% over the week, and unchanged at 7% on P4s. AMD processor prices slipped 1% over the week.
Following last week's Intel Developer's Forum, this week is another busy one in the PC segment, with AMD formally launching their 64-bit Athlon64 processor for desktop systems on Tuesday (9/23/03), and the Computex exhibition in Taiwan starting on Monday. Recall that this year's Computex was postponed from its original early-June schedule to September due to SARS related travel restrictions. With 60 to 70% of the world's motherboards now being built in Taiwan, along with a very strong presence of PC OEMs and ODMs, Computex has lately become the premier exhibition for the PC industry. Its normal early-June schedule allows manufacturers to preview products for the back-to-school and Christmas sales period. With this year's delayed Computex, the exhibition will largely be a pre-Christmas sampler. Since both Intel and AMD will soon be launching new processors, we expect to see a number of motherboards and systems based on these new parts. In addition a number of chipset manufacturers will likely demonstrate new chipsets for these processors. Also expect to see notebooks using Intel's latest 855GME chipset that was announced at the Intel Developer's Forum last week. About 1200 exhibitors are expected to be present at this years show, which is up about 8% from last year.
Based on the final tally of shipments from the top-tier Taiwanese motherboard manufacturers in August, Smith Barney's Hong Kong based PC hardware analyst estimates that motherboard shipments grew only 3% mom, compared to his up 6 to 8% mom estimate. Kirk ascribes much of this shortfall to one company, Elitegroup, but expects overall September shipments to grow 12% mom, up from his prior estimate of +9% due to the lower base in August. Due to the August shortfall, however, Kirk trimmed his 3rd quarter motherboard shipment estimate slightly to +21% quarter over quarter, down from +25% qoq previously. Demand out of Europe has been improving recently, and early indications for Christmas orders appear promising.
AMD is scheduled to formally launch its 64-bit Athlon64 processor for desktops this Tuesday (9/23/03) in San Francisco. While this new processor's 64-bit capability will likely be the focus of AMD's PR campaign, as these capabilities are yet to be supported by Microsoft's operating system, the Athlon64 will be used primarily in the 32-bit mode. At this point analyst are not aware of any major PC-OEM that will introduce Athlon64 based systems at the launch on Tuesday, although some of them are currently evaluating this part. As a result, the primary market for the Athlon64 for the next two quarters will be the enthusiast/gaming community. Industry contacts have confirmed that a number of motherboard manufacturers have bought these processors from AMD, and will bundle the processor along with their motherboards, thereby ensuring their customers have access to the processor. Intel continues to believe that the 32-bit architecture is adequate for desktop applications at least for the next few years, and at least publicly denies that it is currently looking to move its desktop processors to a 64-bit architecture.
DRAM Spot Prices Mixed as DDR Slips While SDRAM Remains Firm
DRAM spot pricing was mixed last week with DDR spot prices slipping 2%-3% while SDRAM spot prices firmed by 1%-2%. This marks the third consecutive week that DDR was flat to slightly lower while SDRAM was flat to slightly higher. The DDR 256Mb 333MHz and 266MHz parts each declined 3% to $4.43 and $4.41, respectively. The DDR 256Mb 400MHz declined 2% to $4.51. Note that these DDR spot parts have decline by 5-6% over the past 3 weeks. SDRAM spot pricing remained firm with the 256Mb increasing 2% to $4.21 and the 128Mb increasing 1% to $2.91.
Channel contacts inform us that activity in the spot market was tepid but some noted a slight pickup late this past week -- though spot prices have yet to reflect any material increase in vendor activity. By all accounts, DRAM supplier inventory levels remain lean, and we hear that neither Samsung nor Micron are shipping any product to the spot market. Samsung may even be short shipping its contract customers.
However, OEMs are not tapping the spot market in any measurable degree, which suggests that their actual consumption needs are being met. Since September has been surprisingly lackluster in the spot market (most brokers expected a demand pickup after August), this may also suggest that OEM inventories are high enough to bridge any supply limitations in the near-term. Seasonally, DRAM demand momentum tends to crest over the next month or two.
