1Q09 Better than 4Q08, but still terrible:

Memory makers continued to suffer horrendous losses in 1Q09, effectively wrapping about $1.75 around each of the 4B or so units of DRAM, NOR and NAND that they shipped to end customers during the quarter. According to our compilation of memory makers' 1Q09 results, they collectively lost about $6.7B on sales of $9.3B. (The "losses narrow at Elpida, Hynix and Micron" headlines were hardly reassuring, as can be seen in the table below.) It was better than 4Q08 on the loss side by more than $4.3B, to be sure, but on revenues that were likewise down 15% from about $11.0B. Write-offs for various purposes continued to play a large role in these losses, so badly vendors misjudged the market dynamic and demand outlook.

In 1Q09, both DRAM and NAND prices recovered compared to 4Q08, but only made it a fraction of the way back to where they need to be to restore profitability. Supply restrictions by both DRAM and NAND vendors worked to improve prices, but un-utilized or under-utilized capacity still remained a heavy financial burden for everyone: all memory makers have idle wafer start capacity for which they were and are takng 'wasted' depreciation in 1Q09 (charges without compensating output/revenue), and everyone permanently took down non-competetive capacity for writeoffs in 4Q08 and 1Q09.

Yes, someone made money! Again, there were two companies that made money in memories in 1Q09, and they were the same two as last quarter. Macronix banked about $18.4M on sales of $145M for first place, and GSI Technology, a smaller Specialty SRAM company, eked out profits of $1.2M on sales of $13.6M, making it perhaps the ONLY network-centric chip company (except those which sold theselves to larger companies when things were still hot in 1998-2001), to have a net positive financial experience over its lifetime.

And yes, someone else took huge write-downs: The other side of the coin saw IDT take a comprehensive $686M 'good will and asset impairment charge' during what was their Fiscal 4Q ending 3/31...against sales of $107M. But, in later news, see the "Mother of All Special Charges" in Spansion's late release of their 4Q08 results, described below.

Quotable quotes: From GSI Technology 1Q09 Statement:

... "The sequential decline of approximately $665,000 in military/defense sales was not unexpected; nor was the decline of $1.1 million in sales to Cisco Systems, given Cisco's earlier announcement that it anticipated substantially lower sales in our quarter ended March 31, 2009. These declines, however, were partially offset by a $1.8 million increase in sales to Huawei Technologies, the result of an acceleration in the 3G build-out in China."

Is this the first step towards a "Changing of the Guard"? A bona fide competitor to Cisco?

Taiwan DRAM makers: We have had an on-going BLOG commentary about the dire situation for Taiwan's DRAM makers, which gets worse by the day, and for which we see no viable end in sight (for most of them.) We should see some resolution in either the ProMOS or Powerchip matter before the end of 2Q, or when their bonds become due. The longer they are rescue-financed, the more money their creditors will eventually lose before they actually DO throw in the towel. (Well, I said that about Hynix a few years ago, and they eventually made a government and bank-orchestrated comeback, so there MIGHT be a chance. But, by doing this, Hynix effectively forced others to join in their losing life-support initiatives with unnecesssarily low prices in the DRAM business for extended periods, costing the industry billions of dollars but extending Hynix life...which was good for Hynix and its creditors and shareholders, but not very good social policy for the industry.)

Frankly, I like the Anglo-American way of getting out of business...the TI way, the "Toshiba DRAM" way, the United Technologies way (Mostek, 1985), the Intel DRAM-SRAM way, IBM Micro DRAM way...argue about it, look at the numbers without rose-colored glasses...take your lumps, shut it down or sell the assets and leave. Get on with life!

Let the oppressed go free! Give the men and machines a chance to produce something the market values.

The nearly departed, Qimonda and Spansion: Both are in various stages of bankruptcy, have clamped down most of their production to a trickle and are seeking buyers for all or parts of their operations; both will face liquidation, within a few months (Qimonda) or several months (Spansion) if they cannot find a suitor (or White Knight), or present a acceptable plan for Business Reorganization to the Bankruptcy Court. More losses are yet to be recognized, in both cases. Unlike Taiwan's DRAM makers, also, neither has received recent money from their original and majority shareholders: Infineon on the Qimonda side, AMD and Fujitsu on the Spansion side. Rumors abound, but no money is apparent. These are good companies with good technology, (once) substantial market positions, caught in tough markets...and have had the door slammed shut behind them by the financial crisis: "Sorry, no more money." In another time, they would have been snapped up into larger entities and 'made right'. But not today.

Company Financials 4Q08 and 1Q09
All Values in Millions of US Dollars
BOLDFACE entries are Denali Estimates

Notes:
"Others" include Sony, Matsushita, Renesas, Toshiba, Fujitsu, NEC, Elite Semi., G-Link, Alliance Semi, Sharp
Profits = divisional operating profits for Samsung, STMicro, Intel; profits are after-tax profits for all others
Mosys, SanDisk, SST, and Rambus include substantial IP licensing revenues as % of sales;
Intel and ST Micro combined their flash businesses (NOR + ST's NAND) into Numonyx, effective 1 April 2008
Qimonda has not reported for past three quarters (Cal 3Q & 4Q08, 1Q09); Spansion did not officially report 4Q08 or 1Q09 until 5/13/09, and is now current (see below)
Intel's NAND flash takeaway from IMFT still included in Micron's sales here.

Late posting of 4Q08 results by Spansion boosts Memory Industry 2008 losses to nearly $23B: On 13 May 2008, Spansion became current on filing their full financial results when they filed their 1Q09 results (sales of $400M and $89M loss) at the same time they released their 2008 Form 10K, which showed a huge 'asset impairment' charge of $1.6B for 4Q08, and a net loss of $2.08B for the quarter. The 1Q09 sales were surprisingly robust, in light of their bankruptcy proceeding.