2nd half September contract negotiations are essentially finalized and early indications show that prices were generally flat on most high volume parts with some small increases on other specialized parts. According to the DRAMExchange.com, DDR 256Mb 400MHz contracts are at $5.25, DDR 256Mb 333- 266MHz are at $5.06, and DDR 128Mb 266MHz are at $2.56. Contract prices are currently about 15% above spot prices. Given this spot/contract spread, doubt future contract prices will be able to post the robust gains experienced in July and August. Over the next two months, generally expect contract prices to be flat to slightly up (with increased seasonal PC trends), and then decline in November or December as the holiday build winds down.
Other DRAM News
SMIC (Semiconductor Manufacturing International Corporation) announced that it is on plan to begin production of 0.14-micron DRAM by the end of 2003, and 0.11-micron DRAM in 1H'2004. Infineon is a partner of SMIC and provides technology licensing to SMIC in exchange for a guaranteed allocation of capacity.
Infineon continues to strive to gain market share and lower its cost structure and announced two separate plans this past week. First, Infineon said they will open a new headquarters in China. The company commented that it currently has 800 employees in China and plans to increase this to 3,000. Second, the company announced it plans to farm out more production of its DDR 256Mb 266MHz parts to its partners. Currently Infineon produces about 80% of this item internally and expects to reduce this amount to 50% or so.
Flash Spot Pricing: NOR Increased Slightly, NAND was Relatively Stable
Spot pricing on NOR flash memory slightly improved as two parts showed slight price increases. Spot pricing on Intel's 128M TSOP (J3A) increased 2% to $10.95 and AMD's 4M PLCC (F040-B) increased five cents to $0.65. All other spot prices for NOR memory parts remained stable with the 64Mb average at $8.90, the 32Mb average at $7.00, the 16Mb at $2.79, and the 8Mb at $3.22.
Spot pricing on NAND flash memory was relatively stable across the board with slight changes among the various densities. Spot pricing on 2Gb NAND increased about 1%, 1Gb NAND remained unchanged, 512Mb and 128Mb NAND parts decreased 1% while the average for 256Mb and 64Mb parts were steady.
Sandisk made two announcements this past week. First the company provided an update for 3rd quarter and 2nd half business outlook that as essentially in line with previous guidance of full year revenues of $950 million. The company noted that NAND flash memory capacity shortages continue in the industry. Second, the company announced plans to sell approximately 8 million additional shares with expected proceeds of approximately of $500-550 million to be used for investment into additional flash memory fabrication and testing capacity.
Boxmakers . . . Apple competes directly with products developed by Microsoft, Intel, AMD and the open source community (i.e., Linux). Microsoft and Intel alone spend 18X as much on R&D annually as Apple, making it extremely difficult for Apple to maintain technology leadership. (The R&D delta would be even larger if it included third party software companies writing applications for the Windows platform.) In addition, Apple's lack of scale relative to the Windows/X86 platform forces the company to charge a significant premium for its products, a premium that most are not willing to pay. In addition to this scale disadvantage, Apple's channel model forces the company to stock 4-6 weeks of channel inventory at all times versus Dell's 0 weeks, which can represent a significant cost disadvantage during periods of declining component prices. In the low or no-growth PC industry, Apple's growth depends on market share gains from Windows/X86 and Linux/X86. In addition to the company's price
premium, there are other reasons to believe that customers may be reluctant to "switch" to the Apple platform, including Windows application lock-in, lack of familiarity with the Macintosh operating system and concerns regarding application availability on the Mac platform. Portions of Apple's target market are becoming primarily replacement opportunities due to technology maturation and saturation, especially very profitable creative markets such as desktop publishing and graphic design. Apple derives roughly one quarter of total revenue from the education market where spending can vary widely depending on state and local government budgets. Apple derives roughly 40% of total revenue from the consumer market where demand can vary widely depending on consumer preference and macroeconomic conditions. Apple plans to open additional company-owned retail outlets in the U.S. during calendar 2003, increasing the company's fixed cost base. Near term, potential risks to our current 3rd quarter 2003 estimates and consensus include potential supply constraints on new PowerMac G5 systems, a potential sharp decline in sales of older PowerMac G4 systems, rising component costs and weak European demand. If the impact of any one of these risks is greater than we currently expect, Apple shares may not achieve $22 target.
RobBlack.com MarketWrap
robblack.com |