Outlook for 2Q09 DRAM and NAND prices have moved up for more than four months now, but there are no guarantees the uplift will last; in fact, pessimism is the more 'conventional wisdom' one gets by reading the trade press. At any rate, price rebounds so far are not nearly enough to make memory makers profitable; at the "4Q08-to-1Q09" rate of improvement, they will break into the black about the end of 2010. Things could move faster than that, for sure. Memory makers have performed near 'feats of magic' in the past two years in terms of getting their manufacturing costs down, to keep losses under what they otherwise would have been. But, those who bet on market and profit recoveries for a livelihood (shareholders and security analysts), have stood by the sidelines, or worse: most memory company share prices still reflect much pessimism about getting things in order any time soon.

The financials for 2Q09 MAY reduce memory-makers losses by $2-3B, as write-off possibilities are nearly exhausted, and costs are coming down. Overall, the industry will still be very red, still not enough growth, and still shadowed by much latent capacity that can be brought into play at a moment's notice.

Memory Companies Suffer More in 1Q09, but Future Looks Better…or so they say:

Many of the world’s leading memory companies have already reported their 1Q09 financial results, which are, not surprisingly, about the same as their 4Q08 results, both in the sales levels and profits, though DRAM was marginally worse and NAND flash likewise a little better. Hynix, Samsung, SanDisk, Micron are all in; Elpida and Toshiba have just finished their Fiscal 2008 on 31 March 2009, and will release 'official' year-end numbers soon, but company-authorized ‘close estimates’ are widely available in advance of fully audited results.

Based on these companies, which collectively make up more than 60% of total DRAM and 90% of NAND sales, one can see how well the industry fared as we moved into 2009:

* Micron's FY ends 8/31, so its 1Q09 was Dec-Jan-Feb; others 1Q09 were all the Jan-Feb-Mar months.

Caveat #1: Several Companies...DRAM makers ProMOS, Powerchip and Qimonda, in addition to NOR market co-leader Spansion...are in various states of restructuring, negotiations for sale of all or part of the operation, bankruptcy, pre-liquidation, and/or receivership, and have not even fully reported their 4Q sales and profits. The remaining companies not shown above are all expected to report within two weeks, when a fuller accounting will be presented.

Caveat #2: Since many companies took large annual write-offs in 4Q08, to correct for inventory over-valuation, impaired assets and Good Will that turned out to be not so good, we expect that once the dust settles, 1Q09 will show some real improvement in results compared to 4Q08...for most, there was nothing left to write down as they rolled into 2009. According to our earlier 4Q tabulation (including estimates sales and profit/loss for not-reporters), the ‘memory’ industry had 4Q08 sales of $11.0B and lost about $8.8B, finishing out the industry’s most painful year ever.

Caveat #3: Exchange rates continue to be volatile, and to muddy any interpretation of results which are converted from ‘native’ currencies of most memory vendors. We reproduce our comments from an earlier article, below:

“As can be seen in the notes at the bottom of the Company Financial Results table, exchange rates changed markedly in 4Q08, compared to earlier in 2008. In particular, the Yen appreciated more than 10%, the South Korean Won dropped more than 35%, and the Euro dropped more than 20%. The NT$, which is loosely tied to the US dollar, remained about the same in the fall quarter.

Elpida has relationships with Rexchip and Powerchip in Taiwan, from whom it buys product. Most of the Koreans’ production is in Korea, except Samsung’s sizable fab in Austin. Hynix’ Oregon fab was decommissioned during 4Q08. All of these realities cause some accounting and ‘financial results’ interpretation challenges. Qimonda is mostly in Germany (Dresden), but has had production agreements with Winbond and SMIC in Asia, and closed its Virginia fab in 4Q08.

Many companies have debt denominated in US dollars, as well.

All these companies, plus the Taiwanese, report their financial results in their native/headquarters currencies, though they all sell internationally, at prices usually denominated in US dollars. Their books, on which they account for assets and liabilities (and depreciation, which makes up large fraction of their cost-of-production), are mostly in native currencies, with offshore accounts converted and rolled into consolidated results.

For Hynix and Samsung, we have converted results back into US dollars for a common basis of comparison; their memory sales were sequentially down by 18% and 21% in 4Q, respectively, measured in Won, and worse still when denominated in dollars, due to the Won’s depreciation against the US dollar.”

Caveat #4: Samsung’s presentation in their analyst slide set for 1Q09 results had some changes in the way they report product lines for 1Q09, making a sequential comparison difficult but not impossible. This complication will temporarily add to interpretation difficulties caused by their international business operations (production in Texas), and the continued Won depreciation against the dollar. Suffice it to say, their 1Q09 results for memory were ‘about the same as for 4Q08.’

Summary: When final tallies are made, we expect that sales for 1Q09 will at about $9.5-10.0B, and an industry-wide loss that is reduced to about $6-7B. The loss is expected to be less due in part to improved pricing and pricing stability, but offset by reductions in GB shipments or shipment growth of both DRAM and NAND, as well as the exhaustive ‘housecleaning’ done in late 2008.

Memory Makers lose $8.8B in 4Q2008, to bring annual losses to $20B: Memory companies lost about 85 cents for every dollar of sales in 4Q08, and, for certain, there’s a lot more bad news coming in 1Q09. According to our survey and analysis, memory makers lost about $20B in 2008, on total revenues of $56B; taking into account transfers from foundry partners to their parents for resale, memory product sales to end customers totaled about $46B, down from about $58B in 2007, despite GB growth for NAND of more than 135% over 2007, and of DRAM by more that 66% vs. 2007.

(Denali tracks companies which make memories. In some cases, those companies also make other non-memory products. We separate the memory portion of sales and profits where it can be done with some confidence, but there remain two complicating factors: (1) Atmel, Cypress, ISSI, SST, Winbond, Macronix and some others, have non-memory products for which we cannot separate P&L, and (2) There are some significant transfers between foundries and their principal customers…PSC to Elpida, Nanya/Inotera to Micron, ProMOS to Hynix, etc., which result in double-counting some revenues when reported sales are used in the summaries. In this case, the losses are strictly cumulative, while the sum of the revenues exceeds the end-market total by the amount transferred within the memory makers community. Therefore, the sum-of-the-company sales are not necessarily representative of the total business volume to the end market.)

When the financial results for 4Q08 for the 25 or so companies that make memories are totaled up, we found that they lost about $8.8B on sales of $11B. While DRAM makers had bled to death all year, NAND makers weakened markedly in 3Q and on into 4Q, when prices dropped about 15%, on the back of 3Q’s drop of 30-35%. (Rising NAND prices did not gain traction until the very end, and have been sustained into 2009).

We calculate that NAND flash makers took losses of about $3.3B in 4Q08, on sales of only $2.4B, and no vendor escaped red ink…which included sizable write-offs from SanDisk and Toshiba.

In the context of this discussion, losses include costs of decommissioning fabs, operating losses, inventory write-downs, ‘asset impairment’ charges. In their usual year-end clean up and resignation that things are really, and irreversibly bad, the fourth quarter is frequently a catch-all for the financial sins of the year before. Those charges aside, however, 4Q08 was without question the largest money losing quarter for memory makers ever, though we will have to wait a few more weeks to see how long the record will last. ..maybe not that long.

Company Financials 2008
All Values in Millions of US Dollars
BOLDFACE entries are Denali Estimates

Notes:
"Others" include Sony, Matsushita, Renesas, Toshiba, Fujitsu, NEC, Elite Semi., G-Link, Alliance Semi, Sharp
Profits = divisional operating profits for Samsin, STMicro, Intel; profits are after-tax profits for all others
Mosys, SanDisk, SST, Saifun and Rambus include substantial IP licensing revenues as % of sales;
Intel and ST Micro combined their flash businesses (NOR + ST's NAND) into Numonyx, effective 1 April 2009
Qimonda has not reported for past two quarters (Cal 3Q and 4Q08); Spansion did not report Cal 4Q08)
Taiwanese PSC, ProMOS and Etron have reported sales but not yet their profits for 4Q08

Two memory companies were profitable in 4Q08. Macronix has been the leading supplier of Mask ROMs for many years, and though its sales were impacted, they still came up with profits of about $25M on sales of $170M, both down from 3Q08 and 4Q07. GSI Technology, a high-speed and networking SRAM maker in Silicon Valley, has picked up the slack in these two markets that other SRAM makers have moved away from, but which continue to support perhaps a few hundred million dollars in sales annually spread among all suppliers.

Remedial actions by DRAM and NAND makers came too late to help 4Q results, but have had some beneficial impact (esp. for NAND) in 1Q09. DRAM supply cutbacks ran into weak PC sales, and the ever-present move to smaller die and higher productivity, despite taking of thousands of 200mm wafer starts/month off line in the past 4-5 months. But, for the most part, demand was dropping faster than supply, and with still some inventory overhang from last fall, supplies were still outrunning demand. Even the NAND prices moving up in the past six weeks is viewed with some suspicion: Will it last, or not?

Discussion #1: Exchange rates’ impact on these results: As can be seen in the notes at the bottom of the Company Financial Results table, exchange rates changed markedly in 4Q08, compared to earlier in 2008. In particular, the Yen appreciated more than 10%, the South Korean Won dropped more than 35%, and the Euro dropped more than 20%. The NT$, which is loosely tied to the US dollar, remained about the same in the fall quarter.

Elpida has relationships with Rexchip and Powerchip in Taiwan, from whom it buys product. Most of the Koreans’ production is in Korea, except Samsung’s sizable fab in Austin. Hynix’ Oregon fab was decommissioned during 4Q08. All of these realities cause some accounting and ‘financial results’ interpretation challenges. Qimonda is mostly in Germany (Dresden), but has had production agreements with Winbond and SMIC in Asia, and closed its Virginia fab in 4Q08.

Many companies have debt denominated in US dollars, as well.

All these companies, plus the Taiwanese, report their financial results in their native/headquarters currencies, though they all sell internationally, at prices usually denominated in US dollars. Their books, on which they account for assets and liabilities (and depreciation, which makes up large fraction of their cost-of-production) are mostly in native currencies, with offshore accounts converted and rolled into consolidated results.

For Hynix and Samsung, we have converted results back into US dollars for a common basis of comparison; their memory sales were sequentially down by 18% and 21% in 4Q, respectively, measured in Won, and worse still when denominated in dollars, due to the Won’s depreciation against the US dollar.

Discussion #2: Taiwan and Government actions: Although no action had been agreed to during 4Q08, there was much posturing and discussion among all interested parties…the six Taiwanese DRAM companies: Powerchip, ProMOS, Inotera, Winbond for Qimonda, Nanya, Rexchip, plus prospective outside technology providers Elpida and Micron. Since this is not yet a done deal, and may not even happen, we’ll leave it at that and pick up the thread in a later edition.

Discussion #3: IP, royalties and license income: This nice nearly fixed-income stream amid declining sales revenues takes on a special tint in bad times, both for users/licensees (like a tax), and for providers, as something akin to a monthly Social Security Check. For the year, here are the reported royalty and license incomes for the major memory makers, SanDisk, Rambus, SST, and Mosys. Other companies get IP income, but it is paid in kind, or hidden in larger joint technology transactions (IMFT, full company IP cross licenses (ST and Intel), or specific transfers (Elpida from Hitachi and NEC then Mitsubishi).

Source: Company Reports

Hidden from view are such IP plays as the licensing of technology and DRAM designs from Elpida to Powerchip, from Hynix to ProMOS and from Micron to Nanya. Each process node installaion and associated designs costs between $125M and $180M. Inotera and Rexchip do not sell their own branded product directly the market and their output goes to their respective owners, so no royalties are involved directly with Rexchip and Inotera.

In the context of self-developed technology, the memory industry's R&D budget is probably in excess of $10B/year, so the costs of purchased or licensed technology, while not insignificant and a useful financial adjunct to some suppliers, are still a small part of the industry's technology acquisition budget.

Discussion #4: Sales by Quarter, by Product, Profit or Loss, NOR, NAND, DRAM, other: And another look, using additional outside data sources, to estimate the sales and profit by product line, across the entire memory product spectrum:

"EDIT: I have corrected Etron's 2007 P&L entry to show a net profit of 39M instead of 13M. Thanks and apologies to my Etron readers who noticed this discrepancy."

“Let’s wait a while more; the problem will take care of itself!”

In the past two years, Taiwanese memory companies that are being considered as candidates for government-driven consolidation have posted losses of about $5.3B on sales of $15B…with more losses coming day by day, as we roll through the toughest DRAM market ever in 2009.

The fullest exposition of the government plan, which has been percolating for about four months already, was disclosed on Friday, 6 March 2009. It did not instill confidence that the ‘shotgun-marriage’ was either imminent (it may take six more months), or was even an eventuality, or that it would be successful in any way except in keeping the Taiwanese firms alive to play another day... provided they are alive when the actual consolidation contract is ready to be signed.

The details about technology transfer are still not being made public, or maybe still undecided; how much government cash will be involved is also not clear, even though the $6B number has been bandied about for months now. (Latest word is that, once the huge bill for “Taiwan DRAM maker salvation” hit the lawmakers, they downsized the operation, hoping to keep government investment under $1B.) Elpida’s and Micron’s respective roles are still unclear and yet to be determined. As well as how the existing shareholders will be compensated for their diluted ownership share is unclear. Rexchip either is or is not included in the list of consolidation candidates. The Founding Fathers gave themselves an extraordinarily long ‘tie-it-all together’ window of six months, in addition to the approximate four months, already invested in discussion.

But if the market stays as bad as it is, ‘natural forces’ will perform their own brutal triage of DRAM makers, and not just the Taiwanese. Who has enough money to ship a $1 bill with each 1Gb DRAM they sell for 85 cents…for the next six months and 4B units of 1Gb DRAM? In fact, memory makers are losing money about 3-4 times that fast at the present time... about $2-3B per month, and perhaps 75% of those losses are among DRAM makers. Indeed, maybe the ‘six months’ clause is a clever exit strategy for the Taiwanese government, as the crisis resolves itself with bankruptcies, free-market consolidations and the ‘natural’ closing of fab doors. Many Hsin Chu citizens, formerly employed in the local high-tech industries, are now outside looking in, waiting for the rebound. As they got deeper into the discussion of costs and prospective consequences, they certainly realized the difficulty in doing what they hoped to achieve... making the industry sustainably whole without hurting anyone’s feelings.

Taiwan Memory Company Financials, all Q08
(All Values in Millions of Dollars)

Notes:
* Corrected number from Etron Corporate Office.
Bold Faces in Profit Table are Denali Estimates
NT$/US$ used in conversions were: 1Q, 31.4; 2Q, 31.0; 3Q, 31.5; 4Q, 32.8
Taiwanese PSC, ProMOS and Etron have reported sales but not their profits for 4Q08
Although Etron is a fabless memory maker, it's product line crosses all those who are in line for Taiwanese Government "assistance", and so will also be impacted in the marketplace.

I am reminded of another DRAM “crisis,” which grew out of the late 1980s DRAM shortages, and the Japanese emergence as a broad competitor to US technical supremacy, esp. in DRAMs. 1985’s DRAM market drove all U.S. companies, except Micron, to the exits, and every day had U.S. claims of "Unfair Trading Practices" lodged at the Japanese. In 1987, a DRAM shortage developed that was (or was not) deliberately engineered by the Japanese to put the final nail in U.S. DRAM maker’s coffin. High prices called into play the need for another U.S. DRAM maker, and an initiative led by IBM’s Sanford Kane, sought $1B in funding (things were cheaper then) to build a new DRAM company around IBM’s ‘donated’ DRAM technology. It was christened “US Memories.” But before they could put it all together, again, natural market forces saved the day: brought DRAM prices down from the stratosphere, made for abundance again, and “US Memories” quickly disappeared into the back pages of the industry’s history books before it even invested its first nickel. I was at the meeting when they finally pulled the plug on the “US Memories” idea; everyone sighed, turned and walked out the door. It was over.

So, too, we may see this for Taiwan Memory Company. Indeed, my bet is that Micron-Nanya-Inotera, plus some additional back-up help from Formosa Plastics (Nanya’s successful parent) will emerge as a very competent DRAM entity, with more technical participation by Nanya, and based in Taiwan. Also, it is likely that Elpida will have to find financial sustenance elsewhere, perhaps go back to Japan and knock on a few more doors.

Hynix, which lost $2.5B in DRAMs and NAND in 3Q and 4Q08 combined, seems to be in the care of its bankers and owners (debt holders), who are reluctant to write it off or refocus its strategy — “We have come too far than to give up now.” They should talk to SanDisk shareholders, who, when the stock was trading at $15 last year, declined an offer of $25 from Samsung... and now it trades at $9.22; or Spansion, which was delisted from NYSE last week, and was last trading at 2 cents a share, despite $2.3B in sales in 2008, strong market positions and a good IP portfolio. Waiting for a sunnier day is often is often a mistake.

Samsung (whose reported memory financials I have some suspicions about), certainly has sister divisions in high-end phones and LCD displays, which can generate cash to buy time for DRAMs and NAND Flash to recover...but only up to a point. Samsung is good, they are not God.

It is not a safe bet today, to assume that ‘we’re in a cyclical industry, and it will come back, eventually.' Cash is king, and will be for a while. And, after two VERY tough years in DRAMs, and maybe one more coming in 2009, every part of the operation is bare bones and reduced to the essentials: no more money for fabs and the next gen 50nm processing, no money for design shrinks, no money for developing new DRAM markets, “hold that DDR3 design”... no new hiring, no bonuses, no raises…

In any case, as Yogi always said, ‘It ain’t over ‘til it’s over’.

Supply Growth Continues to Flood NAND and DRAM Markets, Making Price Stability Impossible: With NAND and DRAM prices showing no signs of bottoming out, memory makers went deep into the red DRAM in 3Q08. The many remedial actions taken by virtually all participants had yet to be reflected in price stability, so out of balance was the memory supply line, compared to demand. What was worse, an abrupt weakening of demand welcomed memory vendors as the fourth quarter and Christmas came into view, largely attributed to the economic uncertainties surrounding everything financial these days. We estimate that memory makers lost about $4B in 3Q, against about $3B in 2Q08. It is hard to imagine the financials not getting significantly worse in 4Q, and continuing out into 2009's first half.

Steady productivity improvements, as the proximate means of cost reduction, only served up more DRAMs and GB of NAND. Capital investments in capacity coming on line merely added fuel to the fires of excess supply.

As with OPEC and oil, it is not clear if production cutbacks can keep up with weakening demand, and chip industry pundits are forecasting both sharp slowdowns in growth for both supply and demand for 2009, compared with today’s run rates.

Trying to Stop DRAM and NAND GB Growth to Balance Both Markets! Methods of slowing DRAM and NAND supply growth were many and varied. Most 200mm NAND lines have now been taken down or will be soon, due to their lack of cost-competitiveness with 300mm lines (IMFT-Boise, Nanya, Hynix Eugene OR and Korea…for a total of five 200mm fabs, several others). Many fab expansion plans were paused (Samsung NAND Austin, Toshiba-SanDisk Flash Alliance, IMFT Singapore, Nanya-Micron’s MeiYa, others). CapEx for 2009 were all cut back, some as a result of the existing overcapacity, some due to creeping cash-flow problems, some due to the financial markets’ turmoil and uncertainties about 2009 overall economic outlook.

In DRAMs, Taiwan has been exceedingly hard hit, in addition to showing the worst financials for 3Q08. Talk of a government bailout or support is making its way into discussions, but there is no clear direction or action yet. Taiwan DRAM makers lost 75 cents on every dollar of sales in 3Q08, counting restructuring costs and write downs. Prices have dropped still further from average 3Q levels, indicating both that a balance point has not been reached and that the 4Q financials will be even worse.

“Cash is King.” Most DRAM makers are cash-flow negative, putting a rather strict time limit on how long some companies can survive without financial assistance. This cash need and inability to generate it internally, has run smack dab into the worldwide tightening of credit. What prudent banker will lend money to a DRAM maker or NAND maker in such an uncertain business climate as we have today, which is already losing a lot of money? This is certainly the most severe test of many memory makers' business hypotheses yet: (1) Do pure play memory makers have a chance against broad-line suppliers (who may have readier access to cash)? (2) Does the Taiwan “Licensed Technology and make PC DRAMs Only” business model have a future? (3) Will these Taiwanese production partners-foundries be pulled back into the Parent/Technology developer fold, one way or another? (4) Ultimately, what number of independent DRAM or NAND production entities will the industry support, with its huge capital demand and associated technology and scale economies? Will there be four players in each market in 2010? Three? Two? One?... No one knows, for sure. But both the NAND and DRAM industries seem to be well on their way to finding a new equilibrium in competition, investment in fabs and technology, and industry concentration. No one thinks things will stabilize before mid-2009, which might as well be an eternity for those companies with thin bank accounts and reluctant bankers.

3Q08 Financial Results for Memory Companies: Below are the collected financial results for the chipmakers in the memory market which we track. All these results need ample background to a full understanding, and we have appended some discussions of key interpretative factors that have come up in 3Q08.

"Others" include Sony, Matsushita, Renesas, Toshiba, Fujitsu, NEC, Elite Semi.,
G-Link, Alliance Semi., Sharp
'Profits' = divisional operating profits for Samsung, STMicro, Intel;
profits are after-tax profits for all others
Mosys, SanDisk, SST, Saifun and Rambus include substantial IP licensing revenues as % of sales;
Exch. Rates 1Q08: 105.8Yen/$; 962 Won/$, 0.659 Euro/$; 31.4NT$/US$Exch. Rates, 2Q08: 105.0=Yen/$US; 31.0NT$/US$, 1040Won/US$; 0.646Euro/$US
Exch. Rates 3Q08: 104.5 Yen/$US; 31.5 NT$/US$;1120Won/$US; 0.676Euro/US$
For 4Q, Won could be as high as 1350/$US; 1 Euro = 1.30 US$ at 3Q end (=0.77Euro/US$)

Three Memory Companies Still Profitable: Macronix, GSI Technology and IDT all remained in the black in 3Q08. Macronix had net profits of $67.3M on sales of $240M, remarkably good even in the best of times, more so in today's dark climate. They are the Number One maker of Mask ROMs.

GSI is nearly the sole survivor of the Network Memory Fantasy Game from the early part of this decade. They recently announced a 450MHz 18M SigmaRAM. GSI Technology earned $3.6M on sales of $17.1M in 3Q08.

IDT, whose specialty SRAM line is only a small fraction of total sales these days, earned $11.7M on $201M for the quarter.

Samsung Results: In their 3Q conference call, Samsung claimed that they were profitable in memories, but we’re skeptical. We believe that if they were, it was through the use of unorthodox fab depreciation and cost-assignment accounting which shifted costs from one accounting period to another, or from one product line to another, in ways not usual in the industry. We will be looking into this and writing more about it in future blogs, but for comparison purposes here, we have shown Samsung as NOT profitable in 3Q08. Once we dig further, it may be that the 3Q results were even worse than shown here. It is hard to imagine that Samsung's basic manufacturing cost structure, based on yields, die sizes, process ground rules, fraction of output run on 200mm lines, expense levels (SG&A and R&D), were that much better than Hynix, which lost nearly $0.90 on every $1.00 sales, and Elpida, which was uniformly running 70nm design rules for the entire 3Q, has a good mix of differentiated products, and still lost 30 cents on every dollar of sales. Samsung even said that their NAND price/GB achieved in 3Q dropped 30% from 2Q, which is quite a price reduction curve to stay ahead of by reducing costs.

Profits, GAAP and extensive nuancing of results: During 3Q08 and again to be expected in 4Q08, great care must be taken in reading and interpreting financial results. It is these times, with rapid price declines and industry uncertainty, that management goes to great lengths to ensure that the financial results accurately reflect the financial status of the company. They also aggressively write down or write-off assets which no longer have value, sell assets at a loss (compared to their book value) to improve cash balances, write-down inventories to the new Pricing Realities..."mark to market". Some assets sales have taken place, and more can be expected.

The incidence of these practices during 3Q08 is more than can be detailed here. Suffice it to say, fair interpretation of current financial results demands a lot of care.

Other Events and Issues:

Samsung-SanDisk: During 3Q08, Samsung made an unsolicited offer for SanDisk of $5.8B ($26 per share when the SanDisk street price was $14/share), but withdrew it after Toshiba bought an additional 30% share of SanDisk's JV Fab Flash Vision for $1B; this made SanDisk richer by $1B, but gave them fewer GB of NAND from their JV fab. With two quarters of losses and losing market share of total NAND GB sales, SanDisk probably needs the cash more than the NAND capacity. In addition, now that SanDisk's stock is trading at under $9, they may rue the day they turned their back on $26/share, no matter that the stock WAS ONCE $54 within the past year.

Exchange Rates Chaos: The South Korean Won was trading at under 950/US$ until a few months ago. The US$ was $1.60 to a Euro as recently as early July. Most recently, the Euro dropped to under 1.35US$ and the Won went up to more than 1400/US$. The NT$ was stable at 30NT$ = 1US$, but now trades at more than 32.5NT$/US$. The global financial chaos, where dollars are still the predominant reserve currency (estimated at more than 65%), and international invoices are billed in US$, tight credit put many of the world’s currencies at a severe disadvantage as businesses rushed to buy US$ to pay their bills...driving up the price of US$ and driving down the value of the Won, the Brazilian Real, and even the Euro. This plays havoc with setting ‘sales’ and financial performance against a stable and common backdrop. Chip prices are given in US$, but most internal accounting of depreciation, R and D, and SGA are in native currencies. We have just seen the beginning of this in 3Q08, but it will bloom fully beginning in 4Q, with the US$ sticky at a high value not seen for six or seven years.

Mind the GAAP: What is a meaningful financial exposition for investors, or what the SEC demands in terms of reporting financial results, may be quite different from what is most useful to corporate management. One recent ‘improvement’ for ‘transparency’ is the reporting of the value of stock option grants and exercises as a part of the expenses, as they are viewed as employee compensation. Deferred compensation takes another toll to reconcile GAAP accounting with whatever non-GAAP method is used by the company.

Cypress also needs five other adjustment line item categories of expense-cost-gains-losses to reconcile their GAAP earnings with those of their own non-GAAP, the latter of which is used internally to measure operating performance.

Write-downs of inventory and ‘asset impairment charges’: Periodically, but not continuously, manufacturers assets...inventory and capital stock...are written down (or less often, up), once it is realized, and recognized, that the value they are carried on the books are no longer valid, and there is no chance that the market price will ever again reflect the book value. Chip prices are volatile, and this write-off or write down step is not taken until it is pretty clear that the value is gone forever.

This is also done with capital equipment and other assets once realized that ‘the world has moved on’. This write-down of assets was done many times by many players in 3Q08, as it was realized that 200mm lines would never again be competitive with 300mm lines. At least a dozen 200mm lines were closed, or announced to be closed, during 3Q, constituting several hundred million dollars worth of write down. It is a tacit admission that they did not charge as much depreciation as they should have in earlier accounting periods...which would have reduced historical profits and earnings per share.

Once an asset is deemed unproductive to the point that its value is not as high as the value it is carried on the books, but it still useful, a company can take an ‘asset impairment charge’; indeed Intel took a $250M asset impairment charge for its stake in Numonyx and ST Micro likewise took a $344M charge for its share of Numonyx during 3Q08. A horse of a different color, is the 70%+ stake that Infineon now has in Qimonda, which is in danger of being delisted from the NYSE due to too-low a share price…the same problem Spansion has, and which AMD and Fujitsu face with their respective stakes in Spansion. We can expect a reverse split to be forthcoming for such companies to avoid delisting.

Finally: We have begun to upgrade our Samsung Financial results to reflect consolidated operations, which adds in the results from their semiconductor operations in Austin (DRAM and NAND flash). In the past, we have reported only the parent company results. In future edition of these results, all prior Samsung results will be restated to be on a consolidated basis; in these results, only 3Q08 is consolidated.

Samsung non-parent memory sales (Austin) and profits have been running about $500M/quarter of sales and mixed Operating Profits or loss of about 8-10% of sales. The older DRAM fab, dating from the late 1990s, runs 60K/mo. of 200mm wafers; the NAND flash fab was launched just a few years ago, now runs about 70K/mo. of 300mm wafers.

Memory Maker’s Profit Shortfall Amid a Time of GB Aplenty…And It Will Get Worse before It Gets Better

Based on reported financials, it appears that only Samsung and Macronix were profitable in memories in the 2Q08 session. And it was questionable for Samsung. NAND prices have headed steadily downward for now a full year, without interruption, at a rate way faster than costs can be reduced. DRAM prices, which have been bad for most of 2008, paused for a few months with stable prices in their own race to oblivion. However, they resumed their march to ever-lower quotes recently, and remain under significant pressure. Strong demand and high GB shipment rates can usually act to stabilize pricing. But not this time, since large production increases have washed out any potential shortfall of supplies. Indeed, the huge expansion of memory demand over the past 24 months has been totally without impact in taking the pressure off DRAM and NAND Flash prices. They have fallen just as much as if demand had collapsed.

For the 25 companies that make up our memory producer database, losses were reduced in 2Q08, compared to 1Q08, largely due to more stable DRAM prices and fierce cost cutting throughout the industry. Cost reductions of 10-15% per QUARTER have not been uncommon for DRAMs, and sometimes more than that for flash, in both the NOR and NAND camps.

For the first half of 2008, memory makers have lost about $6B on sales of about $25B. No one expects 3Q to be much of an improvement (indeed, DRAMs will worsen), and 4Q is anyone’s guess. Well-respected market punditry iSuppli, foresees no market turn until 2H09. Though CapEx has been shaved back, mostly from Taiwanese DRAM makers, productivity is ‘on a roll’, with everyone ramping next-generation designs, pushing the lithography and innovating on more compact cell structures and multi-bit per cell advances.

This is not demand weakness that is driving low prices; it is supply excess, driven by these impressive productivity improvements and technical advances, and hyper-capacity investment, in 2006 and 2007, with momentum into 2008, borne of ‘irrational investment exuberance’ and some probably misguided strategic ambitions to gain market share either by driving the weak players from the market, or adding wafer starts while everyone else is (supposedly) asleep at the wheel.

Of course, this never works and only makes matters worse before they get better. Financially, the industry would have been far better off by pooling their resources, buying “Qimonda” (or some other player), divvying up their assets, taking their fabs off line and dissolving the company. Too late, the damage is done. That $6B in 1H08 DRAM losses would have bought almost any DRAM supplier and all but the largest NAND makers.

In addition to the quarterly dose of red ink, strategic moves to cut costs by technology development sharing, and DRAM CapEx cutbacks to ration cash have been more common. No one has dropped from sight yet, but the pain is palpable and the end is not in sight. As we noted in our earlier BLOG (Memory Industry Humming While Losing Money, 08/11/08), all the excess cost is being squeezed from the production lines, which are running at high utilization rates and uniformly on leading-edge processes. This is like driving on the freeway in Los Angeles: 75mph and with only one car spacing between you and the next car, thin profit margins for too long have left the industry with similarly thin margins for error, or to respond to new and changed business environments.

NOR: Numonyx (formally launched 1 April 2008), and Spansion maintain the lion’s share of this flat market, combining for about 80% share. Spansion continues to bring forth innovations based on Mirror-Bit, which they will take into the NAND Flash market with ORNAND II, with their EcoRAM that seeks to upset the data server memory applecart, R/W performance boosted NOR flash, a significant agreement with SMIC (as a part of their own version of FabLite capital make-up)..but, so far, their financials show no impact from these moves…many of which will require significant market making efforts to ease concerns about unorthodox applications of replacement technologies, i.e., it will be a slow process. Their 8K 300mm wafers a month in their Japanese fab makes hardly a dent in their quarterly silicon output…they could easily use 5x that to reflect 65nm-Mirror-Bit business potentialities and cost advantages, as well as their focused efforts to take on NAND in phones with ORNAND and ORNAND II. Oh, well.

Numonyx is showing a more traditional direction with NOR flash, sticking with the floating gate process (but with PCM in the background), being very tight on capital, and signing an agreement with Elpida to supply back-process generation NOR flash on 300mm wafers. With a ‘Major’ in NOR, Numonyx also has a ‘Minor’ in NAND, brought in from ST and foundried at what was the ST-Hynix JV fab at Wuxi, China.

Below are the gathered company financial reports for 2Q08 (incl. Micron Technology, whose fiscal 3Q ended May 31), ST-Intel-Numonyx is proving a little hard to unravel, so far, but it should become clearer for 3Q08.

Hopefully, this is the bottom of the market, but only time will tell

Memory producers had predictably bad 1Q08 financial results, with DRAM prices low and lower and NAND prices dropping 30-50% from 4Q07 to 1Q08, depending on who is doing the talking. All in all, revenues for memory makers tracked in our worksheet dropped 13% from 4Q07 and profits (or losses, as the case may be) continued their slide, and exceeded about $4B. A few smaller companies and those outside DRAMs managed to keep their heads above water. Sales and profits are shown in Table 1 below (all values are millions of dollars):

"Others" include Sony, Matsushita, Renesas, Toshiba, Fujitsu, NEC, Elite Semi., G-Link, Alliance Semi., Sharp
'Profits' are divisional operating profits for Samsung, STMicro, Intel; profits are after-tax profits for remainder of companies
Mosys, Sandisk, SST, Saifun and Rambus include substantial IP licensing revenues as % of sales;
Exchange Rates 1Q08: 105.8Yen/$; 962 Won/$, 0.659 Euro/$; 31.4NT$/US$

A few additional comments are in order:

Numonyx Launch: At the end of the quarter, Numonyx was launched out of ST's NOR and NAND businesses, combined with Intel's NOR business. Intel will continue to build some NOR for Numonyx on a contract basis for a while, Numonyx will continue to supply Intel-branded NOR product to Intel, for service continuity to Intel customers under existing Intel contracts. All of Intel's NAND business is a part of IMFT, their JV with Micron, and they take half the output for IMFT's several fabs. So, Intel is out of the Flash (manufacturing) business, but still sells NOR and NAND, at least for a while.

Intel's Flash Operating Profits: Estimating the operating loss from Intel's flash business was complicated by several issues, which were not entirely made clear in their 1Q08 call to analysts. They DID take a $275M Asset Impairment charge against their assets transferred to Numonyx; they DID note sales for all flash of $497M, and alluded to an estimate of $300M in 1Q08 for the NOR portion only (the remainder was assumed to be NAND, received from IMFT). We estimated an operating loss of $500M, including the one-time asset impairment. This does not get any easier, even with shedding NOR to Numonyx with the aforementioned continuing buy/make/sell arrangement noted above. Intel also will own a share of Numonyx profits and loss?

Stock Option Transparency, and Adjustments: All companies now seem to have their stock option backdating issues behind them, are in compliance with current laws on Stock Option pricing, and have, where necessary, restated their earnings. Whew, glad that one is done.

Cypress is now "Cypress Semiconductor" only: Cypress Semiconductor is shown separately now, as their solar panel unit, Sunpower, operates and reports as a separate entity.

Deleted m- systems and Saifun from roster: m-systems and Saifun have now been dropped from the worksheet listings, since they have been acquired by SanDisk and Spansion, resp., more than a year ago.

JVs Boil up to Improve Industry Efficiencies, Stay Alive: Joint ventures, discussion and business combinations expanded considerably during 1Q08, in response to weak DRAM pricing. Micron is now partnering with Nanya, displacing Qimonda. The fate of Qimonda and Nanya's Inotera DRAM fab is still undecided. Qimonda is now talking to Elpida about coordinating R&D efforts. Elpida has Powerchip as a foundry and Rexchip as a major 300mm 70K/mo JV (with Powerchip) for DRAMs. SanDisk's JV NAND fabs with Toshiba are stable, but they also announced an extension of their agreement for additional NAND flash manufacturing partnering.

Samsung's Unusual Loss led by DRAMs; Taiwan Pure Play DRAM Makers Lose about 76 cents for every Dollar of Sales: By our best estimates, Samsung memory products lost money for the first time since early 2001, largely due to DRAM price erosion. As a whole, Taiwan's DRAM makers (Nanya, Inotera, ProMOS, Powerchip) lost an estimated $979M on DRAM sales of $1291M. In general, we believe that NAND makers results ranged from being slightly profitable to losses of 25-35 cents per dollar of sales. DRAM makers showed losses ranging from 30 cents per dollar of sales down to "1:1."

Commodities? Playing in the Same Markets? There was a wide variation among companies in what they reported as year/year and sequential price declines for both DRAMs and NAND flash in their conference calls. In addition, what they saw as sequential MB shipment growth for both NAND and DRAM were quite variable, and not easy to understand. Although there are good reasons for some variations from vendor to vendor, the Hi-Lo range was often close to 2x.

Macronix and Mask ROMs Lead the Market: Among all memory makers, Mask ROM maker Macronix starred by having profits of $25M on sales of $161M...and was the memory industry's best financial performer.

Weakening US Dollar's Impacted Results Considerably: The decline in the value of the dollar against (most) major currencies during 1Q08 influenced sales levels and earnings dramatically, as the dollar fell 6-10%. The chart below shows quarter-ending 12/31/07 and 3/31/08 exchange rates for the Euro, Won, Yen and NT$ (not the quarterly averages):

Source:  www.x-rates.com

Consistently tough pricing in DRAMs and on-again off-again tough pricing in NAND Flash throughout 2007 resulted in a dramatic reversal of fortunes for memory makers in 2007. Almost from the start of 2007, prices dropped precipitously, steadying DRAMs in 2Q but a rebound for NAND Flash. The summer was all-around better, but in the fall, price pressures resumed for both DRAMs and NAND Flash, and remain below production costs (DRAMs) or close to it for NAND Flash midway through 1Q08.

The overall memory market stayed about the same size from 2006 to 2007: $58B, +/- depending on who is counting and what is counted. DRAMs were down a bit to $31B, and NAND Flash went up a few $B, to nearly $15B. But the $58B in 2006 had nearly $9B in profit in it for memory makers; in 2007, vendors lost more than $2B, a large portion of which was in 4Q07. And, with prices at their annual lows on 31 December 2007, this sets the stage for another tough year in 2008.

For several companies, so severe was DRAM price pressures, that 4Q07 losses were equal to more than half their 4Q revenues. Major Taiwanese DRAM makers (i.e., Nanya, ProMOS, Powerchip), ...plus Qimonda, the only other pure DRAM supplier, all lost money in 2007, and it is probably safe to say that all DRAM product lines lost money, even though the product line was not singled out for broadline suppliers.

Losses were exacerbated by settlement of long-standing stock options, which gave a dose of losses to Rambus, Atmel, and SST, as well a frequent housecleaning of overdue charges for write-downs of inventory and for 'asset impairment' - equipment that had not been depreciated enough in past quarters, to align the financial statements with today's realities of how much the equipment was actually worth.

Without exceptional DRAM bit growth (which came in at more than 90% and the largest in more than a decade), and without the very impressive cost reductions by DRAM makers (mostly more than30-35% over the year, and as much as 15-20% in 4Q alone), it would have been much worse. The flagship 512M DDR2-667 opened at more than $6.00 and left the year at about 80 cents; with the rise of 70-75nm processing, the market shifted strongly from 512M to 1G for DRAMs, contributing both to cost reductions and to supply growth. Shipments of 1Gb DRAMs in January were fewer than 7M units, but by December had risen to more than 180M per month. 512M, which grew wildly to a record of more than 8B units, were higher than any density of DRAM in a single year - ever! This high peak shipment, which was undoubtedly aided by both strong demand growth and the industry's difficulty getting past 90-85nm processes, really gained momentum as 2Q moved into 3Q as 70nm processes and 1Gb production, really moved into high gear... and continues today.

As we enter 2008, Saifun's sale to Spansion is approved and done; the Numonyx JV between Intel and ST for NOR flash is wrapping up details and will be completed by the end of 1Q08. GSI, now a leading SRAM maker in a market many have walked away from, is now listed in our summary table. Both ST and Intel will continue to have an interest in NAND Flash ventures: ST with Hynix, and Intel as a part of IMFT, with Micron Technology.

End demand is strong, but productivity improvement and the specter of a housing-led recession remain major concerns. DRAM makers have cut back on their CapEx for 2008 to try to stabilize prices and head off an even deeper and longer glut; few expect DRAM GB growth to be as strong as 2007's was, but we thought the same thing last year, too, and got a major upside surprise. In fact, PC sales are strong, prices for memory are low, price elasticity is real, and no one knows with any certainty if 2008's DRAM business will be strong, weak or middling. Many companies are saying 2H08 will be a tight DRAM market, but who knows.

NAND Flash is estimated to be facing 130-165% GB growth in 2008. Investment is unabated.

Here is the Table of Financial results, (see below) comparing 2007 with 2006 for most major memory makers who report individual company or divisional data. We expect to publish more discussion about the 2007 memory market in the next few weeks, looking at cost reduction strategies and successes, DRAM product portfolio strategies, technical line-ups for major vendors in NAND Flash and DRAMs, and more insights (we hope) on what to expect for 2008.

"Others" include Sony, Matsushiuta, Renesas, Toshiba, Fujitsu, NEC, Elite Semi., G-Link, Alliance Semi., Sharp
'Profits' are divisional operating profits for Samsung, STMicro, Intel; profits are after-tax profits for remainder of companies
(ST & Intel will formally launch their NOR flash JV at the end of 1Q08; each will still have NAND business unit ST + Hynix, & Intel + Micron (IMFT)
msystems will be shown combined with Sandisk in 2008

Cypress is now more "Sunpower" = solar cells, than a Semiconductor company, and semi's will be listed separately after 4Q07
Saifun will continue to report separately until acquisition by Spansion is closed, probably early 2008
Mosys, Sandisk, SST, Saifun and Rambus include substantial IP licensing revenues as percentages of sales;
Exchange rates for 4Q07:107Y/$, 942W/$, 0.692 Euro/$, and 32 NT$/US$ for 4Q07